Integrating AI - Business Applications
Download as PPTX, PDF5 likes1,305 views
This document provides guidance on integrating AI into organizations. It recommends aligning AI projects with business drivers to create value, focusing on scaling human capabilities with assisted and augmented intelligence, and taking a portfolio approach to AI innovation around customer needs. It also stresses the importance of developing an effective operating model with the right outcomes, team, tools, and iterative process, as well as understanding the range of AI and machine learning efforts.
Integrating AI - Business Applications
- 1. Integrating AI into your Organization: Business Applications Hal Kalechofsky, Ph.D.
- 2. Agenda • Some Guiding Principles • Thinking about AI in Business • Thinking Big about Big Data • Some Best Practices • Align with a Business Driver • Innovate around Customer Needs • Have an Effective Operating Model • Understand the Range of Efforts • Machine Learning and Deeper • Algorithms and Decision-Making • Value-Strategy-Execution-Metrics • Enterprise AI Readiness Intended Audience: Executives and decision makers
- 3. A Few Definitions AI is: The ability of machines to think, learn, and act like humans Machine learning is a type of artificial intelligence (AI) that provides computers with the ability to learn without being explicitly programmed. Machine learning focuses on the development of computer programs that can change when exposed to new data More like: Amplified Intelligence
- 4. Thinking about AI Assisted Intelligence: AI systems that assist humans in making decisions or taking actions. Hard-wired systems that do not learn from their interactions. Automation: Automation of manual and cognitive tasks that are either routine or non- routine. This does not involve new ways of doing things – it automates existing tasks. Augmented Intelligence: AI systems that augment human decision making and continuously learn from their interactions with humans and the environment. Autonomous Intelligence: AI systems that can adapt to different situations and can act autonomously without human assistance. Human in the Loop No Human in the Loop Image adapted from PWC: Sizing the Prize in Artificial Intelligence Look for the A B problem (Andrew Ng). Input A generates simple response B “If a typical person can do a mental task with < 1 sec of thought, it can probably be AI automated.”
- 5. Data is the new business model. Think Big about Data It is often not about who has the best algorithm, but who has the most data.
- 6. AI Integration: Best Practices #1 • Align with a business case/drivers - Business 101: Keep costs down, revenue strong, happy clients - Relentless focus on use cases (Connect to KPI) • No “One size fits all” solution (Understand effort) - Orient innovation towards customer success - Continuum of ways to implement AI for great outcomes.. • Match the solution to the problem - Don’t be a hammer looking for a nail - Mix business & technology in the right way • Low-hanging fruit - Focus on what is easily do-able - Github world-view • Add/evolve existing platforms/systems/resources - Leverage capital investments • Build towards data-shares - “What can be connected, will be connected” - The output of your program is the input of someone else’s AI Business
- 7. AI Integration: Best Practices #2 • Focus on scaling humans - Listen to customers and stakeholders (be consultative..) - Assist, Augment, and Automate • Operational or Strategic? - Decide broad-brush if you have operational or strategic opportunities - Eg. Robotic process automation can do a lot • Consider a portfolio approach to AI (Broad View) - A mix of quick-win projects tying to transformative long-term projects - Don’t get bogged down by technology, look at what others are doing • Invest in people (talent) - When it gets built, it will be built by a team - Consider partnerships and core competencies • Embrace it - Change leadership - People & AI together. AI is not about destruction, it is about job movement
- 8. Align with a Business Driver Achieving Operational Efficiency Improve Customer Satisfaction / Experience / Expectation Competitive Differentiation Eg. Improve decisions, improve process, Augment humans, improve production support, Removing legacy Eg. Predictive modelling, recommendations, decrease cycle time, anti-fraud Eg. New business model, Enhance business models
- 9. Portfolio Approach: Innovate with AI Around Customer Needs Business Value definition “Who, What, Why?” Strategic Customer/Market Needs Focused Use Cases Governance and Investment Technology Solution Prototype Deployment Architectural alignment Business à Technical requirements Innovate everywhere “Succeed fast or fail fast” Executive oversight Prioritization (value, complexity); Agility, Domain Knowledge Right tool for the right job Sourcing, solution evaluation, implementation “That which is measured, improves” KPIs, tangible value, learnings
- 10. You Need an Effective Operating Model Outcomes The Business Model Data Science Data Data Engineering Compute IT Tackle the right problems Build the right team Have the right tools Iterate the right way 1 23 4 It’s about aligning the right model, data, and infrastructure with the right outcomes Large, clearly defined business value • What “job” would someone “hire” your solution to do (Christensen)? • Who is the customer? Interdisciplinary • Hybrid, not unicorns • Don’t existing pattern match Deploy platforms/pipelines for efficiency • Be polyglot • Data Science |= Software Development Success is more “agile” than Agile • Data science is exploratory • The future is heterogenous Adapted from Carlsson, Wang, Forrester, Anaconda, 2018
- 11. Understand the Range of Efforts
- 12. Types of Learning • Supervised (inductive) learning – Training data includes desired outputs Ex.: Have labelled photos, train model to recognize new photos • Unsupervised learning – Training data does not include desired outputs Ex.: Have unlabeled photos of N people, divide into N clusters • Semi-supervised learning – Training data includes a few desired outputs Ex.: Supervised learning on labeled data only, then apply classifier to unlabeled data to generate more labeled examples • Deep Learning - Inspired by neural brain structure, methods based on learning data representations and abstractions, supervised or unsupervised • Reinforcement learning – Rewards from sequence of actions – Ex.: Game-playing, or robot putting object in a box
- 13. Things that AI does and does not do well • Classify/recognize images and sounds • Search the Web • Games: Chess, Jeopardy • Translate in many languages • Identify fraudulent trends/patterns • Work in deadly environments • Product recommendations • Personal Assistants • CRM leads, contacts, LTV predictions • Plagiarism checkers • Map/traffic applications • Reasoning • Problem solving • Speech recognition (still a “D”) • Moving in arbitrary environments • High-level planning and control • Manipulating objects or balance AI does well AI does less well* * But getting better fast ..!
- 14. Consider AI/ML Pipelines • Most of your deployments will resemble a pipeline more or less like this • There are large amounts of good open-source packages and models available • There are lots of niche AI companies that do certain specifics very well • There is also a lot of public data out there, as well as pre-crunched model APIs (Don’t re-invent the wheel..)
- 15. Algorithms and Decision-Making 4 Models (Michael Schrage, HBR 2017) 1) The Autonomous Advisor Algorithms are your strategic advisor, with human oversight 2) The Autonomous Outsourcer Algorithms are your business process outsourcing 3) The Autonomous Employee Software is a valued colleague, “machine-learning” first enterprise 4) All-in-one Autonomy Algorithms run company decisions; human leadership defers to algorithms
- 16. Expectations WHILE ALL THE RISKS OF AI ARE VERY REAL, THE APPROPRIATE BENCHMARK IS NOT PERFECTION BUT THE BEST AVAILABLE ALTERNATIVE.
- 17. AI-Powered “Healthcare” Enterprise Operations Smarter, Lower cost, Better Customer Outcomes V E S 5+ Years 2-4 Years 1-2 Years M Business Drivers: Operational Efficiency, Patient Outcomes, Competitive Differentiation No Humans in Loop Automation Humans in Loop Assisted Intelligence Sense Comprehend Predict A VSEM Model – Healthcare Example Predict Act • Machine/Human • Data at Rest/In Motion • EMR, Papers, Journals,… • Sensors / IoT • Interactions • Observing process, workflows • Aggregate • Search, Compare • Correlate • Troubleshoot • Problem-solve • Hypothesize • Model, Simulate • Statistical likelihoods • Reason, Diagnosis • Advise • Recommend • Evaluate • Claim rejections • Denials • No-shows • Prior authorizations • Patient re-visits • Wait times • Unnecessary procs • Post-treatments • Clinician admin time • User satisfaction • Patient volumes • Care quality • Disease incidence • Error rates • Preventative care • Disease prediction • Rev. per Dr/Employee • Cost per unit • Precision medicine • Population health Operational Experience Financial • Operate • Integrate/Implement • Continuous Improvement • Evaluate • Adapt, change • Repair, fix • Monitor, Follow-up
- 18. Enterprise AI Strategy Includes… • Data Life Cycle Management • Enterprise Knowledge Management • Enterprise Automation Strategy • Infrastructure Upgrades • Legacy Systems Deprecation • Security • Command Center • Governance Structure • Policies & Procedures • Change Management • Executive, IT & Business Buy In • Skilled Resources and External Experts
- 19. Enterprise AI Readiness WHAT: • Executive, IT & Business Buy In • Enterprise AI Strategy – Enterprise Data Strategy – Data / Intelligence Lifecycle Management – Enterprise Knowledge Management – Data Governance – Policies &Procedures • AI/ML Frameworks & Algorithms • New AI & Automation Infrastructure & Infrastructure Upgrades • Enterprise Automation Strategy • Change Management • Security Strategy • AI & Automation Command Center HOW: • Innovation Life Cycle / Innovation Management • Current State Assessment • Analysis of Use Case Commonalities • Creation of Future State AI & Automation Architecture • Creation of Future State phased roadmap • Deprecate Legacy Systems • Skilled Resources and External Experts • Legal, IP & Ethical Considerations
Editor's Notes
- #4: Objectives * Make it easier to scope, design, & build Cognitive/AI applications (Potential solution and vendors would go here) * Scalable process for Business & Technology mutual success * How to specify requirements to a Cognitive/AI engineering team * Document business value, capabilities, & information sets * Common language description for AI projects
- #5: Forms of AI in use today include digital assistants, chatbots and machine learning amongst others. Automated intelligence: Automation of manual/cognitive and routine/nonroutine tasks. Assisted intelligence: Helping people to perform tasks faster and better. Augmented intelligence: Helping people to make better decisions. Autonomous intelligence: Automating decision making processes without human intervention
- #6: Objectives * Make it easier to scope, design, & build Cognitive/AI applications (Potential solution and vendors would go here) * Scalable process for Business & Technology mutual success * How to specify requirements to a Cognitive/AI engineering team * Document business value, capabilities, & information sets * Common language description for AI projects
- #7: Think of AI, if you will, as intelligent people Where would you deploy them to best manage your business?
- #15: For example, CRISP-DM https://en.wikipedia.org/wiki/Cross-industry_standard_process_for_data_mining
- #18: V = Vision (the “Idea”, the future state, along with the “Why”) S = Strategy (the “What” we are doing) E = Execution (the How) M = Metrics, success indicators, measurements, value statements