When Milliseconds Matter:

Transcription

1 When Milliseconds Matter: Architecting Real-Time Analytics Into Systems Jose Castano April 28, 2015 Leaders know they must leverage data to outperform, but are struggling to hit the mark Drive top-line growth via transformational new services Dramatically maximize yield on investments in standard electronic business practices 80% 6% -4.3% $226M 16% $100M of marketers send same content to all subscribers of businesses extremely satisfied with ability to use customer data for decisions annual decrease in customer lifetime value for firms that don t use engagement analytics estimated fraud loss to healthcare government tax revenues lost to noncompliance typical banking regulation noncompliance fine More sophisticated analytics techniques must be put in play IT must be exploited as a business strategy 2 Sources: Aberdeen Group, NHCAA, Association of Certified Fraud Examiners 1

2 An enterprise analytics strategy must be able to support a wide variety of decision types Decision Flow Strategic Tactical Feedback Loop Guide overall direction of the enterprise Should we expand overseas? Manage and control operations What product should we promote today? Handle every customer interaction Is this payment request valid? How often is this decision made? What is the value/risk? How repeatable is this decision? What type of analysis does the decision require? How quickly must the decision be reached? Source: James Taylor, Decision Management Solutions: Transforming Systems From Transactional Systems to Decision Management Systems, 2015 But, outdated views of infrastructure can impede progress Live Significant complexity Separated data warehouses Analytics latency Transactional data is not readily available Lack of synchronization is not easily aggregated and fresh duplication Multiple copies of the same data Predictive Scoring Predictive Modeling Excessive costs Of moving data around 4 2

3 An enterprise analytics architecture is required to transform systems into Decision Management Systems Strategic Right-time analytics Decision Flow Tactical Feedback Loop Consistent, accurate view of data for uniform decision results Richness of capability to support every decision type Range of responsiveness for time-to-value from days to msecs 5 Right-time analytics Real-time analytics Focus for today: injecting insights into operational decisions decision management is a business Strategic discipline, supported by operational and analytics software, that enables organizations to automate, optimize and govern repeatable Tactical business decisions to improve the value of customer, partner and internal interactions Decision Flow Feedback Loop decisions are high volume, low latency, highly repeatable and high value in aggregate They require analytics that are closely linked to transactional systems so that decisions can be automated, in real-time Key considerations: richness of analysis and ability to maintain SLAs 6 3

4 Focus on Analytics in Business Applications What is realtime analytics? It s about leveraging the power of analytics in the transactional moment to achieve a more favorable outcome for a transactional event, while the event is in progress 7 This is the time during which a transactional event is still occurring Someone is shopping at a store Someone is on the phone with a customer service representative An insurance claim is being adjudicated on a processing system A purchase using a credit card is being processed An electronic payment is being processed The person visiting a store buys more than he or she otherwise would have What would have been an over-payment is stopped before it gets out the door The person on the phone, who was about to cancel a service, instead re-ups A commission of fraud is stopped before it is effected Use Case: enhanced suggest-selling at minimarts A gasoline retailer operates mini-marts at its stations The retailer has a loyalty program, and had collected a huge amount of data about customers purchases, but didn t really see how this data could be effectively leveraged Mini-mart employees were encouraged to suggest-sell, i.e., to suggest to customers additional items they might want to buy Success of this program depended largely on individual employees knowledge of customers buying habits 8 8 4

5 Enhanced suggest-selling at mini-marts: Getting smarter, and getting faster The gasoline retailer implemented SPSS to make sense of the vast amount of purchase data for sales in general and for sales to loyalty-program customers Patterns were detected, and predictive analytics could be enabled through models The company also acquired a DB2 Analytics Accelerator, to get suggest-sell queries executed in seconds in time to enable suggestions before the instore checkout had completed Sales went up Customer satisfaction improved data-driven purchase suggestions were often what a customer actually wanted to buy but had forgotten Everything is done on the customer s timetable, in seconds (CIO) 9 9 Real-time predictive customer intelligence leads to smart business decisions at the point of impact 1. Build long-term customer relationships 2. Drive one decision one interaction at a time 3. Maximize customer lifetime value Telco Providers When a high-value, long-tenure customer calls with a bill query within six months of the end of his / her contract, offer him / her a complimentary handset upgrade if he / she renews in advance. Customer repeatedly called tech support with mobile Internet issues comes to the website to check process to transfer his / her number. Initiate a chat, check problems are now resolved, refund two months data charges by way of apology, and offer discount on a new handset. Banking Providers Customer calls for password reminder on self-trade account. Pattern of usage shows limited activity and few returns; offer upgrade to managed investment offering. Customer s transactions show multiple trips to Asia; offer international emergency cover, and provide information about commission-free ATM withdrawals from our regional partner. Insurance Providers When confirming to a customer that his / her claim will be settled, direct him / her to a repair shop which, although it may take 1 2 days longer than other options, has higher quality ratings. When a customer calls in with a coverage query, let them know they can save money by combining their two policies into a single multi-car one. 5

6 The worldwide cost of occupational fraud has risen 9% over the past two years, to $3.7T USD annually The cost of fraud is considerable and rising, not falling The typical organization loses 5% of its revenues to occupational fraud each year (no change) The median loss caused by the occupational fraud cases in our study was $140K (up 4%) 20% of the cases were greater than $1M (up 2%) Fraudsters are good at hiding their behavior The frauds reported to us lasted a median of 18 months before being detected (no change) It is nearly impossible to recover fraudulent losses 58% of the victim organizations had not recovered any of their losses due to fraud, and only 14% had made a full recovery Yet, most organizations still rely on after the fact methods Over 40% of all cases were detected by a tip more than twice the rate of any other detection method 11 Sources: Association of Certified Fraud Examiners (ACFE) Report to Nations, 2012 and 2014 IBM s fraud management point of view Detect Detect in real time if a transaction, request, application, document, etc. is potentially fraudulent by applying models and rules in real time Detect Detect fraud within a business process Prevent Take action in real time when it matters Prevent Stop processing known fraud, or encourage fraudsters to abandon their objective by showing more is known than they think should be known about their activities and intentions Investigate Gather data about fraudsters and/or schemes DETECTED or DISCOVERED; build cases for prosecution, recoveries, or denial of payments. Build watch lists and rules Investigate Confirm fraud for prosecution, recovery, rules and watch lists Fraudster Discover Find Fraud within the Discover Discover fraud by retrospectively reviewing past data and identifying individuals or organizations that may be conducting fraudulent activities 12 6

7 IBM s fraud management point of view z Systems optimization focus: best results based on location of operational data Detect Detect in real time if a transaction, request, application, document, etc. is potentially fraudulent by applying models and rules in real time Detect Detect fraud within a business process Prevent Take action in real time when it matters Prevent Stop processing known fraud, or encourage fraudsters to abandon their objective by showing more is known than they think should be known about their activities and intentions Transaction / Batch Workload Investigate Gather data about fraudsters and/or schemes DETECTED or DISCOVERED; build Orchestration cases for prosecution, recoveries, or denial of payments. Build watch lists and rules Investigate Confirm fraud for prosecution, recovery, rules and watch lists Fraudster Discover Find Fraud within the Business Rules Discover Discover fraud by retrospectively reviewing Predictive past data and identifying Analytics individuals or organizations that may be conducting fraudulent activities 13 Use Case: Bank: incrementally enhance existing fraud detect with predictive approach Banking Transaction Initiated Transaction Determine fraud risk Continue Pend/investigate, etc. scores Historical evaluates Transaction Fraud rule engine In place today Extend with additional insight from predictive analysis Transaction Merchant Model Account Results of large, complex queries Historical Account IBM DB2 Analytics Accelerator Business Goal: reduce loss due to card fraud, reduce card deactivation, grow revenue associated with card purchases, reduce call center costs, improve service yielding preferred card usage Approach: incorporate aggregate data from geographic location, merchant, issuer and card history into existing card authorization business flow to reduce fraud while preserving transactional SLAs Integrated high performance query optimizations enable client to aggregate data several times a day and use this complex data as part of real-time fraud detection process Enhance with predictive scoring integrated with fraud detection transaction for even more preventive capabilities 7

8 Use Case: Insurer: minimize loss from claims overpayments Intake Claims Prepare Claims 2 System of Record Adjudicate Claims Deliver Benefit 1 Recover Overpaid Claims Predictive Analytics leveraging IBM SPSS Modeler with Scoring Adapter for z Systems Scoring adapter Claims DB (DB2 z/os) Descriptive Analytics leveraging IBM DB2 Analytics Accelerator Challenge 1: Complex reports for overpaid claims not completing on time, result is monetary losses Solution: Integrate optimized analytics of IDAA with overpayment reporting of transactions Benefits: Up to 2000x improvement in speed of overpayment reports LOB users enabled to respond with more agility to overpayment trends Informed decisions at the right time Challenge 2: Stop improper payments prior to payment, avoid pay & chase, meet SLAs Solution: Integrate predictive analytics into claims adjudication for analytics in place Benefits: Very efficient scale for analytics Scale requirements only achievable with analytics as part of transaction flow Expected results of efficient in-transaction analytics can be multi-million dollars per year 15 Predictive analytics should be a key element of your real-time decision strategy Predictive models detect patterns Deviation from expected behavior can isolate bad (or good) behavior, trigger additional actions or new targeted marketing and up-sell / cross-sell offers Like the real world, predictive models are not binary Understanding how closely a pattern of behavior matches a known pattern of bad (or good) behavior can help uncover crimes or non-obvious opportunities Predictive models can have many variations Can be built to assess only specific transactions or more generically for all transactions Multiple layers of models can be invoked for increasing sophistication of analysis, triage leading to further inspection of contributing factors and weights 16 8

9 Starting point: where are the transactions and data that support real-time operations located? Systems of Record: Customer Accounts Payments Purchases Card Claims etc. Distributed data environments There s a very high probability that your operations and data of record are maintained on a mainframe Worldwide, 92% of top banks, 84% of top insurers, 92% of top retailers, 90% of top airlines rely on the mainframe As do 71% of the Fortune Global 500 The mainframe processes roughly 30 billion business transactions every day 55% of ALL enterprise applications need the mainframe to complete transactions How can we leverage the mainframe for incorporating analytics into operational decisions? 17 The infrastructure you need: compare approaches Move the data to the analytics Systems of Record Customer Accounts Payments Purchases Card Claims etc. Considerations: network Process Orchestration Predictive Models Business Rules Can performance / throughput SLAs tolerate data movement and network traffic? Can models integrate large volumes of historical data with incoming transactions to deliver the most accurate outcomes? Can security for sensitive data be maintained across multiple zones? Can audit trails be maintained to satisfy regulations? Can availability and BC/DR objectives be met? Can 100% of transactions be richly analyzed without user impact? Place the analytics with the data Better approach? Systems of Insight Customer Accounts Payments Purchases Card Claims etc. Process Orchestration Predictive Models Business Rules 18 9

10 The key elements behind real-time operational decisions Right-time analytics Real-time analytics Decision Flow Strategic Tactical Feedback Loop MODELS RULES INTEGRATED PROCESSES Predictive modeling, business rules and process flows together enable the most effective decisions Integrating analytics with transactions transforms operational systems into Decision Management Systems 19 decisions (Real Time Analytics) Webinar decisions: high volume, low latency, highly repeatable and high value in aggregate Require analytics that are closely linked to transactional systems so that decisions can be automated, in realtime Key considerations: richness of analysis and ability to maintain SLAs 10

11 10/9/2015 z Systems: when analytics are business-critical Analytics as part of the flow of business Insights on every transaction z13 is the first mainframe system with embedded analytics providing real-time insights on all transactions. This capability can help guarantee the ability of the client to run real-time fraud detection on 100 percent of their business transactions. * 21 *z13 is the first mainframe system with embedded analytics providing real-time insights on all transactions. This capability can help guarantee the ability of the client to run real-time fraud detection on 100 percent of their business transactions. [2] Based on IBM internal tests comparing IBM zenterprise Analytics System 9700 with a comparably priced, comparably tuned competitor Eighth Unit configuration (version available as of 12/31/2014), executing a materially identical 10 TB BIDAY Fixed Execution workload in a controlled laboratory environment. Competitor configuration: Eighth Unit including competitor recommended software options and features. IBM configuration: z13 platform with 1CP and 3 ziips with 128GB memory and DB2 Analytics Accelerator Full Rack (N ) with 7 S-blades (140 Intel E5-2680v2 2.8GHz cores and 128 GB RAM), 2 Hosts (1 active 1 passive) with 20 Intel E5-4650v2 2.4GHz cores each and 12 disk enclosures, each with GB SAS drives. Results may not be typical and will vary based on actual workload, configuration, applications, queries and other variables in a production environment. Users of this document should verify the applicable data for their specific environment IBM Corporation 22 11

12 Copyright IBM Corporation 2015 IBM Corporation Systems and Technology Group Route 100 Somers, NY Produced in the United States of America February 2015 IBM, the IBM logo, ibm.com, AIX, BigInsights, Bluemix, Cognos, Stage, DB2, Domino, DS4000, DS8000, Easy Tier, FileNet, FlashCopy, Global Technology Services, GPFS, IBM FlashSystem, IBM MicroLatency, IBM Pure, IBM SmartCloud, ILOG, InfoSphere, Lotus, Netezza, Power, POWER, POWER7, POWER8, Power Systems, PowerLinux, PowerVM, ProtecTIER, PureApplication, PureFlex, PureSystems, SoftLayer, SPSS, Storwize, System i, System Storage, Tivoli, WebSphere, XIV, z/os, z/vm and z Systems are trademarks of International Business Machines Corp., registered in many jurisdictions worldwide. Other product and service names might be trademarks of IBM or other companies. A current list of IBM trademarks is available on the web at Copyright and trademark information at Intel is a registered trademark of Intel Corporation or its subsidiaries in the United States and other countries. Linux is a registered trademark of Linus Torvalds in the United States, other countries, or both. Microsoft and Windows are trademarks of Microsoft Corporation in the United States, other countries, or both. UNIX is a registered trademark of The Open Group in the United States and other countries. This document is current as of the initial date of publication and may be changed by IBM at any time. Not all offerings are available in every country in which IBM operates. THE INFORMATION IN THIS DOCUMENT IS PROVIDED AS IS WITHOUT ANY WARRANTY, EXPRESS OR IMPLIED, INCLUDING WITHOUT ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND ANY WARRANTY OR CONDITION OF NON-INFRINGEMENT. IBM products are warranted according to the terms and conditions of the agreements under which they are provided. Actual available storage capacity may be reported for both uncompressed and compressed data and will vary and may be less than stated