The Nielsen Company (US), LLC. Confidential and proprietary. Do not distribute. AUGMENTED ANALYTICS Practical approach to Business Intelligence automation Yana Salimova Sep 17, 2020
and proprietary. Do not distribute. AGENDA WHAT MACHINES CAN AND CANNOT DO NOW? HOW TO CHOOSE THE BEST CASE FOR AUTOMATION? HOW TO DELIVER ANALYTICS 39X FASTER LESSONS LEARNED 1 2 3 4
and proprietary. Do not distribute. WHAT MACHINES CAN AND CANNOT DO NOW? If process is well defined and can be split into sequence of simple “1 second” tasks it’s feasible to automate it with a combination of tools Rule 1: You need to have a lot of data to teach an algorithm Rule 2: Anything you can do with 1 second of thought can be automated FEASIBLE: • Image recognition • Translation • Visual inspection for defects • Requests triage NOT FEASIBLE: • hire an executive • write a novel • prepare 50 page report on market research
and proprietary. Do not distribute. HOW TO CHOOSE THE BEST CASE FOR AUTOMATION Technical due diligence Data available Repetitive Well-defined process Existing Best Practices Business impact Large and/or expensive group of people Spend considerable amount of time on the task Perform task often
and proprietary. Do not distribute. FROM BI TO AI ANALYTICS Our key objective is to automate causation and prediction Descriptive What happened? BI platform, from manual visualization to smart one. Which learns from user interaction Causal Why happened? AI-driven insights. Generic algorithms to drill down through the data and find key drivers Predictive What will happen? Variety of generic forecasting algorithms. Linked heavily to anomaly detection Prescriptive What shall i do next? Custom build to specific use cases like Marketing Mix or Price optimizer. Autonomous AI executes as per prescriptions RTB (real-time bidding) eCOm pricing Manufacturing facilities automation
and proprietary. Do not distribute. THE VISION The Augmented analytics COE was formed to modernize our Analytical teams by leveraging modern technologies to unlock the power of our data From Static presentations with low automation and topline insights that take 2-3 weeks to produce (& labor touch of ~35 hrs) To Augmented Analytics powered by Artificial Intelligence and returned in seconds. Natural Language Interface and Deep Insights generated through Data science and Machine Learning for tools that improve every day Thousands Associates globally Data Science Integrating data science Into our everyday analyses Artificial Intelligence Integrate Data sources Natural Language Processing (NLP) Multiple analytical paths and model boosting to analyze Big data Natural Language Generation (NLG)
and proprietary. Do not distribute. STEP1: CHOOSING THE PILOT USE CASE Use data Time tracking & Lean diagnostics measurements to identify most frequent use case with cycle time & touch time Deep dive into process Best Demonstrated Practices and best analysts helped to describe analytical process in details
and proprietary. Do not distribute. STEP 2: FIND PARTNER AND DEVELOP THE POC Build or buy Vendor selection Agile development process, joint work with our experts team Detailed SOW: what else can we do with AI what’s not possible now?
and proprietary. Do not distribute. STEP 3: TEST THE SOLUTION AND MEASURE THE RESULTS Alpha testing: few users, ensure data is correct and tool works correctly Beta testing: more users, convenience and results assessment • Time per analysis reduced from 35h to 0.9h → 39x improvement! • Quality of insights remained the same Next Steps: deploy as part of core analytical platform
and proprietary. Do not distribute. DIFFICULTIES AND LESSONS LEARNED 1. Buy-in and support from ALL stakeholders 2. Independent vs embedded R&D team: depends on independence level 3. Data access and data preparation: buy-in from data owners is key 4. Results assessment: set game rules from the start 5. Business readiness to follow standard process is key 6. Feedback from users: start early and frequently
aiconforg
buyselltechnologies
ktkn1129
kakehashi
re9010matsubara
kassy1127
enechain
kimpachi_d
viva_tweet_x
3sunny
ashisuto_career
wakayamaguchi
chrisshort
shpigford
paulrobertlloyd
lara
tammyeverts
chriscoyier
myddelton
sachag
notwaldorf
davidbonilla
thekraken
aarron
