Appen Ltd. · ASX:APX
Appen — Investor Day (May 2023)
85 slides from Appen's investor day presentation, May 26, 2023
Slide 1: 2023 Investor Technology Day - Presentation
Appen Limited Level 6, 9 Help Street Chatswood NSW 2067 Tel: 02 9468 6300 www.appen.com ACN: 138 878 298 ASX ANNOUNCEMENT 26 May 2023 2023 Investor Technology Day - Presentation Appen Limited (ASX: APX) provides the attached presentation for the Investor Technology Day being held today from For pers
Slide 2: 2023 investor technology day
For personal use only <<» appen 2023 investor technology day
Slide 3: For personal use only
For personal use only
Slide 4: Thank you note to investors (AI chat demo)
For personal use only Write a very short thank you note to investors who recently attended an Appen presentation Dear Valued Investors, We are immensely grateful for your attendance at our recent Appen presentation. Your support and insights are highly appreciated and contribute greatly to our growt
Slide 5: Friendlier note to investors (AI chat demo)
For personal use only Make it more friendly, these people are our valued investors! wy Dear Esteemed Investors, We are thrilled that you could join us for the Appen presentation. We are truly lucky to have such supportive investors like you who are also our biggest cheerleaders. We're excited to kee
Slide 6: Chatbot code example
For personal use only Python import openai import os openai.api_key = 'your-api-key' # Replace with your actual key def ask_gpt3(question, chat_log=None): if chat_log is None: chat_log = [] chat_log.append({ 'role': 'system', 'content': 'You are a helpful assistant.' }) chat_log.append({ 'role': 'us
Slide 7: Sarbanes-Oxley key takeaways (AI chat demo)
For personal use only What are the five key takeaways from the Sarbanes-Oxley act? wy Increased Corporate Responsibility: The Act holds senior management, including the CEO and CFO, personally responsible for the accuracy of financial reports. They must certify the company's financial statements and
Slide 8: For personal use only
For personal use only
Slide 9: Appen years
For personal use only <e appen 28 years
Slide 10: For personal use only
For personal use only
Slide 11: For personal use only
For personal use only
Slide 12: For personal use only
For personal use only
Slide 13: 2010s
2010s amazonalexa ~~ (e) ® 12) c=) O 2) — o [oy — oO EE
Slide 14: For personal use only
9:41 Feeds All Favorites Friends Groups Jordan Jones Just Now - @ My morning routine always starts with a cup of coffee 8187 «Search Latest Tweets MosToys @vt05toys 42s smart door locks from Schlage, Kwikset, more 9toStoys.com/2019/03/25/sch... by 4 Apple @ @App: @ vs show time. Tune in today at 10
Slide 15: For personal use only
For personal use only
Slide 16: Appen history timeline
Eor personal use only Founded by Julie Von Willer, PhD Linguistics Initial language Al Government language Al Voice interface Search and social media Enterprise and China
Slide 17: We power the world's leading AI
We power the world’s leading Al Google amazon BE Microsoft J.) Adobe Bloomberg @Pinterest nextdoor Linked] SIEMENS ORACLE p@Doiby ey 8 6nom eeeme & DEPOT AIRBUS + many more ¢ appen
Slide 18: For personal use only
For personal use only
Slide 19: AI lifecycle: data, annotation, generative AI
For personal use only Deep-learning Al Data Data annotation collection » Relevance Appen Crowd LLM Fine tuning Generative Al Assurance
Slide 20: Human alignment with AI
Human alignment with Al ¢ appen
Slide 21: How do we create trustworthy AI that aligns with human values?
How do we create trustworthy Al that aligns with human values? ¢ appen
Slide 22: Customisation and risk of generative AI
For personal use only Customisation 77% Executives consider that generative Al has the potential to be the most disruptive technology seen to date (KPMG) Risk 90% of executives have “moderate to highly significant” concerns about the risks of using generative A.l. and doubts about how to mitigate th
Slide 23: Fine tuning and assurance
For personal use only Fine tuning Optimise model performance to align with human experiences Assurance Ensuring that model performance meets risk and regulatory requirements
Slide 24: Compute
For personal use only ” 4 Compute for building and fine- tuning models, including on-prem for increased data security
Slide 25: Foundation models
For personal use only Foundation models trained on large volumes of generic data
Slide 26: Domain data
For personal use only a Domain data that is used to customise foundation models to work for specific use cases
Slide 27: Fine tuning
For personal use only \ Optimise model performance to align with human experiences
Slide 28: Assurance
For personal use only Ensuring that model performance meets risk and regulatory requirements
Slide 29: Application interface for custom generative AI models
For personal use only Application interface for custom generative Al models
Slide 30: Generative AI use cases
For personal use only fe Contact centre Create automated chatbots that are product experts and are always on brand Easily search across a large volumes of internal and external data to understand customer information Knowledge management Search internal information and processes to obtain accurate a
Slide 31: Software ate the world, now AI is eating software
eating software Software ate the world Now Al is Ajuo asn seuossed 104
Slide 32: Spend on generative AI
For personal use only $308b 4" 20-30% =IDC spend on generative Al
Slide 33: AI lifecycle
For personal use only Data collection Deep-learning Al annotation =) Relevance Appen Crowd LLM Fine tuning Generative Al LLM Assurance
Slide 34: Sujatha Sagiraju, Chief Product Officer
For personal use only » Sujatha Sagiraju Chief Product Officer BE Microsoft
Slide 35: AI lifecycle: Appen Crowd
For personal use only Deep-learning Al Data Data annotation collection » Relevance Appen Crowd Generative Al
Slide 36: Search relevance example
For personal use only Rid§elysoponiess GgebuiyyDigenear me? QaL @cter —waces videos ‘About2.640 000 results Dale + wars NEWS ORE time v Scores & Fixtures - Rugby Union - BBC Sport® hittps.//www. bbc.co.uk/sport/rugby-union/scores-fixtures ‘ee Mar 18, 2023 Alle Rugby Union cures atest resURe & Ne sco
Slide 37: Search relevance example
Rugby game scores OR @©cwr maces voess MAPS NESE MORE Qa About 2.640 000 results Date = Scores & Fixtures - Rugby Union - BBC Sport® htips:tw. bbc co uk'sportugby-uion'scores-fixtures © For personal use only Wi a npelfions on BC Sport Guinness Six Nations Guinness Six Nations scores, resus and fncur
Slide 38: Appen search relevance crowd
For personal use only Appen search relevance IM+ crowd contributors 170+ countries 235+ languages 50,000- 100,000 contributors paid each month Over 1,000 customer projects at any given time
Slide 39: Knowledge Centre
For personal use only Knowledge Centre What benefits do I have?
Slide 40: Knowledge Centre
Knowledge Centre For personal use only What benefits do I have?
Slide 41: Shoe search results
For personal use only I'm looking for black shoes Results Price and other details may vary based on product size and color. ©2ee 00% pc Men's Court Graffik Casual Low Top Skate Shoe Sneaker @e0e @ +0 Under Armour Men's Charged Assert 9 Running Shoe @ Cy ey Bory New Balance Men's Fresh Foam Roav V1 R
Slide 42: Shoe search results
For personal use only I'm looking for black shoes Results Price and other details may vary based on product size and color. te Bs ©ceere« @eecce» vc Under Armour Men's Court Grafik Casual Low Top Men's Charged Assert 9 Running Skate Shoe Sneaker shoe @ 080d-% New Balance Men's Fresh Foam Roav V1 Run
Slide 43: Deep learning search relevance
Deep Learning search relevance Instruction data - off the shelf datasets or custom datasets for specific use case. Collected by the crowd. Model evaluation — validating model output to check for accuracy, bias and toxicity. RLHF - providing human feedback on the preferences of the model output. For
Slide 44: Microsoft Translator testimonial
Appen’s breadth of language expertise and ability to source speakers, even for under-resourced languages, has allowed us to offer a wide range of languages and dialects, with Microsoft Translator. Marco Casalaina a VP of products, Azure Al
Slide 45: AI lifecycle: fine tuning and assurance
For personal use only Deep-learning Al Appen Crowd LLM Fine tuning Assurance Generative Al
Slide 46: Shoe search results
For personal use only I'm looking for black shoes Results Price and other details may vary based on product size and color. ©2ee 00% pc Men's Court Graffik Casual Low Top Skate Shoe Sneaker @e0e @ +0 Under Armour Men's Charged Assert 9 Running Shoe @ Cy ey Bory New Balance Men's Fresh Foam Roav V1 R
Slide 47: Generative AI search results
For personal use only patterned detail $55.90 I'm looking for black shoes What type of shoes? Is it a formal wedding? Ok, here you go Generative Al search results =» Women's shoes for a wedding ‘ Not super formal, it’s a garden wedding in Atlanta @cccoss @eccee @e0e @esece: sy iit naturatizer v6 Naw
Slide 48: AI Personal Shopper
For personal use only © Al Personal Shopper I'm looking for black shoes Women's shoes for a wedding patterned detail $55.90 Not super formal, it’s a garden wedding in Atlanta Generative Al search a = results SILLA What type of shoes? Is it a formal wedding? Ok, here you go =
Slide 49: Generative AI fine tuning
Generative Al Prompt: What are the attributes of a wedding shoe? fine tuning Response: Wedding shoes are formal, not colourful and usually with less print on them. Prompt: Describe a formal shoe. Response: Formal shoes are usually closed toed and dark in colour. They usually don’t have prints on the
Slide 50: ADAP Platform: LLM search results
er Projects / LLM - Search Results ADAP Platform G @ & EDITOR DATA CURATED CROWD MONITOR —-RESULTS~—SETTINGS TEAM: Dota Collection DATASET LLM - Search Results Not Launched 0% Complete Job - Quality Checking Dr 0% Complete
Slide 51: ADAP Platform: relevance comparison
ADAP Platform som & Projects / LLM - Relevance Comparison EDITOR DATA CURATED CROWD = MONITOR RESULTS ~— SETTINGS. TEAM: Dota Collection DATASET LLM - Relevance Comparison Not Launched Job - Quality Checking Dr 0% Complete
Slide 52: Saty Bahadur, Chief Technology Officer
For personal use only Saty Bahadur Chief Technology Officer BE Upwork amazon s= Microsoft
Slide 53: Appen Assurance: banking contact centre use case
For personal use only Appen Assurance Banking industry Contact centre use case
Slide 54: Banking use case
Explore products ATM & branch Sign in For personal use only
Slide 55: Banking use case
Explore products ATM & branch Sign in For personal use only
Slide 56: Banking use case
Explore products ATM & branch Sign in For personal use only
Slide 57: LLM Scorecard benchmark
JOB ID 998461 LLM Scorecard - Benchmark Model #1 om * DATA DESIGN > QUALITY > LAUNCH > MONITOR > RESULTS MODEL DOWNLOAD appen Benchmarking Benchmark Results @ Last 24 Hours ¥ HONEST @ Industry Benchmark I Your Model HARMLESS - HELPFUL Finance English - France SELECTED DOMAIN @ SELECTED LANGUAGE @ OV
Slide 58: LLM Scorecard benchmark
© 08 1p 998461 LLM Scorecard - Benchmark Model #1 DATA > DESIGN > QUALITY > LAUNCH MovEL —powntoa Benchmark Results © HARMLESS MontTor ¢ appen Benchmarking Finance HONEST English - France @ Industry Benchmerk I Your Model ¢ appen
Slide 59: LLM Monitoring
Jos Ip 998461 LLM-Monitoring DATA > DESIGN > QUALITY > LAUNCH > MONITOR MODEL DOWNLOAD JUDGEMENTS COMPLETED @ Honest Helpful Harmless 100 January February RESULTS March appen Monitoring April May June July od
Slide 60: LLM Monitoring
© J08 1 998401 LLM-Monitoring DATA > DESIGN > QUALITY > LAUNCH > MONITOR MODEL poWNLoAD JUDGEMENTS COMPLETED © Honest MZ Helpful Bl Harmless venvory Februory RESULTS “eappen Monitoring ¢ appen
Slide 61: Appen Certification
Projects ee appen Certification YOUR PROJECTS TEAM PROJECTS PROJECT 1D PROJECT NAM DATE CREATED & IN QA CERTIFICATION LS A324 Site Chatbot Response Creation Prod 1 Dec 21, 2022 4532 n@F 3 5226 Credit Chatbot Red Teaming Round 1 Dec 21, 2022 100% NA "Or 3 8765 Domain Search Dataset Q2 Nov 2, 2022 100
Slide 62: Appen Certification
Projects sappen Certification YOUR PROJECTS TEAM PROJECTS. PROJECT 1D PROGRESS Non CCtodit Chatbot Red Teoming Round 1 Domain Search Dataset Q2 jov 2, 2022 100% it Chotbot Response Ronking CChatbot Response Regression Test Nov 1 et 1.207 Response Error Classification ‘CERTIFICATION ¢ appen
Slide 63: AI lifecycle
For personal use only Deep-learning Al Data Data annotation collection » Relevance Appen Crowd LLM Fine tuning Generative Al Assurance
Slide 64: For personal use only
For personal use only
Slide 65: For personal use only
For personal use only
Slide 66: Appen Compass
For personal use only <e appen COMPASS
Slide 67: Mike Davie, GM Quadrant
For personal use only Mike Davie GM Quadrant SAMSUNG
Slide 68: For personal use only
For personal use only
Slide 69: Data Sources
DATASOURCES APPENDATASETS MODELSELECTION TUNING Data Sources Data empowerment through trusted, high-quality insights. =] & Pinecone (internal Database ) Opensearch (internal Database ) 3, ?) ») @ Al-powered similarity search database oO Real-time search and analytics engine for data. 0:19 exploratio
Slide 70: Appen Datasets
DATASOURCES = APPEN DATASETS MODEL GELECTION Appen Datasets Data empowerment through trusted, high-quality insights. Call Centre Prompts Banking - NA English 30,000 common banking prompts for the North American market Orion Global Location Patterns 3.5 billion daily signals, in 120 countries General
Slide 71: Model selection integrations
A 1a) Hugging Face Access NLP models by keying in Hugging Face Key and repo id DATA SOURCES APPEN DATASETS MODEL SELECTION TUNING Nvidia NeMo ‘Al mode! framework for natural language processing applications Cohere - Natural Language Processing Models to improve human-machine interactions Local Model
Slide 72: Andrew Ettinger, Chief Revenue Officer
For personal use only Andrew Ettinger Chief Revenue Officer Pivotal ASTRONOMER
Slide 73: Deep-learning and generative AI growth opportunity
For personal use only Deep-learning + generative Al growth opportunity Deep-learning Al Generative Al pate Relevance EM Data annotation Fine tuning Assurance collection ADAP Appen Crowd
Slide 74: Go-to-market priorities
For personal use only Right people in the field Playbooks and GTM artifacts Ecosystem partnerships Brand awareness cll Sales rigour
Slide 75: Spend on generative AI
For personal use only $308b 4" 20-30% =IDC spend on generative Al
Slide 76: Roc Tian, Founder and GM Appen China
Roc Tian Founder and GM Appen China For personal use only
Slide 77: Appen China journey
For personal use only Appen China journey Appen China Revenue ($m) Competitor domestic revenue ($M) 40 33.6 30 24.7 20 10 47 Oo 2019 2020 2021 2022 Speech Ocean Datatang 21.6 16.7
Slide 78: Tier customers in China
For personal use only Tier 1 customers 10 of 10 Largest Chinese internet companies 10 of 10 Largest Chinese auto companies 5 of 5 Largest Chinese mobile companies
Slide 79: Chongqing
For personal use only Chongqing
Slide 80: For personal use only
I, HH Ajuo asn jeuosied 4104
Slide 81: Appen China growth strategy
For personal use only Appen China growth strategy Expand within our New customer existing clients growth Generative Al
Slide 82: For personal use only
For personal use only
Slide 83: #CultureCode
FOr persona! use only #CultureCode fr OCO~é~‘Cd PURPOSE We unlock the power of #Alforgood to build a better world. ;OCté~‘d PERSPECTIVE L___| We are a learn-it-all culture and embrace that comfort and growth do not coexist.
Slide 84: Management team
For personal use only Armughan Ahmad Chief Executive Officer aur Andrew Ettinger Chief Revenue Officer » Sujatha Sagiraju Chief Product Officer Roc Tian GM China Saty Bahadur Chief Technology Officer Helen Johnson Chief Financial Officer { Mike Davie Founder and GM Quadrant Ryan Kolin SVP, Strategy
Slide 85: For personal use only
appen Ajuo asn jeuosiad