4 Facts Every Exec Must Know About A.I. Automation

Please download to get full document.

View again

of 27
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
1. 4 Facts Every Executive Should Know About A.I. Enterprise Automation 2. Source: A.T. Kearney Learn Analyze inputs to create own knowledge, identify underlying patterns…
  • 1. 4 Facts Every Executive Should Know About A.I. Enterprise Automation
  • 2. Source: A.T. Kearney Learn Analyze inputs to create own knowledge, identify underlying patterns Make decisions Make decisions and predictions autonomously Perceive Understand various forms of inputs Artificial intelligence Machine Learning Support Vector Machines Decision Trees … Deep Learning1 (neural nets) Bayes Nets Genetic Algorithm Ontology Reasoner … What is Artificial intelligence? AI is a collection of technologies that enable machines to perceive, learn and make decisions How does it work? Source: A.T. Kearney / / Rules based Automation Supervised Learning / / Contextual Self Learning Automation Under Certainty Automation Under Uncertainty RPA
  • 3. We are now in the 4th AI boom cycle Source: A.T. Kearney •Specific applications •Supervised and unsupervised learning paradigms •Nature/inspired architecture (e.g.. genetic algorithms, neural nets) •Training by relatively large data sets Machine Learning Deep Blue vs Kasparov Breakthrough Machine Learning Enablers Deep Learning •Supervised, unsupervised and hybrid learning paradigms •Advances in algorithms (e.g. one shot learning, adversarial neural nets, etc.) •Training by big data and specialized hardware 1970’s 1990’s1980’s1960’s Knowledge-based intelligence Early expert system platforms •Specific applications •Pre-programmed and rule based learning paradigm (if- then reasoning) •Limited data •No commercial applications •Pre-programmed and rule based learning paradigm (if-then reasoning) •Limited data First AI Algorithms Academic research AI winters • In the 70’s: cumbersome expert systems fail to achieve mass adoption • In the 80’s: collapse of the market for first general purpose computers (Lisp Machines) 2012-2017 AI Boom AI Winter
  • 4. Significant acceleration of venture funding 280 185 130 104101 88 81 57 464240 $2.1 $1.4 2017 189 $0.6 $1.0 $0.7 121 $0.9 $0.8 $1.5 132 2015 $0.5 2014 $0.4 2016 $1.1 $1.0 96 $0.3 160 $1.4 153 169 60 $0.7 63 2013 $0.2 $0.3 $0.2$0.3 $0.4 2012 23 $0.2 Deals Disclosed Funding AI Funding History ($Bn) AI Acquisitions Source: CB Insights, A.T. Kearney 2016 2017 Jul Aug Sept Oct Nov Dec Jan Feb Mar Apr May Jun Jul
  • 5. But, jury is still out on AI’s practical impact on the enterprise Questions from the C-suite: Is the current “hype” sustainable? What is real? What are the near term opportunities? What are the implications on capabilities and organization? How to get started?
  • 6. 4 key facts - A.I. and its practical near term impact Fact 2: AI Is Being Used across Organizations but with a Limited Scope Fact 4: Adopting AI Is about More Than Technical Feasibility Fact 1: The AI Boom Is Sustainable and Should Not Be Ignored Fact 3: AI Is Ready for Deployment on Select Activities
  • 7. For the first time in history, AI is beating humans in certain tasks Image Recognition and Voice Transcription Achieve Human Performance By 2014-15 Voice TranscriptionImage Classification Fact 1
  • 8. Recent AI performance boom is driven by 3 underlying drivers Proliferation of Big (Learning) Data Massive Parallel Computing Better Learning Capabilities • Massive increase in “context rich” digital footprint, social network content and interactions • Amount of data is predicted to increase from 4.4 Zettabytes (1021 bytes) in 2013 to 44 ZB in 2020 • GPU chips allow cheap and fast parallel computing (10-20X faster learning) • Tensor chip could speed up learning by 100X Machine learning capabilities reached new highs along various dimensions: • Speed: faster learning and increasing use in real time applications • Accuracy: “human level” is achieved in increasing number of tasks • Versatility: rise of AI platforms that can be used for various applications • Agility: evolving and self-improving Advancing learning algorithms Deep learning neural networks create breakthrough in Supervised Learning algorithms 10-100X Faster Learning 10-100X Training Data ~3X patent applications Source: Teqmine, Nature, Press Releases, A.T. Kearney Fact 1 Drivers of Supervised Learning
  • 9. The underlying drivers are not slowing down but accelerating! Fact 1 Source: Cisco, Tractica, Nextbigfuture, Teqmine Analytics, A.T. Kearney Processing Power Advanced learning algorithms 12,200 513 2016 2025 +42.2% Proliferation of Big (Learning) Data • By 2020, 92% of workloads will be processed by cloud data centers • By 2020, 68% of workloads will be processed by public cloud data centers; 32% by private cloud data centers (2016E: 56% vs. 44%) Deep learning chipset market revenue ($m)15.3 12.9 10.8 8.6 6.5 4.7 20162015 201920182017 2020 Global Data Center IP Traffic (Zettabytes per year) +26.6% Processor trends 0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 2,200 2,400 2,600 2,800 20042000 2016E20122008 # of annual USPTO patent applications for Machine Learning
  • 10. Implication Executives can no longer ignore A.I. Fact 1
  • 11. Fact 2: AI Is Being Used across Organizations but with a Limited Scope Fact 4: Adopting AI Is about More Than Technical Feasibility Fact 1: The AI Boom Is Sustainable and Should Not Be Ignored Fact 3: AI Is Ready for Deployment on Select Activities
  • 12. AI breakthroughs largely limited to narrow AI, with structural limitations Fact 2 Narrow AI (e.g. Deep Learning) … …Human Brain in Reality VS Source: A.T. Kearney, New Yorker, MIT Tech Review Limitation of Narrow AI • Requires lots of training data to teach and calibrate • AI trained solutions for one problem cannot be easily generalized to related problems • Certain problems cannot be handled by AI – even if given infinite training data (see next section)
  • 13. Consensus expert view is that progress will be limited to “narrow AI” in the near term Fact 2 Narrow AI IntelligenceRule-based RPA Broad/General AI Cognitive Modes Rule-based inference Supervised Learning Unsupervised Narrow Learning Unsupervised Context Aware Learning Self Aware Unsupervised Learning Natural Language Processing • Spell and grammar check • Voice to text dictation • Personal assistant apps with basic voice Q&A • Real time dialogue and translation • Idiom, sarcasm, nuance articulation Computer Vision • Scanning typed characters in format forms • Facial recognition • Scanning handwriting • Complex classification (e.g. video segment search) • Vision systems in complex settings (i.e. vehicles) • Autonomous exploration agents Pattern Recognition • Loans risk inference based on rules • Fraud detection (e.g., based on known patterns) • Product recommendation based on hidden customer preference • Real time clinical diagnosis •Anticipate cyber attacks • Mimic intuition and creative connecting the dots Reasoning and optimization • Historic based predictive forecasting • Forecast using demand sensing input with learned segmentation • Discover hidden biases from forecasting data and input • Beat best-in-class human forecaster in specific domain • Beat best-in-class human forecaster in several domains Today By 2025 Beyond 2025 Source: A.T. Kearney, WEF expert panel interviews, Press releases, Company websites
  • 14. Fact 2 Implication Training data and domain insights are the “new AI oil”
  • 15. Fact 2: AI Is Being Used across Organizations but with a Limited Scope Fact 4: Adopting AI Is about More Than Technical Feasibility Fact 1: The AI Boom Is Sustainable and Should Not Be Ignored Fact 3: AI Is Ready for Deployment on Select Activities
  • 16. Narrow AI is suited for automating A  B type applications Fact 3 Input A Response B Use Case Picture Are there human faces? (0 1) or Photo Tagging Loan Application Will they repay the loan (0 or 1) Loan Approvals Ad + User Information Will user click on ad? (0 or 1) Targeted online marketing; web page optimization English Sentence French Sentence Language translation Market Pricing Data Buy or Sell Commodity Trading What Narrow A.I. Is Good At: Learning AB Type Activities Source: ”What AI Can and Can’t Do”, HBR; SME interviews; A.T. Kearney analysis
  • 17. Most of the success stories in the news solve AB problems Fact 3 AB activity: automated buy/sell decisions AB activity: aortic valve blockage classification AB activity: SKU/location demand forecasting
  • 18. AI is not good at complex activities requiring context Fact 3 Personality Episodic Memory Social Learning Context • Exhibit human like emotions and quirks • Perform tasks based on output from past activity events or sessions • Interpret and communicate via tacit cues and signals • Perform complex inference based on context rather than direct training data Learn topics from unstructured data Episodic Memory Source: SME interviews; A.T. Kearney analysis
  • 19. Fact 3 Implication Focus on identifying high impact AB type activities for priority opportunities
  • 20. Fact 2: AI Is Being Used across Organizations but with a Limited Scope Fact 4: Adopting AI Is about More Than Technical Feasibility Fact 1: The AI Boom Is Sustainable and Should Not Be Ignored Fact 3: AI Is Ready for Deployment on Select Activities
  • 21. Organizations must consider adoption drivers when deploying AI solutions Fact 4 SNAPCHAT, WHATSAPP UBER CLOUD One-time Cost − Capex − Talent − Asset specificity of Capex and Talent ➢Limited capex; digital app based ➢Core software engineers; UX/design/consumer talent also key ➢No physical capex investments ➢Software engineers and increasing data scientists ➢Limited capex – SaaS flexible pay Switching Cost − Education / learning cost − Social & Cultural acceptance ➢Minimal cost to download and install app ➢Users get social media ➢Minimal cost to download and learn app ➢Easier to use than cabs ➢ Regulatory challenges ➢B-case benefit take time ➢Paradigm shift (legacy transition; security risks) Eco-System Needs − Need for complementary technologies/processes ➢Smartphone, mobile already in place and mature ➢Cars, smartphone already widely adopted; smart maps/traffic apps exist ➢Data storage and bandwidth low cost exponential growth Unlocking Network Externality Value ➢Very strong externality effects (friends, celebrities, content) ➢Strong externality effect (more cars, more users and vice versa) ➢Modest synergies with longer lead time to observe externalities Time to Mass Adoption 2-3 Years 3-5 Years 7-10 Years Adoption Drivers A.I. Application / Use Cases • What are the underlying hurdle for adoption dimension? • Do the adoption driver complexities appear more like Twitter or Cloud? Source: A. T. Kearney Analysis
  • 22. Example: both use cases are technically feasible today – one will take only 2-3 years while the other will take 5-10 years to adopt Fact 4 75% 25% Narrow AI use cases Challenges of technology adoption A.I. enabled use-case One-time cost Switching cost Eco system needs Hurdle to system externality Use Case 1. AI bot for extracting data from structured context and recognizing letter shapes Use Case 2. AI bot for behavioral profiling and sentiment analysis for customer contact engagements HighLow – 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 Use Case 1 Drivers of A.I. use case adoption A.I. solution adoption (BPO/contact center client example) 100% Use Case 2
  • 23. Enterprise automation is more than just having an A.I. algorithm – require an E2E solution + adoption strategy Fact 4 Technology Data Mgmt. Process Design Team Structure Change Mgmt. • What algorithms (supervised learning neural nets vs. genetic etc.)? • Which solution (Azure vs. Tensor)? • Which technology solution provider? • Should we buy external data to augment internal? • Can we pool data from data lakes? • Should we automate existing process? • Or design a new process from a clean slate? • What is the business case to do this? • Who are the key stakeholders whom we need to get on board? • What organizational and cultural hurdles do we need to address? • What teams & roles need to be established? And, with what skillsets? • Do we have people internally who can develop into the roles? Or should we hire from outside?
  • 24. Implication Design AI holistically – revamp the entire operating model Fact 4
  • 25. Thank you Michael Hu Partner, A.T. Kearney michael.hu@atkearney.com www.linkedin.com/in/mhuspace @mhu_snowcrash
  • 26. Recent IC and perspectives
  • 27. A.T. Kearney is a leading global management consulting firm with offices in 40 countries. Since 1926, we have been trusted advisors to the world's foremost organizations. A.T. Kearney is a partner-owned firm, committed to helping clients achieve immediate impact and growing advantage on their most mission-critical issues. For more information, visit www.atkearney.com. Americas Atlanta Bogotá Boston Calgary Chicago Dallas Detroit Houston Mexico City New York San Francisco São Paulo Toronto Washington, D.C. Asia Pacific Bangkok Beijing Brisbane Hong Kong Jakarta Kuala Lumpur Melbourne Mumbai New Delhi Perth Seoul Shanghai Singapore Sydney Tokyo Europe Amsterdam Berlin Brussels Bucharest Copenhagen Düsseldorf Frankfurt Istanbul Lisbon Ljubljana London Madrid Milan Moscow Munich Oslo Paris Prague Rome Stockholm Stuttgart Vienna Warsaw Zurich Middle East and Africa Abu Dhabi Doha Dubai Johannesburg Riyadh
  • We Need Your Support
    Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

    Thanks to everyone for your continued support.

    No, Thanks