20240523 Japan Investment Conference 2024

OVERVIEW
The Japan Investment Conference (the JIC) is the flagship event held by the CFA Society Japan (the Society), and each year, it features the hottest topics relevant to the asset management sector. The JIC is the Society’s largest conference in terms of both the number of attendees and the number of people involved in its operation. We are confident that we provide helpful information to attendees with various backgrounds, including those in the financial industry. This year’s theme is “Unlocking the potential of AI.”
In the financial industry, including the asset management sector, the potential power of AI has been widely recognized, and it has already become a powerful enabler for innovating operations across the marketing, middle and back office, and front office functions. Given the rapidly changing environment surrounding AI, the Japan Investment Conference 2024 (The JIC 2024) will introduce the latest trends on how AI is applied in the asset management business and invite practitioners and experts from both Japan and overseas as guest speakers and panelists to discuss the background and impact of these developments on the asset management business.
The JIC 2024 is divided into three sections. The first section, the marketing section, introduces how AI changes Sales Force Automation/Customer Relationship Management in the asset management sector. We also have a panel discussion with guest panelists from a Fintech company, a security firm, and an XaaS company to discuss the measures to deliver effectiveness in the asset management business and showcase their best practices on the customer experience. In the second section, the middle and back office section, the guest speakers explain how AI can contribute to higher productivity and enhanced risk management from the practitioner’s point of view. In the last section, the front section, we will focus on how advanced AI technologies can be applied to investment strategies and what benefits can be derived from the application of AI to ESG and SDGs investing. Panelists from an asset management company, academic, and data provider also discuss their initiatives on these issues and global trends from each viewpoint to find out practical implications.

Transcript

1. JAPAN INVESTMENT CONFERENCE 2024 Thursday May 23 | Tokyo Convention Hall Unlocking the Potential of AI
2. PLAN via digiBank Financial Planning- Reimagined “Helping our customers get better with money”
3. Source: Asia Development Bank The need to get everyone to be serious about Financial & Retirement planning 1 Asia is rapidly aging 2 Declining population 3 Rising cost of healthcare
4. Democratising access to wealth mgmt services leveraging data & technology FINANCIAL PLANNING LITERACY 3 >3.0m customers in SG have started their financial & retirement planning with DBS • Content creation/ curation/ adaptation • NAV.sg (PWEB) • NAV Insights - weekly • NAV University • DBSV daily • InsightsDirect • Partnerships & Collaborations • CPF Board • DBS-SUSS Fin-Lit course • SGX-DBS FLY program Empowering with expert knowledge • Gamification • DBS BINGO • Multiplier program DBS NAV PLANNER Sharpening our advisory tools • Asset & Liabilities • Net-worth • Budgeting • Savings • Protection • Investment & Suitability • Retirement (accumulation) • Equities via DBS Vickers • FX • digiPortfolio • InvestSaver • UT/ Funds • Protection • Endowment DIGITAL INVESTING & INSURANCE SOLUTIONS Making investing & insuring accessible CIO Insights Portfolio tools digi- Wealth NEW NEW NEW NEW
5. Retirement = 62 Accumulation Decumulation Balance Sheet Balance Sheet & Income/ Expense Non-DBS DBS CPF withdrawal = 55 guided by Financial Planning interactions Open banking with a purpose – SGFinDex in Singapore Empowering all Singaporeans to plan for retirement… as well as during retirement Personal balance sheet CDP Insurance Govt (CPF/ HDB/ IRAS) Investment Deposit Lending
6. We are uniquely equipped to help individual investors grow their wealth Insurers & IFAs Protection needs INSIGHTS & ADVISORY Strategic insights straight from CIO World’s best digital advisory solution NAV Planner INVESTMENT SOLUTIONS In-house integrated solutions designed for affordability DATA & TECHNOLOGY Cutting-edge AI/ML to analyse data with tech to scale to millions Banks Savings, Lending, Wealth solutions Brokerages & DIY Platforms Investing & Trading End-to-end financial/wealth planning model Full spectrum of financial advice/content, services & solutions
7. Retirement & Goals Digital Advisory (NEW) digiPortfolio SGFinDex Unit Trusts FINANCIAL PROFILE ADVICE SOLUTIONS Leveraging data from within and outside DBS Trading Insurers & IFAs Banks Brokerages & DIY Platforms DBS helps customers approach planning and investing holistically by considering the cashflow & balance sheet as well as changing life stages & financial circumstances Savings, Protection, Investment, Liability Mgmt Retirement All within DBS, with no hand-offs Financial & Retirement planning - Reimagined
8. Consolidated view via dashboard Key indicators of their financial wellness • Savings • Protect • Invest • Retire (Map Your Money) Savings Protection Investments Budgeting & Monthly cashflow Map Your Money Assets & Liabilities NEW! Connect to SGFinDex Ideas for you Key components to build a robust plan • Budgeting & Cashflow • Asset & Liabilities • Actionable Insights (Ideas for you) Powered by AI/ML Available online & offline
9. Powered by AI/ML Protection Prioritise type of coverage based on needs and identify coverage gaps Discover protection needs based on dependants and financial commitments Highlight blind spots (e.g. inflation, funeral cost, part- time help) 1st in industry to differentiate protection vs. investment. Address common pain points of customers: 1. Understanding types of coverage 2. Premium affordability 3. Amount of coverage required based on lifestage & industry guidance Helps customers improve their cashflow Powered by AI/ML Powered by AI/ML Available online & offline
10. ▪ Personalised and actionable insights to customers on how to deploy idle cash ▪ Seamless journey to digital fulfillment ▪ All within DBS digibank. No transfer of funds for settlement. No white- labelled solutions Real-time product shortlisting based on customer risk profile + real-time hyper-personalised experience as customer interacts with the tool Powered by AI/ML Powered by AI/ML Investment
11. Projection of future cashflow based on current assets (incl. deposits, investments, CPF & SRS) • Project future cashflow based on current assets & liabilities (data within & outside DBS) • Integrate complex CPF & SRS rules (govt pension schemes) into the automatic calculations • Visualise potential gaps resulting from life goals Retirement Powered by AI/ML Actionable nudges to improve forecast of future cashflow Integrate life goals to assess impact on projected financials Powered by AI/ML Best-in-Class
12. Personalised & Actionable Insights • Powered by >100 AI/ML models • Delivered through Next Best Actions (NBA) & Next Best Conversations (NBC) • Persoalised for >1m customers every month Looks like you’re saving just 3.5% of your monthly income. Aim to hit at least 10% of savings ratio. SET A BUDGET Did you know? At 26, you can get life insurance at SGD 0.95 a day for $100,000 sum insured. TALK TO US Consider digiPortfolio to keep your money invested. Your investment proceeds will be credited in 2 days. BUY NOW Finding ways to boost your CPF retirement pot? Evaluate your personal situation before making any decision. READ MORE Insights are personalised and powered by AI/ML models Insights are actionable from end-to-end
13. Over 3m customers in Singapore engaged with 280m nudges S$2.2b AUM growth in digital investment AUM 2.7m DIY transactions generating S$47b trading volume 6X increase in premiums for digital insurance #1 in market share for retirement schemes – SRS & CPFIS Automating >200,000 risk profiles, saving 50,0000 hours
14. PLAN via digiBank Financial Planning- Reimagined “Helping our customers get better with money”
15. 2024年5月23日 上級執行役員 生成AI統括本部長 宮澤 弦 LINEヤフー株式会社 マーケ領域・セールス領域における 生成AI活用の取り組み 日本CFA協会主催 Japan Investment Conference 2024
16. © LY Corporation 宮澤 弦 Gen Miyazawa LINEヤフー株式会社 上級執行役員 生成AI統括本部長 1982年生まれ 北海道札幌市出身 東京大学 農学部卒業 登壇者紹介 略歴 株式会社シリウステクノロジーズを創業し代表取締役に就任 モバイルSEO事業や位置情報に連動した広告配信事業を展開 ヤフー株式会社によるシリウステクノロジーズ買収、同社に入社 執行役員 検索サービスカンパニー・カンパニー長等を経る 取締役 専務執行役員 コーポレートグループ長 現職 大学卒業後 2010年8月 2022年4月 2023年10月 © LY Corporation
17. マルチベンダー戦略でサービス実装を加速 中長期的に1,100億円の売上収益を見込む はじめに FY2023 個人向けサービスを中心に16件で生成AIを活用 FY2024 LINEや広告等へのさらなる生成AI実装を推進 個人向け LINE AI アシスタント LINE AI Q&A1 オープンチャット メッセージ内容要約 LINE プロフィールスタジオ PayPayグルメ ChatGPTプラグイン2 Yahoo!しごとカタログ クチコミAI要約文 β版 Yahoo!フリマ 簡単出品 Yahoo!知恵袋 生成AI回答 Yahoo!検索 一部クエリAI回答 法人向け DS.INSIGHT データ解釈機能 AI Prompt Manager （LINE公式アカウント） 社内向け Yahoo!ニュース コメントAI要約 β版 Yahoo!ニュース めくる・めぐるTHE NEWS CARDS 2023 • キャッチコピー自動生成ツール • GitHub Copilot 導入 • 社内アシスタント LY ChatAI エンジン マルチベンダー戦略で 世界最高のエンジンを活用 データ 国内No.1プラットフォームサービスの 膨大なデータベース ユーザー インター フェース 圧倒的ユーザーベースと LINE公式アカウントや各種サービスで のユーザー接点 （当社2023年度 通期及び第4四半期決算説明資料より） © LY Corporation 1. 2024年5月8日終了 2. 2024年4月10日終了
18. © LY Corporation クリエイティブ生成および営業活動における社内業務効率化において、積極的に活用 マーケ領域・セールス領域における生成AI活用事例 広告営業活動における仕様確認 クリエイティブ 社内業務効率化 広告クリエイティブの制作アシスト 社内クリエイティブ生成用の社内ツール開発、展開 商品説明やメールのタイトル生成 キャッチコピー
19. © LY Corporation 2023年10月からのAdobe社との協業により実現 広告クリエイティブの制作アシスト LINE広告・LINE公式アカウント LINE Creative Lab上でAdobe Expressを 開くことにより、クリエイティブ画像や動画を 手軽に作成し入稿ができるように 画像生成AI機能も利用可能 LINE広告 LINE公式アカウント 200点以上の テンプレート 約240点以上の テンプレート ※一部の素材は「Adobe Express」プレミアムプランのみの利用になります。
20. © LY Corporation 配信面に沿ったバナーのレイアウト生成にOpenAI APIを利用し、作業工数を削減 社内クリエイティブ生成用の社内ツール開発、展開 Yahoo! JAPAN トップページ Googleアプリ広告 さまざまな箇所で 生成AIを利活用
21. © LY Corporation 複数生成分 • 寝るときの暑さに困っていませんか？ • 涼しい寝具で熟睡をサポート • ひんやり気持ちいい寝具で快眠を • 涼しさを求めるなら冷感寝具 • 爽やかな寝心地を冷感寝具で 2023年6月にリリース、OpenAI APIを利用（GPT4） 商品説明やメールのタイトル生成 Yahoo!ショッピング／Yahoo!フリマ等 プロンプトを調整し、枠ごとに「それらしい」 キャッチコピーを生成。複数文言を最適化配信 することでCTR向上 ほか、メール件名生成においても生成AIを活用 人による作成（1文のみ） •「ひんやり寝具」で寝苦しい夜を快適に
22. © LY Corporation 社内アシスタント LY ChatAIの活用により、社内業務効率化を推進 広告営業活動における仕様確認 従業員 社内向けツール （RAGツール構築） 社内 データベース 質問 回答 LLM （大規模言語モデル） 回答 質問 結果提示 検索 RAGツール1 構築の開発に着手。社内データの活用による業務の自動化・効率化の実現 機械学習での離反傾向顧客の特定と抑止 1. RAG：検索拡張生成（Retrieval Augmented Generation）：検索ベースと生成ベースの人工知能（AI）モデルの両方の長所を組み合わせた、自然言語処理（NLP）技術
23. © LY Corporation 「WOW」なライフプラットフォームを創り、日常に「！」を届ける。
24. ジャパン・インベストメントカンファレンス2024 企画部門やオペレーション部門業務への 生成AI活用 PwCコンサルティング合同会社 ディレクター 白取 宏之 2024/5/23
25. PwC 白取 宏之 PwCコンサルティング 合同会社 ディレクター 経歴 コンサルタントとして資産運用領域における戦略策定から施策の実現までを一貫して支援 2022年よりPwCコンサルティング合同会社のディレクターとして資産運用領域を担当 寄稿 「生成AIが変える資産運用業務」 スピーカー紹介 主なコンサルティング実績 • 国内大手金融グループ 運用機能のグループ内での再編（フロント、ミドル、バック全般）の企画から実際の業務・ システム・組織・要員の移管支援 • 国内大手保険会社 資産運用部門の効率化・BPR企画検討、テクノロジー調査、PoC実施、本番導入の支援 • 国内大手運用会社 事業継続体制の高度化の企画検討、実際の事業継続体制の構築、PDCAの支援 • 国内大手保険会社 統合・買収にかかるPMI支援 他運用領域における支援実績多数 2
26. PwC 1. 生成AIとは何か 2. テキストの生成 3. 企業におけるユースケ―ス 4. 社内導入における課題 アジェンダ 3
27. 生成AIとは何か
28. PwC 5 経営陣の視点および生成AI進化の加速度 70% 45% 1万倍 のCFOは今後12～18カ月以内に生成AIに 投資すると回答※4 4年以内に生成AIは進化し、企業はより 低いコストで良い結果を達成できるよう になる※2 のCEOは今後３年間で生成AIが自社の 価値の創造、提供、獲得の方法が大きく 変わると回答※1 生成AIは2030年までに世界経済へ貢献す る可能性がある※3 15.7兆ドル • 出典： • *1. PwC Pulse Survey 2023 • *2. Hugging Face, Large Language Models: A New Moore’s Law? • *3. PwC Publications, Global Artificial Intelligence Study: Sizing the prize • *4. PwC 27th Annual Global CEO Survey
29. PwC 生成AIとは何か？ 6 AI （人工知能） AI は環境に組み込まれたシステムの理論と開発であり、 特定の目的を達成するために感知し、決定・行動 機械学習 AI のサブフィールドは、時間の経過と経験を 通じてパフォーマンスを自動的に向上させる システムの構築に焦点 利用はデータ サイエンティ ストに限定 ディープ ラーニング 人工ニューラルネットワークに基づ く機械学習手法。複数階層の処理 により、データから段階的に高レベ ルの特徴が抽出 生成AI 大量データと大規模な 事前学習済みモデルを 使用して新しいコンテン ツ生成するアルゴリズム • テキスト、コード • 画像、動画 • 音声 などを生成可能 一般的な開 発者やユー ザーが広く利 用可能
30. PwC TEXT マーケティング文章生成、 メール生成、チャットサポー ト、議事録生成、文書要約校 閲など PROGRAM プログラム生成、ドキュメン テーション、text to SQL、 Webアプリ生成など SOUND プレゼンテーション 音声合成生成、 音楽生成 など IMAGE マーケティングコンテンツ、 Webデザイン/市況情報などの チャート生成、スタイル変換など OTHER 3Dモデル（アバター）生成、 など 生成AI 領域 7
31. テキストの生成
32. PwC 生成AIの登場により、大規模な投資をせずとも業務変革に取り組める余地が拡大。資産運用業界における 戦略的な取り組みに対しても生成AIを取り入れることで、競争力確保への貢献が期待されている 資産運用業界を取り巻く環境と生成AIの可能性 生成AIを取り巻く環境 投資家の裾野拡大に伴う 顧客サポート向上 資産運用業界を取り巻く環境 個人や投資初心者を意識した 「わかりやすい説明」 資産運用会社が果たすべき 金融商品の品質管理向上 運用状況やコスト水準に関す るモニタリングの効率化 経済環境の変化に対応する 運用対象の多様化 分析結果やリスクバランスを 勘案した銘柄選択の最適化 対応の方向性(例) 生成AI活用の可能性 クリエイティブ生成により図や写真、文言を挿入 した、わかり易いレポートを作成 各種情報源からのデータを収集・分析し、多角 的な観点で要約されたレポートを自動生成 株式・金利・為替・マクロ経済など大量のレポー トをテキスト要約し、ファンドマネージャーの スピーディな投資判断に活用 即戦力人材を外部採用するこ との競争激化 デジタライゼーションにより 必要な人的リソースを削減 単純な問い合わせなどはロボアドバイザーツー ルと連動させ担当者の対応負荷を軽減 9
33. PwC ポイント① テキスト生成AIの代表的機能分類 現在、企業における業務での利用可能性が高いテキスト生成AIの主要な機能は7つに分類される 10 おさえるべきポイント 入力されたテキストを一つの言語から別の言語に翻訳 既存の知識・Webブラウジングで最新の情報をもとに質問に回答 （事実の提供、概念の説明、比較分析など） ユーザーのプロンプトに基づいて新しい文章やアイデアを生成 （ストーリーテリング、文章生成、創造的な問題解決など） 言語翻訳 情報提供 （検索・抽出） クリエイティブ生成 テキスト要約 対話 長いテキストや複雑な概念を短く、明確に要約 ユーザーとの自然な会話を行い、文脈を理解して適切に応答 コーディング ユーザーの命令通りのコンピュータ―プログラムを生成 添削 入力した文章の正誤判定や添削を行う 1 2 3 4 5 7 6
34. PwC ポイント② プロンプト（入力）による出力結果への影響 生成AIの特徴である「プロンプト」による指示によりアウトプットを生成するが、プロンプトの入力の仕方によって返す内容 は異なる（アウトプット作成における定量的な指示を与えた方が、得たい情報をシンプルな回答で受領することができる） おさえるべきポイント 1.クラウドコンピューティングとデータ分析：企業はクラウドプラットフォームを活用 してビジネスデータを収集し、リアルタイムで分析することで、迅速な意思決定を 支援しています。これにより、生産性の向上やコスト削減、新たなビジネスチャン スの発見が可能となります。 2.オムニチャネル販売：小売業界では、顧客がオンライン、モバイル、店舗などの 複数のチャネルを使って製品やサービスを購入できるようになりました。オムニ チャネル販売を実現するために、デジタルショッピングエクスペリエンスと物理店 舗の統合が重要となります。 3.自動化とロボティクスプロセスオートメーション（RPA）：業務プロセスの自動化 により、効率と正確性が向上し、人間のエラーを減らすことができます。RPAは、 ルーティンワークや繰り返しのタスクを自動化するために使用されます。 4.インターネット・オブ・シングス（IoT）：センサーやデバイスを利用して物理的な オブジェクトをネットワークに接続し、データの収集と相互通信を可能にします。 製造業では、IoTを使用して生産ラインの監視や予防メンテナンスを行うことで、 効率性や品質を向上させています。 5.仮想現実（VR）や拡張現実（AR）：製品のデザインやトレーニング、販売体験な どにVRやARを活用することで、顧客や従業員の参加度や理解度を高めること ができます。また、遠隔地のチーム間でのコラボレーションやリモートトレーニン グにも役立ちます。 これらは一部の例ですが、DXはさまざまな業界やビジネス領域において大きな 成果を上げています。重要なのは、組織のニーズと目標に合わせて適切なデジ タルテクノロジーを選択し、導入することです。 1. AIによる自動顧客サポートシステムの導入（例：チャットボット）：効率的 な顧客対応とコスト削減。 2. データドリブンな予測保守管理システムの構築：設備の故障を予測し、 メンテナンススケジュールを最適化。 3. モバイルアプリを活用したオンデマンドサービス提供：顧客の利便性向 上と新たなビジネスチャンスの創出。 プロンプトのチューニング前 チューニング後 「業務で役立つDXの事例を教えて」 「DX推進部において、業務で役立つDX推進事例を最新のもの から3件教えて。回答はそれぞれ50文字以内で」 ユーザプロファイルを考慮した 新しい情報に絞り込まれた回答が得られる 700文字以上 150文字程度 11
35. PwC ポイント③ 社内外文書参照による回答精度向上 生成AIから社内外文書を参照させることにより、事前に学習されていない情報を引き出すことが可能 おさえるべきポイント 生成AIモデル ① ② ③ ④ ① ユーザーは生成AIシステムにプロンプトを送る ② プロンプトに関連性の高い情報（＝参考情報）を検索 ③ 検索結果を生成AIシステムに返す ④ プロンプトと検索結果を生成AIシステムに入力し回答を生成 プロンプト 生成AIシステム 社内外文書 Web、社内規程など Pros • 最新の情報や社内に特化した情報を引き出すこ とが可能になる • 生成AIモデルの外部情報を参考するため、モデ ルが学習していない情報も生成できる • 生成AIへの追加トレーニングが不要なため、 Fine-tuningと比べてコストが安い • 参考情報の品質が低いと生成物の品質が低下 • 参考情報のデータ量が大きいと制限に抵触し、回 答を生成することができない • 参考情報に機密情報を含める場合、生成物から 機密情報が漏洩するリスクがある • 事前に参照するための参考情報を準備する必要 があり、工数がかかる可能性がある Cons 12
36. 企業におけるユースケ―ス • 活用パターン • ユースケース
37. PwC LLMの活用パターン 大規模言語モデル（LLM、 Large Language Model）の活用は大きく3パターンに分類される。社内データの 活用や他システムとの連携をするほど造り込みが発生する 14 LLMが学習済みのデータ生成 機能の範囲でアウトプットを作 成し活用するパターン 以下は代表的な機能 ・要約 ・翻訳 ・添削 ・ドラフト作成 ・アイデアだし、壁打ち ・論点抽出、分析 など プレーンなLLM LLM+社内文書連携 LLM+他のITツール ＋ 検討 パターン 概要 外部Webのみならず社内特有の 文書（マニュアル、規程集、社員 情報、製品・売上情報など）を生 成AIと連携し、参照させることで、 LLMが学習済みの範囲で社内 の専門性や特有性のある業務に も活用を広げるパターン プレーンなLLMへ他のITツール （プログラミング、演算ソフト、BI ツールなど）と連携させ、より広 範囲でのデータ参照や用途で の活用が可能 右に行く程IT面の造り込みが発生
38. PwC 社内外からの問い合わせ対応の効率化 業務の堅確性と生成AIによる効率化方法 生成AIを組み込んだ効率化イメージ 1 2 1 2 ユーザがSaaS経由 で質問 SaaSが直接回答 ユーザが質問を入力 生成AIがFAQ案を作成 生成AIによる対応 堅確性 マ ニ ュ ア ル ・ 過 去 QA を 参 照 通 常 業 務 1 4 生成AI活用ユースケース：LLM+社内文書連携 業界・業務領域を問わず、LLM+社内文書による情報参照・検索や問い合わせ対応の効率化を実現した後、社内リソー スを専門性の高い分野に配置することを目的としたユースケ―スが数多く検討されている 高 度 な 業 務 規 程 な ど の 社 内 文 書 生成AI 内蔵型 SaaS 1 2 FAQ 担 当 者 生 成 AI チ ャ ッ ト ボ ッ ト UI 2 3 5 3 担当者がFAQ案を確認 4 FAQ内容に問題がなけ れば、生成AIより回答 5 個別対応が必要の場 合、担当者より回答 堅確性 業務内容(例) 効率化方法 高度 • 機関投資家からの 問い合わせ • 販売会社からの問 い合わせ • FAQを作成する 生成AIを構築し、 人による作成内 容のチェックを 経て顧客へ回 答する仕組みの 構築 通常 • システム関連など のヘルプデスク問 い合わせ • 社内規程・ルール の検索 • 正答率90%程 度の品質が担 保される、生成 AIを内蔵した SaaSを導入 実現事項 • 個別性や業界 特有性の高い 業務においても、 担当者の作業 負荷軽減や業 務効率化を実現 • 早期にAIを導入 することが可能 • 社内リソースを 専門性の高い 分野へ配置する ことが可能 15
39. PwC アナリストレポート、運用報告書、ファンド企画および販売用資料のコンプライアンスチェックにおいて、各国の金融規 制・法令に関連するデータ、レポート、その他重要な情報を参照・学習した生成AIを利用することにより、コンプラ担当 者の知識・経験に寄らない幅広く且つ平仄の取れた判断に基づいたチェックが可能となる ユースケース① 各種資料のコンプライアンスチェックの効率化 ◆実現イメージ（案） • コンプライアンスチェックにおいて、現在は担当者の知識や経験を基に 対応しており属人的となっている • 各国の金融規制の変更を網羅的に把握し、最新の情報を用いて生成 AIを活用したコンプライアンスチェックを行い効率化に繋げたい 必要な検討要素 各国の金融規制情報の適時収集 情報の正確性・最新性 専門知識（法関連）の解釈の妥当性 ➢ コンプライアンスチェックでの抽出条件を学習した生成AIへ金融 規制関連の情報や報告書などの文書類を参照させることで、指 摘候補事項を自動的に抽出し当該運用を効率化 活用の概要・背景 1 2 3 ④指摘箇所に印がつ いたレポートが生成 生成AI活用ユースケース 生成AI ③法令・規制情報を基に各 書類のコンプライアンス指 摘候補箇所を抽出 チェック前のレポート・ 報告書・企画書 ②生成AIが各レ ポートを参照 各国の金融規制関連情報 ①生成AIが適時規制 情報を参照 担当者 ⑤精査の上、 適宜更新 × × × △ △ △ 16
40. PwC 企画部署スタッフは社内規程に基づいた判断が求められることも多い中で、どの文書に何が記載されているかを検索 する手間も多いが、チャットボットと生成AIを活用した仕組みを導入し検索業務を効率化 ユースケース② 社内規程や文書、ルールの検索効率化 ◆実現イメージ（案） ➢ 社内規程やルールに係る文章、過去回答内容などの収集と蓄 積を行い一元管理することで、企画部署スタッフからの問いわせ への回答を自動化 • 企画部署スタッフが業務上社内規程を参照する際、確認したい文言を 含む文書の特定と記載内容の検索に多くの時間を費やしている • 規程や手続きについてチャットボットへ問い合わせると、該当する書類 と記載内容を自動的に回答する仕組を構築することで、当該業務負荷 軽減と効率化の向上を図りたい 1 2 3 必要な検討要素 最新の社内規程やルール、約款などの 文書の保存 情報の正確性・最新性 チャットボットUIの開発 活用の概要・背景 企画部署スタッフ チャットボット 生成AI ①質問入力 ②関連情報の検索 ③関連情報の抽出 ・過去の回答内容 ・規程類各種情報 ④回答 【チャットボットによる問い合わせ対応例】 企画メンバの質問： 事業投資を行う際にはどのような手続きや申請が必要ですか？ 生成AIの回答： 1. 1,000万円以下の事業投資の場合、各事業部担当役員決裁になります。 投資実行前までに、稟議回付を完了させてください 2. 1,000万円超の事業投資の場合、経営会議附議と社長承認が必要とな ります。 経営会議の日程調整と投資立案の上、経営会議での決裁が必要です。 ※投資管理規程.pdf p5：3項. 投資基準金額、p6:5項. 決裁者 職務権限規程.pdf p23：7項. 稟議フロー 生成AI活用ユースケース 17
41. PwC 複数レポートからのデータ抽出およびデータ活用効率化 業務内容と効率化方法 生成AIを活用した効率化例 1 2 AI-OCRが各GPの運用レ ポートより定量情報を抽出 生成AIが各GPの運用レ ポートより条文などの文章 情報抽出 生成AI＋ITツールによる対応 インプット 生成AI活用ユースケース：LLM+他のITツール 生成AIだけでなく図や表を作成するロジックを持つITツールとの組み合わせで、レポートなどのアウトプット 作成に活用するユースケ―スも検討されている 3 データローダーを介し定量 情報をDBへ格納 4 計算・可視化ロジックを組み 込んだツールにて図や表を 生成 業務内容(例) 効率化方法 • オルタナファンド運用に係る 運用会社(GP)の市況確認 やパフォーマンス分析 • 運用会社毎に異な るフォーマットの運 用レポートより、必 要な条文や項目を 自動的にデータとし て抽出 • 抽出したデータを用 い計算、図や表形式 に加工する仕組みを 持つITツールの構 築 実現事項 • 業務上必要とな るデータの確認 と抽出に係る作 業負荷軽減 • 抽出データを活 用した各種分析 作業の効率化 5 生成AIが文章と図、表を組 み合わせてレポート作成 各 GP か ら の レ ポ ー ト ( 異 フ ォ ー マ ッ ト ) 1 生 成 AI 3 DB ア ウ ト プ ッ ト 4 5 デ ー タ ロ ー ダ ー AI-OCR XXX 計 算 ・ 可 視 化 ロ ジ ッ ク 2 18
42. PwC オルタナティブファンド運用会社が提供するレポートより、特定の項目値や表、条文をAIが自動的に抽出し、 パフォーマンス状況の可視化や社内向けレポート作成に活用 ユースケース③ オルタナティブファンド運用での社内レポート作成効率化 ◆実現イメージ（案） ➢ オルタナティブファンド運用レポートの文章情報とAI-OCRより抽 出した定量情報を生成AIが参照し、社内レポート用の文章作成 ➢ 数値計算・可視化ロジックを有するツールと生成AIが連携し、文 章＋図・表を組み合わせた社内レポートとして完成させる • オルタナティブファンド運用に係る分析や社内レポート作成業務を進め る際、オルタナファンド運用会社各社が提供するレポートから必要情報 を抽出する作業に、担当者は多くの労力を費やしている • 社内レポート作成に必要となる情報抽出を生成AIとAI-OCRで自動化し、 BIツールと連携させることで業務の効率化を図る 1 2 3 必要な検討要素 オルタナティブファンド運用に係るレ ポート類の情報収集・保存 情報の正確性・最新性 抽出項目・条件の妥当性 活用の概要・背景 生成AI活用ユースケース 19 生成AI ⑤運用レポート の文章情報取得 データローダー オルタナ運用会社 Webサイト RPA XXX ①運用 レポート を取得 ②運用レポート を蓄積 AI-OCR ③運用レポート の定量情報取得 オルタナ運用レポート 生成AI用 DB 市況情報DB ⑥図や表を 作成 社内レポート ④DBへ定量情 報格納 計算・可視化 ロジック ⑦生成AIが社 内レポート作成
43. PwC 資産運用業務における生成AI活用の可能性① リスク管理会議報告書の ドラフト作成 • 過去の会議報告書を学習させ、今後の会議資料のドラフトを自動化 市場リスクの管理 • マーケットデータやリスク指標を読み込ませ市場リスクに係るパターンや トレンドを認識させることで、リスク管理において参考となる情報を生成 商品説明資料のドラフト 作成(デザインアイデア) • 商品説明資料案のドラフト作成やデザインアイデア案の自動生成を行う ことで、本来企画検討に要していた工数の削減 統計データの収集・要約 • 統計データを収集し読み込ませることで、分析・要約された情報を短時 間で生成 企画書のドラフト作成 • 俗人的な暗黙知となっていた企画アイデアが形式知化されてアイディア の質向上 生成AI活用による期待効果 業務 生成AIの機能※ 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 生成AI活用の効果 商品企画における規制・ 法令などの情報収集 • リアルタイムでの法規制情報の収集を行い、それらを基にした分析結 果・要約することによる業務の効率化 〇 〇 〇 リスク管 理 運用企画 商品企画 ディスク ロージャー 運用報告書の文書校正・ 数値整合チェック • 過去の運用報告書を学習させ、今後の報告書の文書校正を自動化 • 運用報告書と比較対象となる文章を参照し、不整合箇所抽出を自動化 〇 〇 ミドル 帳票などのドラフト作成 • 過去の帳票を学習させることで、今後の帳票のドラフト作成を自動化し 効率化 〇 〇 ③オルタナティブファンド運用 での社内レポート作成効率化 • 多種多様なデータソースより必要情報の抽出を自動化し効率化 〇 〇 オペレー ション バック・ コーポ レート 数値の整合性チェック • 比較対象となる資料をそれぞれ読み込ませ、不整合部分の特定を自動 化し作業を効率化 財務諸表の注記文言案 の提示 • 長文で記載された文言をテキスト要約して抽出することが可能 数値の整合性チェック • 比較対象となる資料をそれぞれ読み込ませ、不整合部分の特定を自動 化し作業を効率化 〇 〇 〇 計理・ 経理 20 〇 〇 〇 〇 LLM活用の パターン プレーン LLM活用パターン凡例 +社内文書 連携 +その他IT ツール ※機能1. 言語翻訳、機能2. 情報提供(検索・抽出)、機能3. クリエイティブ生成、機能4.テキスト要約、機能5. 対話、機能6.添削、機能7.コーディング 機能1 機能2 機能3 機能4 機能5 機能6 機能7
44. PwC 資産運用業務における生成AI活用の可能性② 議決権行使のデータ集計 • リアルタイムでの議決権行使情報の収集を行いそれらを基にした分析 結果・要約を抽出することによる業務の効率化 ESG関連レポートの ドラフト作成 • 広報企画案の作成や対内外向けの動画像案の自動生成を行うことで、 本来企画検討にかかっている検討・企画～作成における工数の削減 ヘルプデスクの 問い合わせ対応 • これまでの問い合わせ内容や回答、報告書を読み込ませておくことで、 効率的かつ自然な表現で回答や報告内容のドラフトを生成 ①各種資料の コンプライアンスチェック • 規制の変更を網羅的に把握することが可能 • 最新の情報でのコンプライアンスチェックが可能 法規制情報の収集 • 市場・企業環境の変化に伴う各種規制の変化が発生した際に即時に収 集し自社の規程に適応することで違反リスクの低減 生成AI活用による期待効果 生成AI活用の効果 各部署からの改修要望に かかる情報収集・分析 • 各部署から異なる文言で起票される同じ意味合いである要望を適切な 文言として置き換え、一覧化を行い、集計業務を効率化 CSR/ES G推進 コンプラ イアンス システム 経営企画 業界レポートや統計資料 の収集および要約 • 長文レポートや資料をテキスト化し読み込ませることで、短時間で要約し ポイントを絞った形で記載内容を抽出 バック・ コーポ レート 企画書のドラフト作成 • 俗人的な暗黙知となっていた企画アイデアが形式知化されてアイディア の質向上 企画概要説明資料 • 長文レポートや資料をテキスト化し読み込ませることで、短時間で要約し ポイントを絞った形で記載内容を抽出 ②社内規程や文書、ルー ルの検索効率化 • ナレッジの収集/蓄積を行い一元管理することで問い合わせに関する回 答を自動化 21 生成AIの機能※ 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 〇 LLM活用 のパターン プレーン LLM活用パターン凡例 +社内文書 連携 +その他IT ツール ※機能1. 言語翻訳、機能2. 情報提供(検索・抽出)、機能3. クリエイティブ生成、機能4.テキスト要約、機能5. 対話、機能6.添削、機能7.コーディング 機能1 機能2 機能3 機能4 機能5 機能6 機能7 業務
45. PwC ユースケース選定から実現までの進め方の例（ハイレベル計画） ユースケースを選定した後にPoCによる実現性の確認や、QuickWin施策の実行などを行い、生成AIの社 内展開を目指す 22 ユースケースの ロングリスト作成 展開計画のドラフト、 Quick-win着手 展開計画の最終化、 本格展開体制の組成 本格展開 ゴール • ユースケース案の洗い出し • PoC実施是非、取り組み順序な どの優先度付け、計画策定 • Quick-win施策の具体化 • PoCを通じたユースケース精査 • 貴社としてのVisionと来年度以降 の本格展開計画（第一版）作成 • Quick-win施策の実行 1~2カ月 ～2カ月 3カ月～ 主な 対応事項 アウトプット • 各部門からのアイディア収集 （トレーニングを含む） • ユースケースのロングリスト化 • Quick-win施策の企画策定 • 開発ベンダーの選定 • PoC計画の策定 • ユースケースのロングリスト • Quick-win施策の企画 • PoC計画 • PoCの実施（継続的なユース ケースの追加を含む） • ユースケースの取り組み順序や 時間軸、必要体制などの計画 （第一版）作成 • Quick-win施策実施・評価 • 研修計画・活用ガイドライン検討 • ユースケースのショートリスト • Visionと展開計画（第一版） • Quick-win施策の進捗 • PoCを通じたユースケース精査 • 2024年度以降の本格展開計画 の最終化、体制の組成 • PoCの実施（継続的なユース ケース追加を含む） • ユースケースの取り組み順序や 時間軸、必要体制などの計画の 最終化 • 推進体制の定義と要員確保 • 研修コンテンツ・活用ガイドライン 策定 • 展開計画（更新版） • 推進体制に必要な要員確保 • 活用ガイドライン
46. 社内導入における課題
47. PwC 企業における生成AI社内導入フェーズの進捗 機関投資家においては業務活用を意識した検証を全社的に開始したところもある。運用会社は、社内が許 容する範囲でフロント人員が一部業務でスポット利用しているケースは存在すると考えられる 24 カルチャー 変革 検討カテゴリ 人材育成 業務活用 活用期 発展期 成熟期 導入期 フリー検証 （一般的な）ユースケース検証 （固有業務の）ユースケース検証 社外接点における活用 幹部向け教育 中核メンバー向け教育 一般ユーザー向け教育 アイデア創出イベント 組織変革ワークショップ テクノロジー活用の企画立案 ビジネスプロセス変革（意識） 独自モデル構築のための教育 自走のための教育 （高度専門業務の）ユースケース検証 ビジネスプロセス変革（実業務） 基盤構築 一部社員向けトライアル環境 全社員向けトライアル環境 個別アプリケーション開発 社内外データ連携 プラットフォーム構築 運用体制構築 事業アイデア創出のプロセス整備 新規ビジネス展開 リスク管理 （初期的な）ガイドライン策定 ガイドラインの修正 インシデント集の回付 モニタリングの仕組化 リスク管理拠点の拡張 第三者委員会の設置 モニタリングの全自動化 ポリシー策定 （全社員の）リテラシー向上 生成AIの活用フェーズ 機関投資家 資産運用会社 社内導入フェーズ
48. PwC 25 CoE組織による全社横断的な生成AI展開 DX推進組織やIT部が主導し、全社共通基盤の構築と並行して各事業部のリテラシー底上げ、個別ユース ケ―ス収集とPoC支援を行う企業が増加 社内展開アプローチ 社内独自のLLMを実装、 共通基盤として事業部門へPoC展開 生成AIプラットフォーム System 検索 生成AI 事業部 人事 ・・・ CS 法務 オリジナルLLM 複数事業部への PoC展開 全体イメージ 個別活動 生成AI共通基盤と インフラ構築ノウハウ 生成AI開発における 機能・非機能ノウハウ ニーズ調査表 業務部門が抱える改善ニーズ ワークショップ 生成AIワークショップコンテン ツ整理ノウハウと ワークショップドキュメント PoC推進設計 アジャイルデリバリーモデルに よるエンゲージ・オンボーディ ング・実行に至る方法論 ユースケース案 大量収集
49. PwC 26 効果的なユースケース検討の分類例 生成AIのユースケースは、実務で必要となる業務知見を参照する「助言型」、実務で活用する作成物を生成 AIが生成する「作業補助型」、一連の業務を生成AIが代替する「業務代替型」に類型できる想定である • 従来人が担ってきた一連の業務を生成AI によって代替させるユースケース • 既存技術と組み合わせる可能性が高い • 実務における作業の一部を生成AIに代替 させるユースケース • 生成AIが注目を浴びた一つの要因と想定 • 実務において生成AIを活用し、業務上必要 となる一般的な知見や社内知見を参照する ことを目的としたユースケース 助言型ユースケース 作業補助型ユースケース 業務代替型ユースケース 業務① 業務② 業務③ 業務プロセス 業務① 業務② 業務③ 業務プロセス 業務① 業務② 業務③ 業務プロセス 生成AI 利用者 参照 助言 生成AI 作成依頼 作成物の生成 生成AI 代替 代替 代替 生成AIが得意 既存IT／AIとの組み合わせ ユースケ―ス検討 利用者
50. PwC 27 生成AI特有の導入効果 一連の業務プロセスのある業務における効果として算出されることが多いが、業務プロセス全体としての効 果を測定する必要があり、生成AIの技術的特徴から定性的な効果も大きいため、導入効果の算定に非財 務データなども考慮することもあり得る 導入効果の算定 業務プロセス全体としての導入効果※ 定性的な導入効果※ 業務① 業務② 業務③ 対象の 業務プロセス 生成AI 現状の語られている導入効果 生成AI導入により 業務①が14%②が35％効率化 （例：解決策調査効率化、顧客対応効率化など） 本来語るべき導入効果 生成AI導入により 業務プロセスが16％効率化 （例：ソフトウェア企業カスタマーサポートなど） ➢ 特定業務の生産性向上のみに着目せず、一連の 業務プロセス全体としての効果算定が重要 業務プロセス全体で は効果がない可能性 生成AIによる定性的な効果例 ※ 出典：Stanford Graduate School of Business、Generative AI Can Boost Productivity Without Replacing Workers https://www.gsb.stanford.edu/insights/generative-ai-can-boost-productivity-without-replacing-workers ➢ 導入による間接的なROIへの影響も考慮が必要 生産性 14%up/h 生産性 35%up/h 効率化 変更なし 導入 生成AIが業務の生産性や品質を助けることで顧客および 従業員の満足度が高まり、結果として離職率低減に繋がる 新技術は労働者への教育・業務プロセスおよびその実装 に時間を費やすが、生成AIは停滞なく数カ月で生産性を高 められる少ない苦労で大きな成果） 社歴が浅いスタッフが生成AIの助けを借りることで、生成AI を使わない社歴の長いスタッフと同等のパフォーマンスを 発揮
51. PwC 生成AIを活用する場合のリスク 28 生成AIを導入するにあたって、技術的・法的・倫理的観点を総合的に考慮したリスク対応が必要である AIデバイド 生成AIに対する期待値や精通度が利用者によって異な るため、デバイドが発生する可能性がある プロンプトインジェクション 利用者が悪意あるプロンプトを入力し、生成AIに有害な 生成物（機密情報、違法コンテンツ、虚偽情報、差別表 現など）を生成させる攻撃をする可能性がある 有害コンテンツ生成 生成物に含まれる情報に、有害な情報が含まれる可能 性がある 説明可能性・説明責任 生成AIではアルゴリズムが複雑となっており意思決定 プロセスがブラックボックス化することにより、説明責任 を果たせない可能性がある 著作権侵害 生成物に関する著作権の扱いは、通常の著作権侵害と 同様であり、依拠性や類似性が認められれば、著作権 侵害の対象となる可能性がある ハルシネーション（虚偽情報の生成） 生成物に含まれる情報に、虚偽の情報や真偽が不明 瞭な情報や金融犯罪や相場操縦などの金融市場の変 動を企図した情報が含まれる可能性がある データセットの確保・選定・評価 適切にインプットデータを選択、評価、確保しないと目的 のモデル能力を得られない 機密情報の生成 機密情報を学習していた生成AIが、プロンプトインジェ クションなどを介して、機密情報を生成してしまう可能 性がある データプライバシー・セキュリティ 顧客情報や取引データ、経営情報などから機微な情報 が漏洩したり推測されたりする可能性がある ルールブックを企業が許容できないリスク 生成AIに施された倫理ポリシーと、利用者の組織の ルールブックとの不一致が発生し、システム、アプリ ケーションなどでの対応が必要になる AI規制法 リスクの高いユースケース／プロセス（信用リスクなど） は、各国の規制法において、厳しく規制される可能性が ある 倫理的リスク 法的リスク 技術的リスク 技術的・法的・倫理的観点を踏まえたリスク例 生成AIのリスク
52. PwC 日本政府のAI戦略会議は2024年4月19日に「AI事業者ガイドライン（第1.0版)」を公表。生成AIに関する国 際的なルールの検討を行うための「広島AIプロセス」でとりまとめられた国際指針・行動規範が反映されて いる AI事業者ガイドライン 生成AIを取り巻く環境 AI事業者ガイドラインの位置づけ※ AI事業者ガイドラインの構成※ 法的拘束力はないが、ガイドラインに準拠していなかったために、一般市民の権利を不当に侵害してしまう、 対外的な説明責任を十分に果たすことができないといった場合には、ブランド棄損を招いてしまうという懸念が生じる 29 Why (基本理念) What (原則) How (実践) 本編 別添 (付属資料) はじめに 第1部 AIとは 第2部 AIにより目指すべき社会及び各主体が取り組む事項 A. 基本理念 B. 原則 C. 共通の指針 D. 高度なAIシステムに関係する事業者に共通の指針 E. AIガバナンスの構築 第3部 AI開発者に関する事項 第4部 AI提供者に関する事項 第5部 AI利用者に関する事項 (各主体の取組事項の解説などを記載) 全員が参照すべき 共通指針 主体別の指針 ※出典：経済産業省 AI事業者ガイドライン第1.0版
53. www.pwc.com/jp Thank you © 2024 PwC Consulting LLC. All rights reserved. PwC refers to the PwC network member firms and/or their specified subsidiaries in Japan, and may sometimes refer to the PwC network. Each of such firms and subsidiaries is a separate legal entity. Please see www.pwc.com/structure for further details. This content is for general information purposes only, and should not be used as a substitute for consultation with professional advisors.
54. Integrating Artificial Intelligence into Index Design Jason Ye, CFA Director, Strategy Indices May 23rd, 2024 S&P Dow Jones Indices – For Financial Professionals Only
55. S&P Dow Jones Indices – For Financial Professionals Only 2 AI ML NLP ChatGPT Neural Networks What is Artificial Intelligence (AI)
56. S&P Dow Jones Indices – For Financial Professionals Only 3 Artificial Intelligence Applications in Investment Management Alpha Generation Cross-sectional Stock Selection Earnings / Valuations Prediction Multi-Factor Timing / Integration Thematic Construction Summarization Sentiment Analysis Thematic Extraction Trading Executing & Analytics Trading Strategy Design Trading Algorithm Optimization Trading Cost Analysis (TCA)
57. S&P Dow Jones Indices – For Financial Professionals Only What is Kensho? Recognized as one of the most innovative artificial intelligence companies within Fintech Over five years, the team at Kensho built analytical products used by some of the world’s biggest financial institutions and the U.S. Intelligence Community. Within S&P Dow Jones Indices, Kensho: § offers artificial intelligence-led index creation and § represents a family of innovative thematic indices. Founded out of Harvard and MIT in 2013, received backing from Google Ventures ， Goldman Sachs, BOA, JPMorgan. S&P Global acquired Kensho Technologies in April 2018. TECH PIONEER World Economic Forum AI 100 CB INSIGHTS #1 FINTECH 50 Forbes TOP 5 HOTTEST IN FINTECH Fortune Best AI Technology Provider Waters Rankings 4
58. S&P Dow Jones Indices – For Financial Professionals Only The Kensho Index Construction Process + = CUTTING EDGE TECHNOLOGY INNOVATIVE METHODOLOGY UNIQUE INSIGHTS 5
59. S&P Dow Jones Indices – For Financial Professionals Only 6 Introducing S&P Kensho New Economies Composite Index Kensho Subsectors Kensho Sectors SPACE 1 ROBOTICS 2 CLEAN ENERGY 3 CYBER SECURITY 4 SMART BORDERS 5 WEARABLES 6 DRONES 7 CLEANTECH 8 VIRTUAL REALITY 9 SMART BUILDINGS 10 AUTONOMOUS VEHICLES 11 DIGITAL COMMUNITIES 12 ENTERPRISE COLLAB. 13 SUSTAINABLE FARMING 14 GENETIC ENGINEERING 15 3D PRINTING 16 NANO TECHNOLOGY 17 SMART FACTORIES 18 SMART GRIDS 19 ADVANCED TRANSPORT SYSTEMS 20 ELECTRIC VEHICLES 21 ALTERNATIVE FINANCE 22 FUTURE PAYMENTS 23 DISTRIBUTED LEDGER 24 DIGITAL HEALTH 25 Intelligent Infrastructure 10 19 Smart Transportation 7 20 11 21 Future Security 1 4 2 5 6 9 7 Clean Power 3 8 Final Frontiers 1 7 Future Communication 9 13 12 Democratized Banking 22 24 23 Human Evolution 2 9 6 15 17 16 Advanced Manufacturing 2 16 9 18 Sustainable Staples 1 7 2 14 16 15
60. S&P Dow Jones Indices – For Financial Professionals Only Capture Early-stage Potential with Natural Language Processing Skating to where the puck is going, rather than where it’s been • Apple first mentions ‘Mobile’ in filings in 2006, iPhone launches in 2007 • <10% of revenue until 2009 – for an interim annualized return of 56% vs 22% following • Alternative data provides opportunities to look ahead and capture the full upside 7
61. S&P Dow Jones Indices – For Financial Professionals Only Evolution of Kensho Kensho 2.0 Kensho 1.0 Universe • Global stocks across 104 exchanges spanning 89 countries • U.S. listed stocks across 5 exchanges Data source • S&P Global Market Intelligence’s internal database • SEC’s EDGAR repository Documents • Non-US regulatory filings,10Ks, 20Fs, 40Fs and S1s • 10Ks, 20Fs, 40Fs and S1s Document Languages • English and English translations • English Document Formats • No specific format • SEC’s preset formats for 10Ks, 20Fs, 40Fs ML Framework • Elastic Search + Language Model • Elastic Search + Recurrent Neural Networks (RNN) Core/Non-Core Classification • Yes • Yes Human Oversight • Yes • Yes Reconstitution • Annually • Annually Source: S&P Dow Jones Indices 8
62. MOONSHOT S&P Dow Jones Indices – For Financial Professionals Only 9 Kensho-Powered Thematic Indices METAVERSE HYDROGEN ECONOMY ARTIFICIAL INTELLIGENCE
63. S&P Dow Jones Indices – For Financial Professionals Only 10 Value Chain of Artificial Intelligence (AI) Software & Algorithms Solution Large Language Model, Generative AI, Deep Learning Chips & Hardware GPUs, CPUs Infrastructure Cloud, Edge and Hybrid Computing, Big Data Technology Data Providers Data curation and data management providers AI Applications AI-as-a-service platforms, cloud-based machine learning development platforms AI Enablers Finance & Insurance . Automobile Industrials, Robotics & Logistics Digital Community Aerospace & Astronomy AI Adopters
64. S&P Dow Jones Indices – For Financial Professionals Only 11 S&P Kensho Artificial Intelligence Enablers (KAIGPT) MARKET CAP 58.4% Large Cap (>$10bn) 23.0% Mid Cap ($2bn-$10bn) 18.6% Small Cap (<=$2bn) COUNTRY 84.8% United States 6.2% China 4.9% Israel 1.8% Canada 2.3% Others GICS SECTOR 85.2% Information Technology 6.8% Industrials 5.5% Communication Services 2.5% Consumer Discretionary Breakdown (as of April 30, 2024) 10 LARGEST COMPANIES Microsoft Corp Nvidia Corp Alphabet Inc C Broadcom Inc Oracle Corp Advanced Micro Devices SAP SE ADR Adobe Inc. Alibaba Group Holding Ltd ADR Intl Business Machines Corp 37 Companies 262B Weighted Avg Market Cap 17% Annualized Return since Inception AI-related hardware AI software developers AI infra- structure services AI application enablers AI training data providers Companies developing and enabling AI 0 50 100 150 200 250 300 350 2018 2019 2020 2021 2022 2023 Performance of the S&P Kensho Artificial Intelligence Enablers Index (USD total return - 5/31/2018-4/30/2024) S&P 500 TR S&P Kensho Artificial Intelligence Enablers Index Source: S&P Dow Jones Indices and/or its affiliates. Data as of April 30, 2024. Total Return in USD version of the indices are used. Charts are provided for illustrative purposes. Past performance is no guarantee of future results. These charts may reflect hypothetical historical performance. Please see the Performance Disclosure at the end of this document for more information about index returns, including hypothetical performance.
65. S&P Dow Jones Indices – For Financial Professionals Only 12 S&P Kensho Artificial Intelligence Enablers & Adopters (KAIEAT) Companies developing and enabling AI, as well as benefit from AI adoption Breakdown (as of April 30, 2024) 77 Companies 952B Weighted Avg Market Cap 16.9% Annualized Return since Inception MARKET CAP 98.7% Large Cap (>$10bn) 1.2% Mid Cap ($2bn-$10bn) 0.1% Small Cap (<=$2bn) COUNTRY 89.2% United States 3.3% Japan 2.6% China 0.9% Italy 0.2% Israel 3.9% Others GICS SECTOR 53.5% Information Technology 21.7% Communication Services 12.4% Consumer Discretionary 5.2% Financials 3.6% Industrials 3.5% Others AUTONOMOUS VEHICLES 11 ROBOTICS 2 SPACE 1 SMART FACTORIES 18 ENTERPRISE COLLAB. 13 DIGITAL COMMUNITIE S 12 ALTERNATIVE FINANCE 22 0 50 100 150 200 250 300 2018 2019 2020 2021 2022 2023 Performance of the S&P Kensho Artificial Intelligence Enablers & Adopters Index (USD total return - 6/15/2018-4/30/2024) S&P 500 TR S&P Kensho Artificial Intelligence Enablers & Adopters Index 10 LARGEST COMPANIES Microsoft Corp Nvidia Corp Meta Platforms, Inc. Class A Alphabet Inc C Broadcom Inc Tesla, Inc Mastercard Inc A Toyota Motor Corp ADR Oracle Corp Advanced Micro Devices Source: S&P Dow Jones Indices and/or its affiliates. Data as of April 30, 2024. Total Return in USD version of the indices are used. Charts are provided for illustrative purposes. Past performance is no guarantee of future results. These charts may reflect hypothetical historical performance. Please see the Performance Disclosure at the end of this document for more information about index returns, including hypothetical performance.
66. S&P Dow Jones Indices – For Financial Professionals Only 13 Additional Research • S&P DJI Kensho Goes Global: –https://www.indexologyblog.com/2024/02/09/sp-dji-kensho-goes-global/ • S&P Kensho Artificial Intelligence Enablers & Adopters Index Selection Process Overview: –https://www.spglobal.com/spdji/en/education/article/sp-kensho-artificial-intelligence-enablers- adopters-index-selection-process-overview/ • The Case for Indexing Thematics with the S&P Kensho New Economies: –https://www.spglobal.com/spdji/en/research/article/the-case-for-indexing-thematics-with-the-sp- kensho-new-economies • Handbook of Artificial Intelligence and Big Data Applications in Investments: –https://rpc.cfainstitute.org/en/research/foundation/2023/ai-and-big-data-in-investments- handbook
67. https://www.spglobal.com/spdji/en/ For more information S&P Dow Jones Indices – For Financial Professionals Only 14
68. © 2024 S&P Dow Jones Indices. All rights reserved. S&P, S&P 500, SPX, SPY, The 500, US500 , US 30, S&P 100, S&P COMPOSITE 1500, S&P 400, S&P MIDCAP 400, S&P 600, S&P SMALLCAP 600, S&P GIVI, GLOBAL TITANS, DIVIDEND ARISTOCRATS, Select Sector, S&P MAESTRO, S&P PRISM, S&P STRIDE, GICS, SPIVA, SPDR, INDEXOLOGY, iTraxx, iBoxx, ABX, ADBI, CDX, CMBX, MBX, MCDX, PRIMEX, HHPI and SOVX are trademarks of S&P Global, Inc. (“S&P Global”) or its affiliates. DOW JONES, DJIA, THE DOW and DOW JONES INDUSTRIAL AVERAGE are trademarks of Dow Jones Trademark Holdings LLC (“Dow Jones”). These trademarks together with others have been licensed to S&P Dow Jones Indices LLC. Redistribution or reproduction in whole or in part are prohibited without written permission of S&P Dow Jones Indices LLC. This document does not constitute an offer of services in jurisdictions where S&P Dow Jones Indices LLC, S&P Global, Dow Jones or their respective affiliates (collectively “S&P Dow Jones Indices”) do not have the necessary licenses. Except for certain custom index calculation services, all information provided by S&P Dow Jones Indices is impersonal and not tailored to the needs of any person, entity or group of persons. S&P Dow Jones Indices receives compensation in connection with licensing its indices to third parties and providing custom calculation services. Past performance of an index is not an indication or guarantee of future results. It is not possible to invest directly in an index. Exposure to an asset class represented by an index may be available through investable instruments based on that index. S&P Dow Jones Indices does not sponsor, endorse, sell, promote or manage any investment fund or other investment vehicle that is offered by third parties and that seeks to provide an investment return based on the performance of any index. S&P Dow Jones Indices makes no assurance that investment products based on the index will accurately track index performance or provide positive investment returns. S&P Dow Jones Indices LLC is not an investment advisor, and S&P Dow Jones Indices makes no representation regarding the advisability of investing in any such investment fund or other investment vehicle. A decision to invest in any such investment fund or other investment vehicle should not be made in reliance on any of the statements set forth in this document. S&P Dow Jones Indices is not an investment adviser, commodity trading advisor, commodity pool operator, broker dealer, fiduciary, promoter” (as defined in the Investment Company Act of 1940, as amended), “expert” as enumerated within 15 U.S.C. § 77k(a) or tax advisor. Inclusion of a security, commodity, crypto currency or other asset within an index is not a recommendation by S&P Dow Jones Indices to buy, sell, or hold such security, commodity, crypto currency or other asset, nor is it considered to be investment advice or commodity trading advice. These materials have been prepared solely for informational purposes based upon information generally available to the public and from sources believed to be reliable. No content contained in these materials (including index data, ratings, credit-related analyses and data, research, valuations, model, software or other application or output therefrom) or any part thereof (“Content”) may be modified, reverse-engineered, reproduced or distributed in any form or by any means, or stored in a database or retrieval system, without the prior written permission of S&P Dow Jones Indices. The Content shall not be used for any unlawful or unauthorized purposes. S&P Dow Jones Indices and its third-party data providers and licensors (collectively “S&P Dow Jones Indices Parties”) do not guarantee the accuracy, completeness, timeliness or availability of the Content. S&P Dow Jones Indices Parties are not responsible for any errors or omissions, regardless of the cause, for the results obtained from the use of the Content. THE CONTENT IS PROVIDED ON AN “AS IS” BASIS. S&P DOW JONES INDICES PARTIES DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR DEFECTS, THAT THE CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED OR THAT THE CONTENT WILL OPERATE WITH ANY SOFTWARE OR HARDWARE CONFIGURATION. In no event shall S&P Dow Jones Indices Parties be liable to any party for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including, without limitation, lost income or lost profits and opportunity costs) in connection with any use of the Content even if advised of the possibility of such damages. S&P Global keeps certain activities of its various divisions and business units separate from each other in order to preserve the independence and objectivity of their respective activities. As a result, certain divisions and business units of S&P Global may have information that is not available to other business units. S&P Global has established policies and procedures to maintain the confidentiality of certain non-public information received in connection with each analytical process. In addition, S&P Dow Jones Indices provides a wide range of services to, or relating to, many organizations, including issuers of securities, investment advisers, broker-dealers, investment banks, other financial institutions and financial intermediaries, and accordingly may receive fees or other economic benefits from those organizations, including organizations whose securities or services they may recommend, rate, include in model portfolios, evaluate or otherwise address.. S&P Dow Jones Indices – For Financial Professionals Only 15 General Disclaimer
69. The S&P Kensho Artificial Intelligence Enablers Index was launched on April 21, 2023. The S&P Kensho Artificial Intelligence Enablers & Adopters Index was launched in Sep 25, 2023. All information presented prior to an index’s Launch Date is hypothetical (back-tested), not actual performance, and is based on the index methodology in effect on the index launch date. However, when creating back-tested history for periods of market anomalies or other periods that do not reflect the general current market environment, index methodology rules may be relaxed to capture a large enough universe of securities to simulate the target market the index is designed to measure or strategy the index is designed to capture. For example, market capitalization and liquidity thresholds may be reduced. In addition, forks have not been factored into the back-test data with respect to the S&P Cryptocurrency Indices. For the S&P Cryptocurrency Top 5 & 10 Equal Weight Indices, the custody element of the methodology was not considered; the back-test history is based on the index constituents that meet the custody element as of the Launch Date. Complete index methodology details are available at spglobal.com/spdji. Back-tested performance reflects application of an index methodology and selection of index constituents with the benefit of hindsight and knowledge of factors that may have positively affected its performance, cannot account for all financial risk that may affect results and may be considered to reflect survivor/look ahead bias. Actual returns may differ significantly from, and be lower than, back-tested returns. Past performance is not an indication or guarantee of future results. Please refer to the methodology for the Index for more details about the index, including the manner in which it is rebalanced, the timing of such rebalancing, criteria for additions and deletions, as well as all index calculations. Back-tested performance is for use with institutions only; not for use with retail investors. S&P Dow Jones Indices defines various dates to assist our clients in providing transparency. The First Value Date is the first day for which there is a calculated value (either live or back-tested) for a given index. The Base Date is the date at which the index is set to a fixed value for calculation purposes. The Launch Date designates the date when the values of an index are first considered live: index values provided for any date or time period prior to the index’s Launch Date are considered back-tested. S&P Dow Jones Indices defines the Launch Date as the date by which the values of an index are known to have been released to the public, for example via the company’s public website or its data feed to external parties. For Dow Jones-branded indices introduced prior to May 31, 2013, the Launch Date (which prior to May 31, 2013, was termed “Date of introduction”) is set at a date upon which no further changes were permitted to be made to the index methodology, but that may have been prior to the Index’s public release date. Typically, when S&P DJI creates back-tested index data, S&P DJI uses actual historical constituent-level data (e.g., historical price, market capitalization, and corporate action data) in its calculations. As ESG investing is still in early stages of development, certain datapoints used to calculate S&P DJI’s ESG indices may not be available for the entire desired period of back-tested history. The same data availability issue could be true for other indices as well. In cases when actual data is not available for all relevant historical periods, S&P DJI may employ a process of using “Backward Data Assumption” (or pulling back) of ESG data for the calculation of back-tested historical performance. “Backward Data Assumption” is a process that applies the earliest actual live data point available for an index constituent company to all prior historical instances in the index performance. For example, Backward Data Assumption inherently assumes that companies currently not involved in a specific business activity (also known as “product involvement”) were never involved historically and similarly also assumes that companies currently involved in a specific business activity were involved historically too. The Backward Data Assumption allows the hypothetical back-test to be extended over more historical years than would be feasible using only actual data. For more information on “Backward Data Assumption” please refer to the FAQ. The methodology and factsheets of any index that employs backward assumption in the back-tested history will explicitly state so. The methodology will include an Appendix with a table setting forth the specific data points and relevant time period for which backward projected data was used. Index returns shown do not represent the results of actual trading of investable assets/securities. S&P Dow Jones Indices maintains the index and calculates the index levels and performance shown or discussed but does not manage actual assets. Index returns do not reflect payment of any sales charges or fees an investor may pay to purchase the securities underlying the Index or investment funds that are intended to track the performance of the Index. The imposition of these fees and charges would cause actual and back-tested performance of the securities/fund to be lower than the Index performance shown. As a simple example, if an index returned 10% on a US $100,000 investment for a 12-month period (or US $10,000) and an actual asset-based fee of 1.5% was imposed at the end of the period on the investment plus accrued interest (or US $1,650), the net return would be 8.35% (or US $8,350) for the year. Over a three-year period, an annual 1.5% fee taken at year end with an assumed 10% return per year would result in a cumulative gross return of 33.10%, a total fee of US $5,375, and a cumulative net return of 27.2% (or US $27,200). S&P Dow Jones Indices – For Financial Professionals Only 16 Performance Disclosure/Back-Tested Data
70. The AI-Driven Future of Finance: Opportunities & Challenges Lucas Lepak Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
71. How should we address the hype factor around Generative AI (GenAI)? 2 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
72. 3 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only GenAI could add $200-340B in value annually to banking Source: McKinsey & Company (2023) Software, hardware, and services for AI will be $400B in 2027, financial sector spending will more than double to $97B in 2027 (29% CAGR) Source: International Data Corp (2023) Global AI investment could be $200B by 2025 Source: Goldman Sachs (2023) “Generative AI could increase global GDP by $7-10 trillion, or by as much as 10%” Source: J.P. Morgan Research (2024)
73. GenAI tops the Peak of Inflated Expectations in Gartner’s 2023 Hype Cycle for Emerging Tech Source: Gartner (2023) “Emergent AI Will Have a Profound Impact on Business and Society” 4 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
74. How can we break through the hype? 5 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
75. Generative AI is the latest milestone in an evolving technology landscape 6 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only Cloud Computing Using remote servers for storing and accessing data/services Big Data Extremely large and complex datasets requiring specialized tools Deep Learning Advanced machine learning inspired by the brain's neural networks 2006 2011 2015 GenAI AI that can create new content like images or text 2020
76. The lifespan and magnitude of technology trends is reflected in Google search volume Google Trends (United States) - Search term interest over time scaled relative to peak popularity 7 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only 0 20 40 60 80 100 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 Cloud Computing Big Data Deep Learning 2006 AWS S3 & EC2 2011 Hadoop 1.0 2015 TensorFlow 2020 GPT-3
77. The lifespan and magnitude of technology trends is reflected in Google search volume Google Trends (United States) - Search term interest over time scaled relative to peak popularity 8 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only 0 20 40 60 80 100 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 Cloud Computing Big Data Deep Learning Machine Learning 2006 AWS S3 & EC2 2011 Hadoop 1.0 2015 TensorFlow 2020 GPT-3
78. The lifespan and magnitude of technology trends is reflected in Google search volume Google Trends (United States) - Search term interest over time scaled relative to peak popularity 9 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only 0 20 40 60 80 100 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 Cloud Computing Big Data Deep Learning Machine Learning GPT 2006 AWS S3 & EC2 2011 Hadoop 1.0 2015 TensorFlow 2022 ChatGPT 2020 GPT-3
79. Our current GenAI moment reflects a shift in perception as much as a growth in technology 10 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
80. We’ve educated and empowered all employees to access GenAI and have enabled this via robust governance and monitoring systems 11 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
81. Why FactSet is your GenAI partner 12 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only 1. TRUST: 40+ years as your partner delivering financial data, analytics and open technology to 207K+ users across 8K+ institutional clients globally, 23M+ portfolios, and >90% client retention. 2. RESPONSIBLE AI: Strong ethical foundation committed to address hallucinations and unintended consequences. 3. FACT BASED AI: Models depend on quality and breadth of reliable data. FactSet delivers best-of-breed data with smart, connected data models for seamless connection and integration. Fact-based training data is essential. 4. AI/ML EXPERIENCE: AI/ML team founded 15+ years ago and created a suite of impactful AI solutions available in FactSet today. Leveraging large language models (LLMs) since 2018 and GPT4.0 solutions already in production. 5. ROI: FactSet can deliver for you or partner with you, whatever works best. The maintenance of GenAI solutions can be handled by FactSet with economies of scale across our client base. 6. EARLY ACCESS & CO-DEVELOPMENT: Benefits include early access to new solutions and enhancements and the ability to co-develop bespoke offerings.
82. Security + Governance www.factset.com/ai 13 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
83. Introducing FactSet’s AI Blueprint FactSet is investing to (1) reimagine your workstation experience with GenAI-powered discoverability, (2) automate your workflows to drive productivity, (3) power your GenAI chatbots 14 Copyright © 2023 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only 2: “Mile Deep” Workflow Automation 3: “Mile High” Innovation Acceleration 1: “Mile Wide” Discoverability Transform the usability and discoverability of the FactSet workstation experience with a two- way chat interface, FactSet Mercury Elevate your FactSet experience with tailored AI-augmented workflows for Research, Asset Management, Banking & Wealth Management Power your GenAI workflows and chatbots with FactSet’s best-in- class, connected data, along with access to our Conversational API, powered by FactSet Mercury.
85. 16 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only GenAI applications combine PRODUCT + DESIGN + ENGINEERING
86. Webb’s Depth of Knowledge framework assesses the level of cognitive complexity for a task Level 1: Recall Who, What, Where? Level 2: Basic Reasoning How did it take place? Level 3: Strategic Thinking How can I make use of it to solve a problem? Level 4: Extensive Thinking What will or could happen? 17 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only Source: https://www.structural-learning.com/post/webbs-depth-of-knowledge
87. Gen AI products pitched as a single magic surface are often experienced more like a single slice of Swiss cheese 18 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
88. 19 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only GenAI engineering projects that begin with a large stack of LLM requests may end by optimizing many of them away
89. The promise of GenAI outshines the hype 20 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only
90. 21 Copyright © 2024 FactSet Research Systems Inc. All rights reserved. FactSet Business Use Only May 28, 2024 | Tokyo Kaikan Join us to learn more about “The Evolution of the Buy-Side Workflows” and discuss key developments within the financial industry, from technological innovations to new approaches in financial operating workflows, as well as the latest enhancements to FactSet’s award-winning solutions. https://www.factset.com/apacforum2024/tokyo
91. テキストデータによるESG評価 • 企業に関するテキストデータや数値データ等を入手し、ESG (Environment, Social, Governance)に関する14のトピックの指数を生成 • 今までは、企業のESG評価は人が評価していたため、バイアスが含まれてい たが、客観的なビックデータをもとに評価することで客観性を担保 (Insight360) 1 Environment Social Governance 指標 プレスリリース ニュース記事 報告書
92. WORD2VEC BERT GPT3.5 GPT4 大規模言語モデル(LLM) • 大規模言語モデル（LLM：Large Language Model）とは、大量の テキストデータと深層学習技術に用いて構築された、機械学習の 自然言語処理モデル • テキスト分類・生成や感情分析、文章要約、質問応答などに応用 2 パラメータ数 (＝モデルの複雑さ) 数十万 数億 3550億 １兆?
93. 大規模言語モデル(LLM)はESG評価に 利用できるのか？ 和泉 潔(東京大学大学院 工学系研究科) 3 ESG分析で使えそうな可能性 1. 情報の分類・探索 • [小杉ら2023] [Mishra2023] [Zouら2023] 2. スコアリング・レーティング • [濱田ら2023] [大坪ら2023] 直接的に将来のインパクトを予測するのは 不向きである
94. 情報の分類・探索 • ニュース記事からESG関連の情報抽出 [小杉ら2023] 4 Q A （条件指示）それぞれのトピックを20 文字以内で出力することを指示 （入力）日本語ロイターニュースから環境分野に関する記事を抽出したサ スティナビリティレポート （最終出力）例示に従って、抽象度を段階的に変化させた3つのトピックを 出力することを指示 L1（最も抽象度が低いトピック） 具体的な企業名や取り組み対象地域（固有名詞）と取り組み内容の両 方を含む L2（２番目に抽象度が高いトピック） 抽象的な企業名や取り組み対象地域（非固有名詞）と取り組み内容を両 方含む L3（最も抽象度が高いトピック） 取り組み内容のみを含む
95. 情報の分類・探索 • ニュース記事からESG関連の情報抽出 [小杉ら2023] 5 トピック 1 [ロイター2020/4/16 ] L1：三井住友フィナンシャルグループの石炭火力発電への融資停止方針 L2：日本の金融機関の気候変動対策の強化 L3：環境問題への取り組み トピック 2 [ロイター2020/1/22 ] L1：三菱自動車の排ガス規制違反に関する捜査 L2：自動車メーカーによる排ガス規制違反の問題 L3：環境への悪影響を引き起こす排ガス規制違反の取り締まり トピック 3 [ロイター2020/12/9 ] L1：トヨタ自動車による水素燃料電池車の発売と技術支援計画 L2：自動車メーカーの水素技術とカーボンニュートラル目標 L3：新型コロナウイルスの感染拡大と感染防止策についての見解 抽出例 ◯ 適切 × 抽象度変化なし × L3で別の内容
96. スコアリング・レーティング • サステナビリティ報告書からESGに関する取組の情報開示を評価 [濱田ら2023] • ChatPDFを使って、報告書のPDFについて以下を質問 6 Q A 必ず最初にYesかNoで答えてください。回答には、定量情報とページ番 号を必ず含めてください。この報告書には、正社員の割合に関する情報 が記載されていますか？ Yes、報告書には正社員の割合に関する情報が記載されています。正社 員比率は127ページに記載されています。報告書によると、2022年の正 社員比率は管理職で89.1％、非管理職で90.0％でした[p128]. ESG評価指標(同志社大学 社会価値研究センター) 0 記述されていない 1 簡潔に記述。定性情報のみ。 2 定量情報が記述。測定できない情報。 3 定量情報が記述。測定可能な情報。 情報開示スコア 正答率 0.78
97. 参考文献 • [小杉ら2023] 小杉 他, ChatGPTを用いたニュース記事のESGトピック分 析, https://doi.org/10.11517/jsaisigtwo.2023.FIN-031_36 • [Mishra2023] ESG Impact Type Classification: Leveraging Strategic Prompt Engineering and LLM Fine-Tuning, https://aclanthology.org/2023.finnlp-2.11/ • [Zouら2023] ESGReveal: An LLM-based approach for extracting structured data from ESG reports, https://arxiv.org/abs/2312.17264 • [濱田ら2023] 濱田 他, 大規模言語モデルを活用したESG評価, https://doi.org/10.11517/jsaisigtwo.2023.FIN-032_45 • [大坪ら2023] 大坪 他, 大規模言語モデルを用いた事業文書からESG 指標への影響推論及び統計的因果推論との関係の検証 https://doi.org/10.11517/pjsai.JSAI2023.0_3Xin429 7
98. 8 LLMによる 将来のインパクト予測が難しい理由 • LLMの技術的な課題 LLMのハルシネーション(幻想) • 経済現象の構造による課題 経済現象は多数による人気投票である
99. 9 LLMによる 将来のインパクト予測が難しい理由 • LLMの技術的な課題 LLMのハルシネーション(幻想) • 経済現象の構造による課題 経済現象は多数による人気投票である
100. LLMのハルシネーション(幻想) 10 生成AIが事実に基づかない嘘の情報を生成することがある 〇〇社の製品には有害な物質が含まれていたことがありました。 A 〇〇社の工場では違法な未成年労働が行われていました。 A 〇〇社のCEOの◎◎氏は辞任しました。 A 事実でない情報に基づく回答 専門的に間違った回答 A 原油価格が上がると、一般的には株価には上昇圧力がかかることがあり ます 一般的には原材料コストの上昇により株価が下がる ウソ
101. 事実でない情報に基づく回答への対策 11 外部知識を利用してプロンプト (LLMへの入力文)に関連情報を追加する Retrieval-Augmented Generation（RAG） LLM 〇〇社のCEOは誰 元の質問文 〇〇社のCEOは誰 LLMに入力する文章 〇〇社のCEOリスト 2012-2017 △△ 2018-2023 □□ 2024- ◎◎ 企業 データ ベース 〇〇社のCEOの ◎◎さんは辞任 していません 〇〇社のCEOの◎◎氏は辞任しました。 A 質問文に 関連する情報 ＋
102. 専門的に間違った回答への対策 12 原油価格が上がると、一般的には株価には上昇圧力がかかることがあり ます A 経済等の専門テキストで 追加学習を行って専門分野の言語を強化する ファインチューニング 一般的には原材料コスト の上昇により株価が下 がる 友好的な人間関係、愛、好意、 結婚、平和、調和、分担、贈り 物、芸術的才能。 寝起きも超不機嫌な私は朝食 がないことで更に不機嫌に陥 るのだった コンビニでなんか 買って食べよかな 私は3泊。それで限界ぎりぎり だよー。よく頑張りましたな。 金融専門テキスト 一般テキスト 追加学習 一般的な LLM 金融特化 LLM
103. 13 LLMによる 将来のインパクト予測が難しい理由 • LLMの技術的な課題 LLMのハルシネーション(幻想) • 経済現象の構造による課題 経済現象は多数による人気投票である
104. 経済現象は人気投票である ◆ある現象が経済に与える影響は、一通りに決まってい ない ◆経済への影響は、その時の多数決で決定される ◆ブームや流行廃りが存在する→心理的な相互作用 14 利益の増加 株価の上昇 金融 引き締め 株価の下落 物価の上昇
105. LLMで心理的な相互作用を分析できるのか？ 15 LLMに個性を与える研究例 消費者モデル(マーケティング分野) Q A あなたは以下の属性を持った消費者です。 名前：〇〇 〇〇 性別：女性 年齢：35歳 職業：営業 趣味：… 性格：… あなたは、〇〇の新商品に関してどのような意見を持っていますか。 私は〇〇の商品をずっと使ってきましたが、今回の新商品は以前よりも 香りが改善されたように思います。ただ、他の似た商品に比べて価格が 少し高いように感じます。 …
106. Copyright© 2023 Tokio Marine Holdings, Inc. May 23, 2024 Yusuke Tomishima, CFA Tokio Marine HD Attempts to extract meaningful insights from ESG data using AI/ML
107. Copyright© 2023 Tokio Marine Holdings, Inc. Self-Introduction 2 Yusuke Tomishima, CFA ① Data Scientist at Financial Planning Department, Tokio Marine Holdings ⚫ Development of quant investment strategies using AI/Machine Learning, Alternative Data. ⚫ Agile application development in python for business efficiency. ⚫ Human resource development for quant talent. ② Visiting Professor at Tama Graduate School ⚫ I offer a course titled "Basic Finance Course for Asset Management." ③ Author ⚫ I am an author of books on various topics including asset management, mathematics, and physics.
108. Copyright© 2024 Tokio Marine Holdings, Inc. Introduction of my team (DX Team) at Tokio Marine HD 3 Project 1 Project 2 Project 3 TMHD DX Team ・・・ Project 3 Project 4 Project 5 ・・・ support / advice project management / direct participation DX Team has started at FY21 and provided a variety of solutions for the DX needs of asset management divisions. In Tokyo office, various DX projects are formed to address issues related to asset management. Project members are selected on a case-by-case, cross-functional basis, and the DX team acts as the CoE, providing management and support for these projects and sometimes participating as a project member.
109. Copyright© 2023 Tokio Marine Holdings, Inc. ESG trends in Japanese insurance industry Sustainable investment balance for each investment method in Japan (amount is in MM JPY) FY 2020 FY 2021 YOY ESG integration 204,958,018 422,115,459 +106.0% Negative/exclusionary screening 135,263,369 261,039,802 +93.0% Positive/best-in-class screening 14,643,189 24,867,183 +69.8% Sustainability-themed investing 7,988,505 10,665,994 +33.5% Impact/community investing 140,363 706,280 +403.2% Corporate engagement 187,170,342 261,495,512 +39.7% Net-Zero Asset Owner Alliance Net-Zero Insurance Alliance Increasing Net Zero Momentum Commit to transit the companies that they have invested to net-zero GHG emissions by 2050. Commit to transit their clients that they have (re- )insured to net-zero GHG emissions by 2050. Source:Japan Sustainable Investment Forum ⚫ The sustainable finance market in Japan is expanding significantly. In the insurance industry, many companies have positioned climate change as the top priority theme. Furthermore, efforts toward net zero, ESG-themed investment, and impact investment are increasing. ⚫ Since 2021, many insurance companies have joined the Net-Zero Initiatives based on Scope 3, considering to establish the GHG emission measurement system and set intermediate reduction targets.
110. Copyright© 2023 Tokio Marine Holdings, Inc. Introduction of the case at Tokio Marine: Hackathon Overview 5 • Organized by: Bloomberg L.P. • Event Summary: BQuant Hackathon is the Tech event organized by Bloomberg. Using BQuant, a Python- based coding platform provided by Bloomberg, participating teams develop data analysis applications on the theme of “ESG × Alternative Data” in three months. • Accomplishment In this presentation, we will introduce the ESG app developed by the Tokio Marine Team at BQuant Hackathon held in October 2022. Five teams from insurance companies and asset management companies participated in the event, and Tokio Marine Team developed the ESG scoring AI that bring us the “BQuant Mastermind” award. Bloomberg BQuant Hackathon Gamification is important!
111. Copyright© 2023 Tokio Marine Holdings, Inc. Hackathon: Our challenges and solutions 6 Challenges of existing ESG score Our solutions Challenges Solutions Transparency Machine learning is used to evaluate the materiality of each ESG indicator and calculate a unique ESG score based on fully quantitative criteria. No human judgement is needed. Coverage Connect the model to BBG data center so that the model can cover all companies worldwide that can retrieve data from BBG. Investor’s viewpoint Switch the training data reflecting the perspectives of investors. 1) Transparency 2) Coverage 3) Investor’s viewpoint The scoring logic of rating agency is a black box. Many companies not covered by rating agencies. Does not reflect the different perspectives of investors.
112. Copyright© 2023 Tokio Marine Holdings, Inc. ESG item Data type (1) Base Score* ([0, 1]) (2) Materiality (see p23-26) MARINE Score (1)×(2) XXX Yes or No 1.00 2.5 2.5 XXX Integer 0.73 8.9 6.5 XXX Real 0.25 3.3 0.8 ・・・ ・・・ ・・・ ・・・ ・・・ Sum 100 72 Example Calculation Hackathon: AI-based ESG scoring model “MARINE Score” ⚫ What is the MARINE Score? ⚫ MARINE Score is a proprietary ESG score calculated based on quantitative criteria without relying on human expert judges. ⚫ Quick scoring based on recent data without the need for qualitative judgments. ⚫ Wide coverage: It covers all companies worldwide that can retrieve data from BBG. ⚫ Score calculation method ⚫ Base scores are calculated for each ESG item based on disclosure status and comparison with other companies in the same industry (0 to 1 point for each item). ⚫ ESG Scores are calculated by summing the above base scores with the weight of materiality by country and industry. * Base Score Items indicating disclosure status: 1 point if it is disclosed, 0 point if not Other ESG items: Calc Z-score of the item of the company within the industry and convert it to a distribution of 0-1 points.
113. Copyright© 2023 Tokio Marine Holdings, Inc. Hackathon: AI-based ESG scoring model “MARINE Score” Features Financial Data ◼ Industries ◼ ROE ◼ ROA ◼ Net profit margin ◼ Interest-baring Debt , etc. ESG Data ◼ GHG intensity ◼ Pct women in Mgmt ◼ Employee turnover ◼ Pct outside directors ◼ Avg age of officers , etc. Train the AI given the dummy task of “Find the company with positive future return of selected KPI”. Gini importance Training Data MARINE Score Binary data KPI future return is positive: 1 KPI future return is negative: 0 KPI candidates (select one) ◼Profitability (ROE, ROA) ◼Valuation (PBR, PER) ◼Credit (D/E ratio) Feature Materiality ⚫ Assessing corporate ESG initiatives with AI-based scoring: “Marine Score”. ⚫ A unique feature of this scoring algorithm is we can switch the training data reflecting the perspectives of investors. ex) Bond managers may choose DE ratio, while Equity managers may choose ROE or PBR...
114. Copyright© 2023 Tokio Marine Holdings, Inc. Overview of the scoring AI Candidates Linearity F-score (test data) Accuracy (test data) Elastic Net Linear Very low Very low Random Forest Classifier Non-linear 0.3~0.4 0.8~0.9 Gradient Boosting Classifier Non-linear 0.4~0.6 0.8~0.9 XGBoost Classifier Non-linear 0.4~0.6 0.8~0.9 Light GBM Classifier Non-linear 0.6~0.9 0.8~0.9 ◼ Forecasting target and features ⚫ Forecasting target: Whether selected KPI (ex. ROE) improves 3 years later. ⚫ Features: ESG and Financial data (around 400 features) ◼ Tried 5 models and select Light GBM Classifier considering below ; ⚫ F-score on test data ⚫ Computational load ⚫ Generalization performance (F-scores of different countries/industries should be at high and similar levels.) Hackathon: AI-based ESG scoring model “MARINE Score”
115. Copyright© 2023 Tokio Marine Holdings, Inc. Job Training Women empowerment, D&I Comfortable and safety working Environment The materiality of the item accounts for 3.7 points out of a total of 100 points in the "S" category. Hackathon: AI-based ESG scoring model “MARINE Score” ⚫ Materiality is assigned a total of 100 points for each of E, S, and G, with points assigned according to the level of importance of each item. ⚫ The table on the right shows an example of materiality calculation* for "S" items in the capital goods sector (Industrials) when corporate value (PBR) is used as KPI. ⚫ Only the top 150 “S” related items are shown.The AI captured the characteristics of the industry, and as a result, items related to safety ranked high. In addition, universally important items such as women empowerment and diversity also ranked high. Example of materiality for "S" item (only top-level items are shown) * Universe: EuroStoxx600 constituents / Industry: Industrials / KPI: PBR
116. Copyright© 2023 Tokio Marine Holdings, Inc. Hackathon: AI-based ESG scoring model “MARINE Score” ⚫ Hierarchical display of materiality for easy understanding of the overall picture Example of UI for materiality evaluation 1) Breakdown of E/S/G/F ➢ E/S/G only ➢ E/S/G/F 2) Top-level items of E/S/G 3) Details of all items
117. Copyright© 2023 Tokio Marine Holdings, Inc. Hackathon: AI-based ESG scoring model “MARINE Score” Example of peer comparison using Marine Score ⚫ Compare scores with other companies in industry to understand strengths and weaknesses compared to peers. ⚫ The table below shows an example of comparison in Japanese retail industry, and we can find that the reason for ZOZO‘s low S-score can be attributed to the lack of progress in women empowerment (low number of female employees and female managers).
118. Copyright© 2023 Tokio Marine Holdings, Inc. Top 5 SUPPLIERS Ticker MARINE E Score Denso corp 6902 JT Equity 44 Aisin Corp 7259 JT Equity 40 Toyota Boshoku Corp 3116 JT Equity 31 JTEKT Corp 6473 JT Equity 40 Toyoda Gosei Co Ltd 7282 JT Equity 46 Average 40・・・(1) Toyota’s E score 58・・・(2) MARINE Score before considering supply chain level initiatives 49 （Down -9） Average of (1)&(2) Toyota’s E score Hackathon: AI-based ESG scoring model “MARINE Score” ⚫ Challenges ⚫ Ideally, ESG initiatives should be active not only within the company but also throughout the supply chain. ⚫ Therefore, we would like to quantify the status of the company’s ESG initiatives toward its major business partners and utilize this information in investment decisions and engagement. ⚫ Solutions ⚫ Using supply chain data provided by BBG, we can modify the Marine Score based on ESG initiatives at the supply chain level. ⚫ The example below shows the modification of Toyota’s Marine Scores based on the top five suppliers, which would be relatively easy for Toyota to encourage them to do something more for ESG. ⚫ Toyota Boshoku's score is low among major suppliers, indicating that there is room for further efforts.
119. Lee Davidson, CFA Real Applications for AI in Financial Data: Unveiling Morningstar’s Medalist Ratings, Intelligence Engine, and the Future for Data Collection
120. ©2024 Morningstar. All Rights Reserved. 2 The volume and complexity of data have surged exponentially, presenting both challenges and opportunities. Amidst this data deluge, AI emerges as a transformative force, reshaping how financial data is collected, analyzed, and leveraged. The Rise of AI in Financial Data
121. The industry’s rushing toward generative AI, but our machine learning efforts go back years. ©2024 Morningstar. All Rights Reserved. 3 How We’re Innovating Today Mo Our AI-powered digital research assistant answers questions from over 200,000 advisors every day. Medalist Ratings The Medalist Rating system lets us cover 10 times as many managed products than we could with Analyst Ratings alone. Data Ingestion Every month, we process data from 16,000 sources and bring in 200+ million documents.
122. ©2024 Morningstar. All Rights Reserved. 4 Automated reports for packaged retail and insurance-based investment products. § Faster process for 150,000 annual Packaged Retail and Insurance-based Investment Products Key Information Documents (PRIIP KIDs) § Large Language Model (LLM) collects data via text, images, and a tables § Uses GPT 3.5 Turbo on keywords and regular expressions in 20 languages § Saves approximately 2,000 hours annually Data Ingestion—More Data, Less Labor
123. ©2024 Morningstar. All Rights Reserved. 5 Behind the Scenes of Data Ingest Our Data Collections team acquires sophisticated, dispersed data from filings with Natural Language Parsing (NLP). PDF document Table Finetuned LLM (Mistral-7b) deployed at EC2 locate table with regex Azure OpenAI Service Database Consensus Solution 1 Solution 2 Human Validation Yes No
124. ©2024 Morningstar. All Rights Reserved. 6 ©2024 Morningstar. All Rights Reserved. 6 Human analysis and machine learning combine for powerful results at scale. § 28% growth of managed investment over the past 5 years. § 21,400 vehicles are covered by analysts and 383,0000 are covered by our model. § We use a random forest algorithm to fit a relationship between a fund’s pillar ratings and its attributes. § Gold, Silver, and Bronze medals include funds where 𝛼!"# > 0 X § 𝛼!"# = 𝛼$%&'' − fees § 𝛼$%&'' = Ψ( ( ) 0.45 x Process + 0.45 x People + 0.1 x Parent § Ψ( = peer group alpha poten_al Medalist Ratings—Simplifying Strategy Search and Selection
125. Training the Model ©2024 Morningstar. All Rights Reserved. 7 The Methodology for Medalist Ratings To make a prediction, take a new fund “x” with all the requisite data (alpha, tenure, fees, etc.) and feed it through each decision tree Each tree contributes one prediction Average all the predictions together (e.g. “vote”)—the class with the highest tally wins and provides the prediction (325 trees vote Positive, 110 vote Neutral, and 65 vote Negative → Positive) Making Predictions We use a Random Forest model composed of combined predictions from submodels primarily comprising regression or classification trees.
126. ©2024 Morningstar. All Rights Reserved. 8 Mo—A New Way to Query Morningstar’s Wealth of Insights Meet Mo, an AI-powered digital research assistant built on the Intelligence Engine platform. § Empowers users to engage with our trusted research and data through a modern AI chat experience. § Expands the context with client-specific research and documentation. § Enhanced accuracy and breadth of knowledge through retrieval paths across additional structured data (both Morningstar and client- owned). § Prompt engineering paired with custom content retrieval to customize for specific audiences (i.e. product-specific or client-specific).
127. ©2024 Morningstar. All Rights Reserved. 9 Anatomy of a Digital Research Assistant The brain of Mo is made up of several existing technologies, but it’s unique because of our content. Content Data Morningstar Intelligence Engine Content Retrieval (Weaviate) Create Embeddings Data APIs Retrieval Layer Process Layer Azure OpenAI Service Chat UI Intelligence Engine API
128. What’s Next? 10
129. We are enabling teams to get started with text, code, and image generation and building GenAI models and applications. ©2024 Morningstar. All Rights Reserved. 11 Play & Create: Constructing the AI Playground for Innovation Engineering Upskill developers with domain specific knowledge and skills for faster production. Sales Transcript summarization and email handling for a more efficient and consistent client experience. Investor Relations Analyze what’s top-of-mind for MORN investors based on questions over time.
130. ©2024 Morningstar. All Rights Reserved. 12 On the Horizon for Mo and the Intelligence Engine Intelligence Engine Management Console Intelligence Engine users can leverage a self-service console to create, test, and deliver applications Portfolio Analysis Applications can analyze and summarize a portfolio’s strategy and its holdings Conversation Memory Support for follow-up questions that require in-session context memory Context Awareness Applications take a question’s context—such as the page the user is on— into account to generate a response Charting Responses Applications can respond with charts (such as return comparisons) when relevant Expanded Screening and Data Access Ongoing expansion to the engine’s data tool and supported screening filters
131. ＜重要事項＞ <Disclosures> © 2024 Morningstar, Inc. All Rights Reserved. 本レポートの全ての著作権はモーニングスターに帰属します。モーニングスターから事前の書面による承諾がな い 限り、当資料の一部あるいは全ての複製ならびに再配布等の使用はできません。また、損害賠償、著作権法の罰則の対象となります。 本レポートに記載されてい る情報、データ、分析、意見は、（1）モーニングスターの機密情報ならびに専有情報が含まれます、（2）コピーならびに再配布はできませ ん、（3）作成日時点の情報 です、（4）予告なしに変更することがあります、（5）モーニングスターによる投資助言ではありません、（6）情報提供のみを目的としたも のです、（7）銘柄の売買を推 奨するものではありません、（8）情報の正確性、完全性や適時性を保証するものではありません。モーニングスターは、法律により定 められている場合を除き、本レ ポートの情報、データ、分析、意見を利用して行ったいかなる投資の判断、損失、損害に責任は負いません。 当資料にある指数はそれ自体運用商品ではなく、直接 投資することはできません。過去のパフォーマンスは将来のリターンを保証するものではありません。本レ ポートは営業用資料ではありません。投資を行う際は、目 論見書や他の開示情報をご確認ください。 Morningstarの商号、ロゴはMorningstar, Inc.の登録商標です。 ©2024 Morningstar, Inc. All rights reserved. This presentation includes proprietary materials of Morningstar; reproduction, transcription or other use, by any means, in whole or in part, without prior written consent of Morningstar is prohibited. The information, data, analyses, reports and opinions presented herein include proprietary materials of Morningstar; are not allowed to be duplicated or redistributed; do not constitute investment advice; are provided as of the date written; solely for informational purposes; and subject to change at any time without notice. This content is not an offer to buy or sell any particular security and is not warranted to be correct, complete, accurate or timely. Except as otherwise required by law, Morningstar shall not be responsible for any trading decisions, damages or other losses resulting from, or related to, this information, data, analyses or opinions or their use. The indexes shown are unmanaged and not available for direct investment. Past performance is not a guarantee of future results. This report is not a sales material. Please refer to the Prospectus or other disclosure information in making investment decisions. The Morningstar name and logo are registered marks of Morningstar, Inc.
132. Balancing Innovation and Oversight – Navigating the Future of AI 1 1
133. The AI hype cycle: AI has many forms, each going through its technology maturity cycle. Narrow AI Vs. General AI • Narrow AI is good at automating manual, repetitive and structured tasks • General AI is creative and can expand beyond traditional limitations
134. Ecosystem Approach to AI People What are the outcomes we are looking for - Economic Growth? Inclusion? Competitiveness ? Hub Do we have adequate AI computing capacity? (GPUs, etc.) Geopolitics Do you have adequate AI specialists? Economy Do we need a dedicated Hub infrastructure to foster innovation? Compute Do we have secure access to cutting-edge AI technologies and chips ? 1 2 3 4 5
135. Financial Sector Perspective Hypothesis 1: AI will push the frontier of Financial Services and will disrupt foundational financial system architecture Hypothesis 1: AI will come with new technology risks (Fairness, Ethics, Accountability and Transparency) Research Compute & Tech Stack Data & Models Talent Innovation Ecosystem Capabilities Governance Security Strategy & Geopolitics Incentives & Oversight Conversational Intelligent Foundational Resources FI Role Policy Role Hypothesis 2: AI will accelerate productivity within Financial services Hypothesis 3: AI will reduce fraud, financial and compliance risks Hypothesis 4: AI will radically improve regulatory risk monitoring Opportunities Risks AI Hypothesis 2: AI will have technology concentration risks Hypothesis 3: AI will have technology IP access risks due to geopolitics Hypothesis 4: AI will have IP access risks Budget
136. Applicability of Generative AI across a bank Function Advise Create Assist Automate Protect Marketing Targeted product recommendations Personalized marketing copy & creatives creation Content creation for social media campaigns Brand identity by analyzing online content, reviews, etc. Sales Sales Tactic recommendations for Agents Generate content for follow up emails, conversations Real time response suggestions based on customer’s sentiment Lead Qualification Customer Onboarding Chatbots for onboarding support Human friendly loan decline reasons Credit Underwriters with summarized application info Information extraction from documents KYC and Compliance Checks Customer Engagement Personalized investment & savings advice Customer personas based on behavior, needs, preferences from transaction data Financial Assistance – track expenses, bills, budget Customer feedback analysis Customer Service Query resolution recommendations for agents Personalized financial literacy content Assist agents in extracting relevant information from company knowledge base Credit & Market Risk Portfolio management in line with macro economic factors Generate scenarios for stress testing for ECL models Model validation via code replication Regulatory & Compliance reporting Collections Advise agents on strategy for incentive maximization Personalized collections emails, letters, notice etc. Assist agents with summarized case histories & suggested scripts Follow-ups as per last interaction Fraud & AML Assist Fraud Managers in investigations Synthetic data generation for rare & unseen fraud events Investigation of emerging money laundering scenarios Automate L0 investigations, auto-fill forms Talent & HR Personalized skill development trainings Generate first draft of policies, employee engagement programs etc. Assist employees in writing email drafts, task lists, action plans etc. Resume screening & matching Decisioning models Transparent AI systems Synthetic data generation Model documentation Model monitoring Legal Draft contracts Legal document review & analysis Due diligence & compliance monitoring IT Predictive maintenance Code generation Server logs analysis Cyberthreat detection & response Degree of Gen AI Usage Low High Medium AI Financial Sector Use Cases with Generative AI
137. Technology Vision 2023 Global Business and IT Executive Survey. ASEAN N= 320 In the next 3 to 5 years, in which of the following business activities does your organization plan to experiment with AI foundation models? 68% 60% 55% 48% 25% 24% 21% 20% 15% Customer support Process automation Information/data analysis Product design Software engineering Content creation Tackling computer-science problems Writing assistance/automation Scientific research AI foundation models - Large AI models pretrained across data modalities (like text and images), rather than on a specific task, that can learn to complete new tasks within these data types with little to no extra training. Examples include large language models, such as GPT-3, and text-to-image generators, such as DALL-E 2 and Stable Diffusion. Organizations plan to experiment with AI foundation models in the next 3 to 5 years in various ASEAN business activities with most executives citing customer support (68%) and process automation (60%). Industry experimenting Generative AI across many domains
138. Generative AI potential and Risks Intellectual Property How will the business protect its own IP? And how will it prevent the inadvertent breach of third- party copyright in using pre-trained foundation models? Data Privacy & Security How will upcoming laws like the EU AI Act be incorporated in the way data is handled, processed, protected, secured and used? Intellectual Property Data Privacy & Security Discrimination Is the company using or creating tools that need to factor in anti-discrimination or anti-bias considerations? Product Liability What health and safety mechanisms need to be put in place before a generative AI-based product is taken to market? Trust What level of transparency should be provided to consumers and employees? How can the business ensure the accuracy of generative AI outputs and maintain user confidence? Identity When establishing proof-of-personhood depends on voice or facial recognition, how will verification methods be enhanced and improved? What will be the consequences of its misuse?
139. 8 Navigating the future of AI: MAS AI Response and Regulations
140. To establish Singapore as a leading AI in Finance Hub by anchoring impactful use cases and promoting deep and pervasive adoption of AI within the financial sector
141. Three Pillars Support Industry AI Adoption Governance (foundation & enabler) Data & Infrastructure (foundation & enabler) AI Talent (raising baseline AI adoption) AI governance is the foundation to enable AI innovation and adoption. The value of AI can be unlocked only when AI innovation is pursued within the defined guardrails to protect customers, FIs and stakeholders. Data availability, quality & privacy are the key challenges for the AI adoption in the financial sector. Without solving the data challenge, the potential value of AI will not be realized. Hence, we will develop secure infrastructure with proper data sharing protocol to enable AI solution development with industry relevant data. Guided by the GenAI JTM, to develop local AI talents and to attract global top AI talents to Singapore to support AI adoption in the financial sector. AI strategy (Financial Sector AI Adoption) AI Talent Development Programme Industry wide AI solutions:
142. 11 11 AI Governance
143. Global landscape on AI standards & regulations UK • Executive Order on Promoting the Use of Trustworthy AI in the Federal Government – 2020 • The National AI Initiative Act – 2020 • FTC | Aiming for Truth, Fairness and Equity in Company’s Use of AI – 2021 • New York | Bill on AI Employment Tools – 2021 • Draft NIST AI Risk Management Framework – 2022 • Draft US Algorithmic Accountability Act – 2022 • Draft Digital Platform Commission Act – 2022 • Draft Health Equity and Accountability Act – 2022 • Draft American Data Privacy and Protection Act – 2022 • Data Ethics Framework – 2018 • AI in the Public Sector – 2019 • Understanding AI Ethics and Safety – 2019 • National AI Strategy & Standards Hub – 2021 • Algorithmic Transparency Standard – 2021 • Defence AI Strategy – 2022 UNITED STATES AUSTRALIA • AI Ethics Principles – 2019 • AI Action Plan – 2021 • Issues Paper: Automated Decision Making and AI Regulation – 2022 SINGAPORE • MAS FEAT Principles – 2018 • IMDA Model AI Governance Framework v1 – 2019 • IMDA Model AI Governance Framework v2 – 2020 CANADA • Directive on use of Automated Decision- Making by Federal Government – 2021 • Draft Digital Charter Implementation Act – 2022 • Draft Automated Decision Systems (Quebec) Bill 64 – 2022 EU • Social Principles of Human-centric AI – 2019 • AI Governance Guidelines – 2022 JAPAN CHINA • Governance Principles for Responsible AI – 2019 • Recommendation Algorithm Regulation – 2022 Sample of Responsible AI related policies July 2022 FURTHER SOURCES Global: OECD. AI Policy Observatory Europe: JRC and OECD 2021 AI Watch OECD • Principles on AI – 2019 NORWAY • DPA Sandbox on AI – 2020 FRANCE • CNIL (DPA) Sandbox – 2021 INDIA • NITI Ayog Exploring AI Principles – 2021 • General Data Protection Regulation (GDPR) – 2016 • HLEG Ethics Guidelines for Trustworthy AI – 2019 • Draft EU Regulation on AI – 2021 • Consultation on Product Liability Framework – 2021 KENYA • Data Protection Act – 2019 CHILE • National Policy on AI – 2021 BRAZIL • AI Bill proposal – 2021 ~ REGULATIONS ARE NON-EXHAUSTIVE ~ Copyright © 2023 Monetary Authority of Singapore. All rights reserved.
144. Importance of Responsible AI Responsible AI Responsible AI is the practice of designing, building, and deploying AI in a manner that empowers people and businesses, and fairly impacts customers and society. Responsible AI enables companies to scale their use of AI with confidence. It’s the foundation to enable further AI innovation and adoption. Fortify existing operational infrastructure Improve customer experience Increase investor confidence Engender customer trust Boost employee commitment by ensuring responsible systems Responsible AI’s impact on the organisations
145. MAS Responsible AI Journey Fairness Ethics Accountability Transparency MAS released FEAT principle in 2018 to provide guidance to FIs for Responsible AI adoption Not a Regulation FEAT is a regulatory guidance, rather than a regulation Project Veritas Published Veritas Methodology and use case study (9 whitepapers), and open-source Veritas Toolkit to evaluate AI systems’ alignment with FEAT principles. 2018 2019-2022 2023 Project MindForge Published a GenAI Risk Framework (1 whitepaper) FEAT+ and new risks include: § Inaccuracies § Economic & Social § Governance § Cyber Security § Stability § Privacy § Legal
146. Risk Dimension Description Fairness & Bias Setting fairness objectives to help identify and address unintentional bias and discrimination. Ethics & Impact Ensuring responsible and ethical outcomes in AI use against a clearly defined set of core values and practices. Accountability & Governance Enabling accountability and governance for the outcomes and impacts of data and AI systems. Transparency & Explainability Enabling human awareness, explainability, interpretability, and auditability of data and AI systems. Legal & Regulatory* Identifying any legal or regulatory obligations that need to be met or may be breached by the use of AI, including issues with compliance, data protection and privacy rules. Monitoring & Stability Ensuring the robustness and operational stability of the model or service and its infrastructure. Cyber & Data Security Protecting data, AI models and systems, and other enterprise information technology (IT) assets from unauthorized access, data loss or leakage, and misuse by malicious actors. *This would typically cover regulation, legal liability, indemnification, related legal obligations, risks and requirements. MindForge GenAI Risk Framework Risk Risk Dimension Potential impact or harm of GenAI to an FI or its stakeholders Group of related risks that share common characteristics and respond to common concerns
147. Examples of New Risks from GenAI Risk Dimension Examples of New Risks Fairness & Bias • New risk could come from the GenAI training data which is unrepresentative or biased, especially from the internet. On the other hand, FIs normally know the training data when training traditional AI models. (FEAT Fairness principle: data and models used for AIDA-driven decisions are regularly reviewed and validated for accuracy and relevance, and to minimise unintentional bias.) Ethics & Impact • GenAI systems could create new content which may be misaligned with societal values, and the rationale or root cause may be difficult to pinpoint due to the difficulties to explain the GenAI model. • Generation of synthetically created deceptive or manipulative content that may trick or mislead users into taking certain actions without fully understanding the consequences (example, nudging children towards certain content or services). Accountability & Governance • GenAI systems could generate new content and it is difficult to understand how the new content are generated. Hence, it is difficult for third party GenAI system providers to take the accountability. • Big volume of new content generated by GenAI systems may lead to insufficient human intervention when human validation is required. Transparency & Explainability • The data used to train GenAI systems may not be convincingly traced. This brings challenge for audit and disclosure. Current FEAT Transparency principle requires FIs to provide clear explanations on what data is used to make AIDA-driven decisions. Legal & Regulatory • New risk could come when GenAI systems’ training data is from internet and violates IP rights. • GenAI systems’ training data may include personal or sensitive data. This may violate some regulation. Monitoring & Stability • GenAI systems may generate outputs that contradict the source content due to lack of understanding of the real-world (hallucination/fabrication/confabulation). This requires to monitor GenAI systems to ensure their robustness and operational stability. Cyber & Data Security • The intentional manipulation of GenAI systems by a malicious actor, through the introduction of malicious data either during initial training or during use, could result in security vulnerabilities or produce inaccurate and harmful outputs. Limited or No FEAT/Veritas Coverage Strong FEAT/Veritas Coverage
148. How to turn private equity firms into AI-based learning organizations The example of Jolt.Ninja Japan Investment Conference 2024 CFA Society of Japan May 2024
149. Disclaimer This presentation was prepared by members of Jolt Capital and it remains the property of Jolt Capital. The Presentation must be returned to Jolt Capital promptly upon request by Jolt Capital. This Presentation was provided to you on a confidential basis for information purposes only and is not and may not be relied upon in any manner as legal, tax, accounting or other advice or as an offer or a solicitation of an offer to buy an interest in any securities referred to herein. This Presentation and the information contained herein must be strictly treated in a confidential manner and may not be reproduced, used, redistributed or disclosed in any format, in whole or in part. Any reproduction of distribution of this Presentation, in any format, whether in whole or in part, or the disclosure of its contents is prohibited without the prior written consent of Jolt Capital. The information set forth herein does not purport to be complete and is subject to change. No representation is therefore given with respect to the accuracy or completeness of this Presentation and it may change without notice. Jolt Capital on its own behalf and on behalf of its representatives as well as its relevant services providers likely to circulate the Presentation disclaim any and all liability relating to this Presentation, including without limitation, any express or implied representations or warranties for statements or errors contained in, or omissions from, this Presentation. Recipients should make their own investigations and evaluations of the information contained in this document. Each recipient should consult its own attorney, business advisor and tax advisor, acting under a duty of confidentiality as such persons shall also be held to the same confidentiality obligations as set forth therein, as to the legal, business, tax and related matters contained in this Presentation. This Presentation is not intended to be, and shall not be, a basis on which an investment decision is taken. Certain information from this Presentation has been obtained from published and non-published sources prepared by other parties, which in certain cases have not been updated through the date hereof. In addition, certain information contained herein has been obtained from companies in which investments have been made. While such information is believed to be reliable for the purpose used in this document, Jolt Capital and its representatives do not assume any responsibility for the accuracy or completeness of such information, and such information has not been independently verified. Except where otherwise indicated herein, the information provided in this Presentation is based upon matters as they exist as of the date of preparation and not as of any future date. There is no guarantee that an investment objective is achieved nor that there will be any return of capital. Moreover, past performance of the investment team should not be construed as an indicator of future performance. Statements contained in the Presentation that are not historical facts are based on current expectations, estimates, projections, opinions and beliefs of Jolt Capital. Such statements involve known and unknown risks, uncertainties and other factors, and undue reliance should not be placed thereon. Additionally, the Presentation contains "forward-looking statements” which can be identified by the use of forward-looking terminology such as “may”, “will”, “should”, “expect”, "anticipate”, “target”, “project”, “estimate”, “intend”, “continue” or “believe” or the negatives thereof or other variations thereon or comparable terminology. This Presentation has not been approved by any regulatory body. None of the information contained herein has been filed with the U.S. Securities and Exchange Commission, any securities administrator under any securities laws of any U.S. or non-U.S. jurisdiction or any other U.S. or non-U.S. governmental or self-regulated authority. No such governmental or self- regulatory authority will assign on the merits or the adequacy of the information contained herein. This Presentation does not constitute an offer of shares nor an invitation to subscribe for or to purchase any fund shares by any person in any jurisdiction in which such offer or invitation is not authorized or in which the person endeavoring to make such offer or invitation is not qualified to do so, or by any person to whom it is unlawful to make such an offer or invitation. 2
150. Jolt Capital: a leading growth deeptech investor Scaling deeptech champions since 2011 § The largest and most experienced growth deeptech(*) investor based in Europe § Superior returns and track record, demonstrated ability to scale tech champions § Responsible and impact investing § Data-driven investor leveraging proprietary AI Jolt.Ninja Founded in 2011 Team 35 Portcos located in 8 European countries Presence in 5 European countries 1,000+ Jobs created/year 3.5x Average cash on cash exit multiple 3.8M Companies sourced in Jolt.Ninja 1€=1kg CO2 Avoided €1.7B Largest exit 50%+ Pertaining to EU taxonomy >500M€ AUM >20% IRR Since inception (*) A deeptech company is a business based on an original IP 3 Jolt Capital’s current locations (as of May 2024) Tokyo Munich Athens Milan Copenhagen Lausanne London Paris Jolt Capital’s planned new locations (by 2025) Montreal Boston Palo Alto Portfolio companies North America Europe Asia
151. What is AI ? 4
152. From AI to Deep Learning 5
153. Machine Learning vs Deep Learning Both Machine and Deep Learning rely on mathematical operations to approximate a complex function. In Machine Learning, a human chooses the most important features to use as a classification basis. In Deep Learning, the system works directly on the raw input. As a consequence, Deep Learning needs enormous quantities of data to converge but is fully automated. The most advanced language model from Microsoft/Nvidia has 530 billion parameters that must be tuned during the learning phase. 6
154. AI in PE 7
155. Progressive evolution towards AI 8
156. Most PE firms don’t really invest in (their own) tech 9
157. Main applications Market research • Map value chains automatically • Analyze competitive environments • Analyze funds activity, exits, etc. Deal Sourcing • Process large quantities of data to identify trends, create clearer representations enabling easy comparisons • Identify actual companies and find data qualifying them from multiple unstructured/structured sources • Score companies according to fund’s strategy Deal Flow Management • Identify targets for CRM • Learn from daily partner decisions Due Diligence • Automatically analyze reports covering competition, markets, news… • Analyze vast amounts of data in order to model targets financial history Accelerate Value Creation for Portfolio Companies • Identify build-up candidates 10
158. Challenges No definitive playbook in Private Equity • Each fund has its own criteria • No magic bullet => No rule to implement Lack of Data • Difficult to gather large amounts of PE data to train algorithms, as the number of transactions is limited and information is not widely divulged. • It takes years to know the outcome of a PE investment decision, and performance is usually impacted by convoluting factors, such as interest rate movement • Unstructured data plays a much more critical role in PE investing, as there is no “efficient market” governing valuations. Cultural • Partners are not used to collaborative work • Need for a full stack solution that addresses the whole process from origination to exit Not just AI • Lack of structured data in Private Equity (compared to public equity) requires particularly sophisticated data collection techniques • Solution will be used only if great UX and full stack service • Fintech level security • Costs and latency 11
159. The main solutions Lots of technologies Full collaboration of all partner at inception • Teach the system what is a good or a bad company by example (thousands) • Explain clearly what would save time and costs accross all company processes (complex internet research can be fully automated, summary of business plans, multiple sources of data collected analyzed and presented in a powerpoint, etc.) • Make sure that all interactions with the system are used as learning inputs No place to hide: all processes must leverage AI • The system must integrate all key processes, it must be the central point of the company • Learning from any interaction/input on any process • The system needs to give back 100x in intelligence, time saving, fun, make look incredibly smart, etc. 12
160. Jolt.Ninja 13
161. History Started in 2016, 3FTE, 4 trainees at EPFL • R&D team embedded in the investment team • CTO part of the investment committee • Over 2m companies profiled automatically since 2017 Productivity • Increased efficiency in deal origination, monitoring and due diligence • Valuable insights on companies (patents, descriptions, relevant similar firms, news feed), enabling Senior Team to dedicate significantly more time to value-added activities AI-based platform • Deep Learning & Natural Language Processing algorithms are underlying Ninja’s performance • Jolt’s team efforts (~25.000 companies reviewed) boost Jolt.Ninja's ability to automatically identify and recommend daily increasingly relevant companies 14
162. One of the largest technology companies databases >3k deal opportunities suggested per year 3.8m companies profiled and updated in real time 20k+ new companies discovered/month 54k investors and their funds profiled 4m+ patents analyzed and scored 5m+ key executives, including management transitions Unique AI/data driven Jolt scores for companies, technologies, strategic positioning, etc. Jolt.Ninja business intelligence publications Greek Deeptech Startups & Scaleups Shipping Industry A European comparative mapping by Jolt.Ninja 2023 Danish Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 European Deeptech Startups & Scaleups A comparative mapping by Jolt.Ninja 2023 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 ∑ 1,378 1,445 1,531 1,613 1,704 1,808 1,931 2,033 2,125 2,210 2,279 2,323 2,333 Aero 689 722 760 798 842 903 955 1,007 1,048 1,090 1,124 1,155 1,163 Land 194 205 216 224 232 253 259 273 281 291 295 298 299 Water 148 155 162 168 177 181 191 197 201 206 207 207 208 Other 780 820 870 917 972 1,032 1,112 1,171 1,227 1,279 1,325 1,345 1,348 0 600 1,200 European Defense Industry companies/domain exposure Σ domains ≠ Total 4% CAGR 4% 3% 5% 4% Strictly Confidential 12 European Number of Companies in Defense Industry Companies founded in 2021/2022 may not have been identified yet o As seen, there are about 2,300+ European defense companies, while participating NATO countries account for 80% of them. o Approximately, +60 European defense companies are founded yearly in NATO participating countries. o European defense companies are concentrated in the aerospatial and other warfare (cyber, C4ISR, new warfare, etc.) domain. - 800 1,600 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Participating NATO countries Non-participating NATO countries Europoean Defense Industry companies/region CAGR 462 1,871 5% 4% Strictly Confidential 10 Deeptech Investment Universe in Canada/Quebec WORLD EUROPE Cleantech & Energy AI & Cloud Computing Auto & Aerospatial Strictly Confidential 8 European Number of Companies in the Biocontrol Perimeter WORLD EUROPE Biocontrol Companies Company HQ in Europe SEIPASA (ES) Operator of an agriculture biotechnology company intended to sell solutions of botanical and microbiological origin for crop protection, bio stimulation and nutrition. PHYTOWELT (DE) Developer of industrial and plant biotechnology intended to produce Phyto enzymes in productive bacteria strains and a variety of biosynthetic possibilities that cannot be performed by chemical synthesis. CERTIS EUROPE (NL) Certis Europe is a development and distribution company marketing a range of crop protection products for the arable, horticultural and high value speciality crop markets. Three companies automatically selected by Jolt.Ninja’s AI/NLP-based algorithm : Total number of Biocontrol Companies by countries Aggregated number at country level : 1. United Kingdom (204) 2. France (137) 3. Netherlands (101) 4. Italy (87) … 15
163. Jolt.Ninja: 100 years of experience and all processes embedded in our AI platform Market research Deal sourcing Pre Due Diligence Investment Decision/ Committee Due Diligence Value Creation Reinvest Buildup Exit Data collection : fully automated § Brute force data collection: thousands of scrapper are collecting data from the open internet § Reinforce the data with databases (paying or free) § Make sure AI links everything : people, patents, companies, investors, funds, LPs, etc. Process management: An AI-enhanced productivity engine § AI-generated deal flow § GenAI: report summaries, value chain creation from examples § IP quality § M&A suggestion Firm scalability: expanding Jolt with total continuity § Covers all continents § No dependencies or inefficiencies : Ninja learns from all interactions § As the main workflow tool, Ninja enhances open communication and process transparency 16
164. Architecture Key Technologies § Fullstack Serverless App § Tensorflow § Open.AI § Transformers § AWS SageMaker § AWS Lambda § Neo4J Graph Database Knowledge Graphs § 40 millions nodes, 110 millions vertices § Graphs are used to model real world complexity § Stored in a Neo4J Data Deep Learning § Does this company fit Jolt’s investment thesis? § Service or Product? § PyTorch, Sagemaker, HuggingFace Semantic and NLU § Extract structured information from texts and documents § Automatic domain identification (taxonomies) § GoogleNLP, OpenAI, HuggingFace 17
165. Academic interest for Jolt.Ninja • Performance analysis • Bias Detection • Data Quality • Augmented Investor An inside peek at AI use in private equity Journal of Financial Data Science, 2021 (Thomas Åstebro, HEC) Humans still required in private equity India Business Law Journal, 2022 (Vandana Pai, Bharucha & Partners) 18
166. Jolt.Ninja examples 19
167. Opening Ninja as a content provider: collection of atlases (available online) Danish Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 Irish Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 Danish Deeptech Startups & Scaleups Shipping Industry A European comparative mapping by Jolt.Ninja 2023 German Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 Irish Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 Polish Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 Quebec Deeptech Startups & Scaleups A Canadian comparative mapping by Jolt.Ninja 2023 French Deeptech Startups & Scaleups A European comparative mapping by Jolt.Ninja 2023 20
168. Insights: pitch summary 21
169. 1’ due diligence report 22
170. Value Chain: with only 3 examples } 23
171. Q&A 24
172. 25 Copenhagen Højbro Plads 8-10 København K 1200 - Denmark Paris 1 impasse Fortin (visitors entrance) 56-58 rue de Ponthieu (postal address) 75008 Paris - France Lausanne EPFL Innovation Park Building I – Office 113 / Route Cantonale 1015 Lausanne - Switzerland Boston Business Dev. USA www.jolt-capital.com contact@jolt-capital.com Jolt-capital @JoltCapital Palo Alto Business Dev. USA
173. © 2024 CFA Institute. All Rights Reserved. 1 Data Science for Investment Professionals Certificate Gain practical knowledge of data techniques and machine learning fundamentals, and how they are used in the investment process. Learn the skills of the future, today. Brenda Hou, CFA
174. Data Science for Investment Professionals Certificate • Designed to provide investment professionals with practical knowledge of the fundamentals of machine learning techniques and how they are used in the investment process. • For those in core investment roles, like analysts, portfolio managers and relationship managers and those who aspire to be in those roles. • Learners will be able to explain clearly and “translate” machine learning concepts and their application to real-world investment problems to a non-expert audience and clients. • Your team will understand the language of Data Science to better serve and anticipate the needs of your clients. Visit this page to watch the overview video. © 2024 CFA Institute. All Rights Reserved.
175. Certificate Courses Earn the certificate on completion of all courses and the coding labs, and the minimum passing score on the final assessment. Data and Statistics Foundation Course 2 Statistics for Machine Learning Course 3 Machine Learning Course 4 Natural Language Processing Course 5 Mitigating Biases in the Data Science Pipeline Final Assessment 90-minute multiple-choice assessment (online) Course 1 • Measures of Central Tendency • Measures of Dispersion Introduction to Distributions • Data Visualization Techniques • Sampling Theory • Hypothesis Testing • Data and Patterns • Randomness and Probability • Linear Regression • Introduction to Advanced Regression Concepts • Introduction to Time Series Analysis • Machine Learning • Supervised Learning • Unsupervised Learning • Deep Learning • The Translator • Cleaning and Wrangling Text Data • Exploratory Data Analysis, Feature Selection and Feature Engineering Selecting, Training, Evaluating, and Tuning an NLP Model • Developing an NLP Model Applications of NLP in Investments • Investment Context, Some Ethical Dilemmas, Biases and Practical Issues • Case Studies and Code Labs © 2024 CFA Institute. All Rights Reserved.
176. Describe and evaluate the statistical methods that underpin machine learning techniques Select appropriate data visualizations and create simple visualizations using Python Describe how machine learning applications can address real-world investment problems Explain machine learning concepts and techniques to a non-expert audience Interpret natural language processing–based classification models and understand their use in investment decisions Evaluate machine learning models for biases and understand strategies to mitigate them Learning Outcomes © 2024 CFA Institute. All Rights Reserved.
177. Certificate Overview 90 – 100 hours 12 – 15 hours per course Members USD 1,272 Non-members USD 1,590 Portfolio Managers, Analysts applying data science techniques, business leaders No experience required, ideal for those with some background in finance and investments Self-Paced, Online Final assessment © 2024 CFA Institute. All Rights Reserved.
178. Participant Profile Ideal for current or aspiring investment professionals including but not limited to analysts, portfolio managers, relationship managers, and traders. Most Common Job Titles That Benefit from this Certificate: Portfolio Manager Equity Analyst Quant Analyst Investment Strategist Financial Advisor Financial Reporting Analyst © 2024 CFA Institute. All Rights Reserved.
179. Scan Me Format: Self-paced online Length to Complete: 90 - 100 hours Mode of Exam: Online 60 multiple-choice questions. 70% pass score. 2 attempts at final assessment. 90 mins, 60 multiple choice questions Data Science for Investment Professionals Certificate Recognition Format: Certificate and digital badge Price: Members USD 1,272. Non-members USD 1,590 Participant Profile For analysts, portfolio managers, and relationship managers seeking practical knowledge of data science and machine learning fundamentals and their applications in investment. Good to have some background in finance and investments. No prior knowledge of Python or R required. Certificate Overview Find out how to apply machine learning and data science concepts in real investment problems and articulate them to non-experts and clients. Learn through immersive code labs with real-world scenarios. The certificate comprises five courses with practical exercises and a final assessment. Enroll now: https://store.cfainstitute.org/data-science-for-investment-professionals-certificate/ © 2024 CFA Institute. All Rights Reserved.