Marina Hess & Yannick Arndt | Peak Ace AG 16 May 2024 NAVIGATING INTERNATIONAL MARKETS: AI'S ROLE IN PAID SOCIAL CONTENT LOCALISATION

2 peakace.agency Let‘s start with a quick introduction Marina Hess Senior Paid Social Consultant Yannick Arndt Senior Paid Social Consultant

3 peakace.agency What we've brought along for you today Quick introduction to Peak Ace Theory: AI localisation and its potential Practice: how we mastered AI localisation for a large international client

4 peakace.agency 4 peakace.agency DIGITAL STRATEGY AI SOLUTIONS PAID SEARCH & SOCIAL DISPLAY & VIDEO CONTENT & DIGITAL PR ORGANIC SEARCH (SEO) DIGITAL ANALYTICS MARKETING AUTOMATION & CRO DELIVERING DIGITAL EXPERIENCES 130+ Awards 200+ Clients 30 Industries 170+ Peak Acers 25+ Languages

5 peakace.agency International Social at Peak Ace Together with our native speakers of 25+ languages, we develop target market-oriented campaigns in order to sustainably expand our global reach. German US-English UK-English Dutch Finnish French Danish Italian Japanese Norwegian Hungarian Polish Portuguese Russian Spanish Swedish Turkish Czech Romanian Arabic Croatian Chinese Thai with more to come

6 peakace.agency International Social at Peak Ace Together with our native speakers of 25+ languages, we develop target market-oriented campaigns in order to sustainably expand our global reach. German US-English UK-English Dutch Finnish French Danish Italian Japanese Norwegian Hungarian Polish Portuguese Russian Spanish Swedish Turkish Czech Romanian Arabic Croatian Chinese Thai with more to come

AI localisation: A little bit of theory

8 peakace.agency Persona/Product research Ad copy creation & optimisation Segmentation Targeting & audience research Competitor research Strategy & media planning Data analysis & interpretation Optimisation of landing pages & feeds Localisation/Translation Market expansion opportunities for social with AI

Localisation customises social media ads for specific regions by adapting content, imagery, and targeting parameters to resonate with local preferences and cultural nuances. What does localisation even mean?

10 peakace.agency Localisation vs Translation "What does your apartment look like?" "What does your apartment look like?" "Where does your magic happen?" Original caption ENGLISH The GERMAN caption has two meanings "How are things going?" Translation & Localisation

11 peakace.agency VS. MANUAL LOCALISATION AI-DRIVEN LOCALISATION 11 peakace.agency

12 peakace.agency Manual localisation can be unique and entertaining … but it presents a few challenges Pros Cons High manual effort High susceptibility to errors Personnel absences of all kinds High financial expenditure Agility & contextual understanding Eloquence and authenticity Naturalness Accuracy

13 peakace.agency AI-Driven localisation is prone to errors … but there‘s so much more Cost efficiency Availability Consistency Privacy & security Dependency on data Limited contextual understanding Quality assurance is necessary Pros Cons High speed and scalability

A human translator remains indispensable for the post-production and post-editing of advertising copy. However, leveraging AI tools can significantly streamline their efforts and accelerate the process

15 peakace.agency Qualitative analysis of different language models DeepL, Google Translate, ChatGPT etc. peakace.agency VS.

16 peakace.agency NMT vs GPT What‘s under the hood? NMT Neural Machine Translation GPT Generative Pre-trained Transformer Tailored for translations Focuses on accuracy Provides a consistent and predictable output based on language patterns The NMT engine can only be retrained to the extent permitted by the available training data An all-purpose-tool Focuses on a successful conversation Varies the output based on conversational patterns The output can be adjusted by adding specific instructions to the prompt

Every AI localisation tool struggles with… ▪ Differentiating between formal and informal "you" ▪ Using inclusive language correctly ▪ Handling memes, trends, and social media language ▪ Translating humour and jokes accurately ▪ Navigating cultural references effectively ▪ Balancing feminine and masculine forms 17 peakace.agency

18 peakace.agency … but some tools do it better than others Navigating through the masculine & feminine forms

19 peakace.agency ChatGPT: Keep talking

20 peakace.agency One more try: Nailing down the idioms Maybe a little more context is all it takes?

21 peakace.agency At least you tried…

For social media copywriting, a dialogue-based approach that prioritises conversational simplicity over translation accuracy is the preferable option. …just be sure to let native speakers handle the final touches! ChatGPT’s language model has proven to be very effective for our needs

Ad copy localisation: A client scenario

24 peakace.agency Recap localisation: this is how we worked previously Manual completion – copy & paste (…intentionally skipped the part on different Excel versions…) Native speakers handle localisations/ translations

25 peakace.agency Key problem Properly localised texts take a lot of time, and we would like to work more efficiently. Manual localisation ▪ Effort: ▪ 150-300 words localised per hour ▪ 6-12 people involved ▪ 6 target languages ▪ Costs are high due to high number of hours needed ▪ In truth, not the agency's focus but it is a strength How can we be more efficient with this? Can we use our time better for our core competencies? How can we reduce costs? What is the value we want to deliver?

AI Localisation: Google Sheets x ChatGPT

27 peakace.agency Google Sheets x ChatGPT Anonymizer case

28 peakace.agency Google Sheets x ChatGPT Numerous different plug-ins enable the use of ChatGPT in Google Sheets… in fact, too many!

29 peakace.agency Google Sheets x ChatGPT One prompt per cell, one cost per prompt

30 peakace.agency Google Sheets x ChatGPT Many varied functions, but too much syntax and too many errors ▪ =GPT(prompt, [value], [temperature], [model]) ▪ =GPT_LIST(prompt, [value], [temperature], [model]) ▪ =GPT_SPLIT(text, split_by, [temperature], [model]) ▪ =GPT_FILL(examples, [inputs], [temperature], [model]) ▪ =GPT_TABLE(prompt, [head], [inputs], [temperature], [model]) ▪ =GPT_MATCH(search_keys, lookup_range, [confidence], [stats], [top_k]) ▪ =GPT_FORMAT(input, target_format, [source_format], [temperature], [model]) ▪ =GPT_EDIT(text, [task], [temperature], [model]) ▪ =GPT_TAG(value, [tags], [examples], [top_k], [temperature], [model]) ▪ =GPT_CLASSIFY(value, categories, [examples], [temperature], [model]) ▪ =GPT_EXTRACT(text, to_extract, [temperature], [model]) ▪ =GPT_SUMMARIZE(text, [format], [temperature], [model]) ▪ =GPT_TRANSLATE(text, [target_language], [source_language], [instructions],[temperature], [model]) ▪ =GPT_CREATE_PROMPT(arg1, [arg2], [arg3], [...argn]) ▪ =GPT_VISION(prompt, image_url, [high_res], [temperature]) ▪ =GPT_WEB(prompt, [value])

31 peakace.agency peakace.agency Google Sheets x ChatGPT TLDR ▪ High setup effort ▪ Error-prone and usability suffers when working with several parties ▪ 1 request per cell → 1 cost per request

Challenge accepted Google Sheets x ChatGPT x AppScript

33 peakace.agency Wait, what? We have a tool designed to fill in large spreadsheets with ChatGPT support?

34 peakace.agency Use case: Localisation Customised to my client project, it offers quick wins and improved efficiency

35 peakace.agency The external connection Customised spreadsheets for each client offer data protection and accuracy

36 peakace.agency Connection is key Automated imports to guarantee budget control and usability

37 peakace.agency This is where the magic happens After the import, the localised content is automatically filled in

38 peakace.agency Landingsheet for a better overview Just for us to proofread and make sure everything worked out

39 peakace.agency Content delivery incoming …to provide the client with a full overview

40 peakace.agency Co-Working Space Proofreading and adjustments so that the import functions are not interrupted

41 peakace.agency Key problem Properly localised texts take a lot of time, and we would like to work more efficiently. Manual localisation ▪ Effort: ▪ 150-300 words per hour can be localised ▪ 6-12 people involved ▪ 6 target languages ▪ Costs are high due to high number of hours needed ▪ In truth, not the agencies focus, but a strength AI driven localisation: ▪ Effort: ▪ Small costs for using API connection ▪ 1000 words per hour for review and adjustments ▪ 6-12 people involved ▪ 6 target languages ▪ Lower costs for localisation ▪ More time for client love and performance checks 6 x faster

42 peakace.agency Visions More ideas to come or to implement More control for clients Improvements in usability Output precision

peakace.agency Ready for the next step in your career? jobs.peakace.agency Marina Hess m.hess@pa.ag Paid Social Consultant Yannick Arndt y.arndt@pa.ag Paid Social Consultant Are you still translating, or do you already have a head start in localisation? Build your own spreadsheet: Questions? Hit us up!