main street, meet mr watson - matt coatney
TRANSCRIPT
Main Street, Meet Mr. Watson
The Accelerating Commoditization of Smart Machines and How Businesses Can React
Matt Coatney Director, WilmerHale LLP Founder, Five Spot Research Ltd
[email protected]/in/mattcoatney
@mattdcoatney
A call to action: smart machines are here to stay
Key technologies and when to use them
How to prepare your organization
Key takeaways
Lily pads
Lily pads
Technology introduction AND adoption are accelerating
Source: Wall St Journal/Asymco
1900 1920 1940 1960 1980 2000 2020
StoveTelephone
ElectricityCar
RadioWasher
RefrigeratorTV
DryerAC
DishwasherColor TV
MicrowaveVCR
Game ConsolePC
CellphoneInternet
SmartphoneHDTV
Tablet
Advanced technologies ARE within reach
1-2 years 3-5 years
Grouping similar itemsClustering and classification
Predicting behavior andmaking recommendations
Classification, regression and association mining
Handling very large dataCluster computing, NoSQL, etc.
Automating decisions andimproving interaction
Deep learning/AI
Grouping similar items
What it is Clustering: groups similar objects together (no pre-assigned categories)
Classification: groups similar objects together based on category samples
Common uses Content categorization and tagging
Customer and market segmentation
Medicine and insurance
Examples of vendors
Full-text concepts
Document title
Document metadata(date, author, practice, etc.)
Grouping similar items: classification example
Brief
Contract
Memorandum
Decision tree
Association rules
Grouping similar items: clustering example
http://www.greenbook.org/Content/TRC/4ExMarketSeg.pdf
Non-Traditionals (internet)
Direct Buyers (mail/phone)
Budget Conscious
Agent Loyals (personal touch)
Hassle-Free (passive)
Survey data
Geodemographic data
Credit information
Self-Organizing Map(Neural Network)
Predicting behavior and making recommendations
What it is Classification: predict new category outcomes based on similar objects
Regression: predict new numeric values based on similar objects/past performance
Association: predict commonly co-occurring objects, e.g. items bought together
Common uses E-commerce – more like this, frequently bought together, people who viewed this also viewed, etc.
Commerce – stocking, supply chain/distribution
Fraud detection and cyber security
Examples of vendors
Predicting behavior and making recommendations: regression example
Doctors’ visits
Procedures
Prescriptions
Hospital stays
Clustering and Regression Models
(Multiple Approaches –Ensemble/Panel of Experts)
Likelihood of hospitalizationin the next year
Predicting behavior and making recommendations: association example
http://readwrite.com/2008/07/16/strands_brings_recommendation http://pivotal.io/big-data/case-study/facilitating-data-analysis-to-better-understand-and-serve-customers-zions-bancorporation
C1 checking biz card merchant payrollC2 biz card merchantC3 merchant payrollC4 merchant biz card checking….
Association Rule Algorithms
Business card => Merchant AcctChecking => Business card…
Handling very large data
What it is Cluster computing: large arrays of commodity hardware
No SQL: efficient storage and retrieval for very large, semi-structured data
Scalable machine learning: clustering, classification, etc. optimized for large data
Data visualization: techniques and tools for meaningful display of massive data sets
Common uses High-volume transactions (e.g. customer interactions, web logs)
Social media interactions
“Internet of Things” (IoT) sensor data
Scientific computing
Examples of vendors
Handling very large data
Source: Capgemini
Handling very large data: Hadoop example
http://www.intelcloudbuilders.com/docs/cloudera_WP_10_Common_Hadoopable_Problems.pdf
Customer churn at a telecom company
Customer information
Call log data
Social media data
Location/cell coverage
Handset replacement and current market options
Hadoop cluster and related analysis components
(e.g. Mahout)
Likelihood a customer would leave the carrier (e.g. friends
leaving, coverage issues)
Handling very large data: more examples
http://www.intelcloudbuilders.com/docs/cloudera_WP_10_Common_Hadoopable_Problems.pdf
Risk Modeling
“… A very clear picture of a customer’s financial situation, his risk of default or late payment and his satisfaction with the bank and its service.”
Ad Targeting
“The model uses large amounts of historical data on user behavior to cluster ads and users, and to deduce preferences.”
Handling very large data: more examples
http://www.intelcloudbuilders.com/docs/cloudera_WP_10_Common_Hadoopable_Problems.pdf
Retail Promotion Campaigns
“Hadoop was able to store the data from the sensors inexpensively, so that the power company could afford to keep long-term historical data around for forensic analysis. As a result, the power company can see, and react to, long-term trends and emerging problems in the grid.”
“The retailer loaded 20 years of sales transactions history into a Hadoop cluster. It built analytic applications on the SQL system for Hadoop, called Hive, to perform the same analyses that it had done in its data warehouse system—but over much larger quantities of data, and at much lower cost.”
Power Failure Prediction
Automating decisions and improving interaction
What it is More processing + better algorithms + much more data
Deep learning: incrementally trained, stacked neural networks; allows more complex, nuanced patterns to be learned
Ensemble/panel of experts: improved performance combining multiple approaches
Common uses Cognitive tasks: answering questions, speech/image/audio recognition, game play
Improved human-computer interaction
Healthcare (e.g. Watson cancer diagnosis and treatment)
Examples of vendors
Automating decisions and improving interaction: an example
The shrinking adoption curve
Then…
10-20 Years
Window ofOpportunity5-10 years
…Now
5-7 Years
3-4 years
Timing is everything
Then Now
Embrace uncertainty
Think use cases before technology
Experiment, experiment, experiment
Fire bullets, then cannonballs
What can we do?
It’s an exciting (and scary) time
WE ARE HERE
Thank You
Matt Coatney Director, WilmerHale LLP Founder, Five Spot Research Ltd
[email protected]/in/mattcoatney
@mattdcoatney