Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Four Eras of Analytics in Government and Elsewhere: From Artisanal Analytics to Augmented Automation Thomas H. Davenport Babson College/MIT/Deloitte University of Maryland April 21, 2017 1 | 2017 © Thomas H. Davenport. All Rights Reserved Four Analytical Eras—Accelerating! 1.0 2.0 3.0 4.0 Artisanal analytics Big data analytics Data economy analytics Cognitive analytics 1975-? 2001-? 2013-? 2017-? 2 Analytics 1.0│The Artisanal Era Artisanal 1.0 Analytics ►Primarily descriptive analytics and reporting ►Internal, small, structured data ►“Back office” teams of analysts ►Internal decision support focus ►Predictive models based on human hypotheses 3 | 2017 © Thomas H. Davenport All Rights Reserved Analytics 2.0│The Big Data Era Artisanal 1.0 Analytics 2.0 Big Data Analytics ►Complex, large, unstructured data ►New computational capabilities required ►“Data Scientists” emerge ►Online firms create “data products” 4 | 2017 © Thomas H. Davenport All Rights Reserved Analytics 3.0│The Data Economy Era Artisanal 1.0 Analytics 2.0 Big Data The Data 3.0 Economy ►Seamless blend of traditional analytics and big data ►Analytics core to the business ►Data and analytics-based products in every business ►Industrialized decision-making at scale 5 | 2017 © Thomas H. Davenport All Rights Reserved Analytics 3.0│Private Sector Goals Developing products and services based on data and analytics—now available to every industry ► “Precision agriculture” offerings for farmers ► Conditional and predictive services for industrial equipment ► In telecom, analytical recommendations and insights from mobile devices Data and analytics-based decisions at scale and supporting the front line of organizations ► Real-time routing ► Granular, targeted marketing programs ► In telecom, treating every customer differently 6 Social Security 3.0—Fraud Prevention Focus ► Primary focus reducing fraud in disability and identity theft contexts ► Text mining and analytics allows “express lane”--rapid approval of 20% of disability claims among very ill and elderly ► Predictive analytics used to identify disability fraud ► Analytics used to identify holders of duplicate SS numbers for identity fraud 7 | 2017 © Thomas H. Davenport All Rights Reserved NYPD 3.0—Situational Awareness Focus ► “Domain Awareness System” takes crime analytics to the next level ► Massive data sources, including: ► 9000 closed circuit TV cameras ► 500 license plate readers, 2 billion reads ► Audio gunshot detectors over 24 sq. miles ► 54 million 911 calls, converted to text ► 100 million summones, other crime records ► “Predictive policing” to 10,000 cops’ smartphones 8 | 2013 © Thomas H. Davenport All Rights Reserved Analytics 4.0│The Cognitive Era Artisanal 1.0 Analytics The Data 3.0 Economy 2.0 Big Data 4.0 Cognitive ►Analytics used to make automated decisions ►Mostly “autonomous analytics” ►Replacement of human tasks—digital/physical ►Augmentation is human focus 9 | 2017 © Thomas H. Davenport All Rights Reserved A Constellation of Cognitive Technologies ► Machine learning ► Neural networks/deep learning ► Natural language processing/generation ► Rule engines ► Event stream/complex event processing ► Robotic process automation ► Custom integrations and combinations of these in a “cognitive cloud” 10 | 2017 © Thomas H. Davenport. All Rights Reserved Codelco 4.0—Cognitive for Safety Chilean national copper mining company has emphasized automation for worker safety Started with remotely-operated rock hammers in 1990s Wide use of autonomous trucks, mine trains Truck loading and smelting increasingly automated Integrated operations center monitors and controls automated devices Goal to eliminate underground human miners by end of this year 11 | 2013 © Thomas H. Davenport All Rights Reserved Vanguard 4.0—Cognitive for Investor Advice “Personal Advisor Services” combines automated and human investment advice Proof-of-concept for Substantially lower cost (30 basis points) and lower wealth thresholds than most human investment advice sources Assets of $50B under management and growing rapidly 12 | 2013 © Thomas H. Davenport All Rights Reserved Defense Health Agency 4.0— Cognitive for Federated Data Used machine learning to read analyst reports and identify machine learning as an important technology for the DHA Used SEMOSS, open source tool developed for the Military Health System, to gather and match patient data across five different electronic medical record systems Used same tool to identify redundant systems that could be shut down with relatively low impact—saved $58M Working on projects to predict patient wait times and disease onset 13 | 2013 © Thomas H. Davenport All Rights Reserved NASA—Cognitive for Back-Office Financials Large-scale implementation of robotic process automation at National Shared Services Center Proofs-of-concept for four financial processes—funds control, funds distribution, technology spending, shared services financials More TK 14 | 2013 © Thomas H. Davenport All Rights Reserved Why Move to Cognitive? Expensive labor Tedious work Too much data Humans poor decision-makers Powerful technologies 15 Are Knowledge Workers Next to Be Automated? Cognitive/ Analytical Computers Transactional Computers Mechanical Systems Knowledge Work Jobs Admin/ Service Jobs Manual Labor Jobs 18th-19th C. 20th C. 21st C. 16 My Answer Is…Yes…and No ► Many knowledge work job tasks will be automated ► Some knowledge workers will lose their jobs, depressing hiring ► 8 lawyers where there were 10 ► There will be a lot of jobs (no one knows how many) working alongside smart machines ► Immense productivity gains could fund retraining and redeployment of people ► But workers can’t afford to be complacent 17 | 2017 © Thomas H. Davenport. All Rights Reserved Ten Knowledge Work Jobs with Automatable Tasks 1. Insurance underwriter—the oldest automated profession 2. Lawyer—e-discovery, predictive coding, etc. 3. Accountant—automated audits and tax 4. Radiologist—automated cancer detection 5. Reporter—automated story-writing 6. Marketer—programmatic buying, focus groups, personalized e-mails, etc. 7. Financial advisor—”robo-advisors” 8. Financial asset manager—index funds, trading 9. Programmer—automated code generation 10. Quantitative analyst—machine learning, etc. 18 | 2017 © Thomas H. Davenport. All Rights Reserved The Impact on People: Automation or Augmentation? ► Like freestyle chess, but applied to business ► Better than humans or automated chess systems acting alone ► Humans can choose among multiple computerrecommended moves ► Humans know strengths and weaknesses of different programs ► Automation is a race to the bottom ► Most current cognitive projects involve augmentation 19 | 2017 © Thomas H. Davenport. All Rights Reserved Five Ways of Stepping ► Step in—humans master the details of the system, know its strengths and weaknesses, and when it needs to be modified ► Step up—humans take a big-picture view of computer-driven tasks and decide whether to automate new domains ► Step aside—humans focus on areas they do better than computers, at least for now ► Step narrowly—humans focus on knowledge domains that are too narrow to be worth automating ► Step forward—humans build the automated systems 20 | 2017 © Thomas H. Davenport. All Rights Reserved What’s Your Entry Point into Cognitive? Mostly Buy • Existing vendor’s software with cognitive capabilities • Pick a small project and a low-hanging fruit vendor • Start with IT automation Some Build, Some Buy • “Autonomous analytics” with statistical machine learning • Go big with “transformative cognitive computing” • Give chatbots a shot Mostly Build • Make an existing application smarter or more autonomous • DIY with open source Becoming a Cognitive Organization ► Pick an entry point, and start some pilots ► Pick the right cognitive technology for your problem ► Take an augmentation perspective from the beginning ► Get good at work design for smart humans and smart machines ► Give your people the options and the time to transition to them ► Put someone in charge of this 22