Predict Behavior in the Scope of a
Transaction
New York City z Systems Technical Summit
Chris Spaight, z Systems Marketing Manager
March 22
Agenda
 Real time analytics: what is it?
 Predictive analytics: why it is important
 The z Systems solution: Zementis
 Why z Systems: architecture advantages
 Use cases: Customer Analytics, Risk Management
© 2015 IBM Corporation
2
“Real-time” is a very subjective term: understand your use case carefully!
• Guide overall direction of the
enterprise
• Should we expand overseas?
• Manage and control operations
• What product should we promote
today?
Tactical
Decisions
Operational
Decisions
Feedback Loop
Real-time
predictive
analytics
Strategic/Tactical Insight
“Right-time” objective: reduce time
from days/hours to hours/minutes
Decision Flow
Right-time analytics
Strategic
Decisions
• Handle every customer interaction
• Is this payment request valid?
Predictive Insight
“Real-time” objective: reduce time from
seconds to milliseconds – or less!
© 2015 IBM Corporation
3
z Systems: vision, strategy and technology to fuse transactions and analytics
Streamlined decisioning process
Act
Faster, more accurate
Analyze
Transact
Report
Transform
modeling
Faster, more accurate
reporting
Integrating predictive scoring with transaction systems
Faster, more accurate scores
Faster transactions
Faster transformation
of data
Integrate
Easier integration of other data sources
© 2015 IBM Corporation
4
The architecture you want: leverage core capabilities to automate, optimize, govern operational business decisions
Predictive modeling, business rules and orchestration together
enable the most effective decisions
• Advanced analytics with classification, association,
segmentation model types (CHAID, NN, C&R, …)
• Rules to define action based on thresholds
• Orchestration to coordinate all activity
Orchestration
Transaction / Batch Workload
Historical Data
Predictive Analytics
Business
Rules
Act
Transact
Business Analyst




Business policy
Regulation
Best practices
Know-how




Risk
Clustering
Segmentation
Propensity
Analyze
Data Scientist
© 2015 IBM Corporation
5
The core software for integrated z Systems solutions
Predictive Analytics
Zementis for z Systems using IBM SPSS Modeler or other
modeling tools (Python, R, SAS, etc.)
• Delivers better, more profitable decisions, at the point of
customer impact
• Improves accuracy by scoring directly within the
transactional application against the latest committed data
• Delivers the performance needed to meet operations SLAs
• Avoid data governance and security issues, save network
bandwidth, data copying latency, disk storage
• Same high qualities of service as operational systems
• Easier to incorporate scoring into applications
Business Rules
IBM Operational Decision Manager for z/OS
• Automate and manage frequently occurring, repeatable
business decisions
• Codifies business policies, practices and regulations
• Enables changes to be easily made by business people
• Automates decision making with the fidelity of an expert
• Centralized, externalized decisions enable consistency and reuse
• Manage business decisions in a natural language
• Decouple development and decision change lifecycle
© 2015 IBM Corporation
6
Agenda
 Real time analytics: what is it?
 Predictive analytics: why it is important
 The z Systems solution: Zementis
 Why z Systems: architecture advantages
 Use cases: Risk Management, Customer Analytics
© 2015 IBM Corporation
7
What can organizations do with predictive analytics?
A
Universe
of Data
Example 




Increase customer
retention
Reduce fraud/loss with
detect and prevent
Mitigate risk based on
predictions
Drive up-sell and cross
sell
Pattern of consumer behavior over last 5 years of
purchase history: card present at time of sale
Current transaction:
o Over $1000, card not present
o Merchant falls into category typical for this consumer’s
spend
Use predictive analytics to determine likelihood of
fraudulent transaction.
Develop insight to questions that are not binary ‘yes’ or ‘no’  likelihood of a pattern
Predictive scores provide complementary insight to rules-based systems
© 2015 IBM Corporation
8
In-transaction, synchronous predictive analytics
• Validate inputs and
channel
• Prepare for
submission to
system of record
Initiate
Transaction
Customer Account
Data

• Continue
processing
• Stop Transaction
• Initiate alternative
action
Determine Risk
and Disposition
• Invoke “existing”
risk scoring
mechanisms
• Get Predictive Score
for Transaction Risk
• Determine actions
Transaction Data
Transaction
History
Determine
Transaction
Disposition
Merchant Data
In-transaction predictive analytics can address issues such as data latency, timely execution, tight SLAs,
governance of sensitive data, access to both transaction and external data, skills gaps.
© 2015 IBM Corporation
9
Agenda
 Real time analytics: what is it?
 Predictive analytics: why it is important
 The z Systems solution: Zementis
 Why z Systems: architecture advantages
 Use cases: Risk Management, Customer Analytics
© 2015 IBM Corporation
10
In-line predictive analytics on z Systems mainframes
What is it?
An integrated predictive
analytics deployment
and scoring capability
for organizations
managing data and
transactions with IBM z
Systems
Why is this important?
• Integrates predictive analytics with the
data, the business process and the
underlying hardware
• Allows organizations to execute predictive
models at the point of greatest efficiency
with respect to transaction performance
or cost efficiency
• Closes the distance between the data, the
computation and the actions that define
the business process
• Net result: creates an optimal IT
architecture to support agile business
operations
© 2015 IBM Corporation
11
Solution snapshot
•
What it Is
What It Does
An integrated predictive
analytics deployment and
scoring capability for
organizations managing data
and transactions with IBM z
Systems
 Scores models developed using
IBM data mining solutions, as
well as third-party statistical
platforms
 Dramatically reduces time-toinsight for critical business
decisions that benefit from
predictive analytics
 Delivers certified execution
engines for applications
compatible with the Java
environment for z/OS (e.g. CICS,
IMS, WebSphere)
 Maximizes the value of existing
IT infrastructure to drive capital
efficiency
•
The joint solution: “Zementis
for IBM z Systems”
•
Zementis product name: “UPPI
for Java on z/OS”
 Fuses predictive decisions and
transaction processing into a
scalable, high-performance,
massively parallel platform that
easily processes data at
petascale volumes
How it Helps
 Enhances governance and
security, not just of the data,
but also of the predictive
analytics
© 2015 IBM Corporation
12
Analytics at scale – Zementis for IBM z Systems
What is PMML
DELIVERS:
•
Compatibility with IBM z
Systems and other IBM
technology solutions and
platforms
•
Enterprise-grade
performance and stability
•
Extreme scalability to
support dynamic analytical
requirements
•
Data management
functionality from Apache
Spark


Uses the Predictive Model Markup Language (PMML)
to import and deploy predictive models
XML-based language
o Used to define statistical and data
mining models and to share these
between compliant applications
o Incorporates data handling and
transformations Clearly separates
model development from model
deployment
o Eliminates the need for custom
code
Mature industry standard
o Developed by the DMG (Data
Mining Group)
o Supported by most leading data
mining tools
© 2015 IBM Corporation
13
Integration via a Java API enhances agility
Java Program
WebSphere App
IMS
Java API
CICS / COBOL
Zementis
Scoring
•
Input to Zementis scoring engine is via Java APIs
•
Model determines the input fields required,
application feeds input fields and invokes
Zementis’ Java API
•
Application integration will depend on:
•
•
•
•
•
Application environment
Programming style of interaction
Organizational guidelines
Other technical factors
Available example & samples for invoking from
CICS / COBOL
© 2015 IBM Corporation
14
Zementis with the z Systems Data Lifecycle Ecosystem
Streamlined decisioning process
Faster, more accurate
Faster, more accurate scores
SPSS
Act
modeling
Faster transactions
SPSS
Analyze
Transact
CICS, IMS
WebSphere, SAP, …
• Generate insights quickly: enable rapid
• Build efficiently: write model
time-to-insight from predictive analytics,
integrated directly into business processes,
using native Spark
once in SPSS, R, Python or SAS
• Deploy fast: eliminate manual
• Drive economic value: scalable, OpEx- &
process & errors
CapEx-efficient, high ROI
Faster, more accurate
Report
Transform
reporting
Faster transformation
of data
Integrate
Easier integration of other data sources
© 2015 IBM Corporation
15
“Transact”:Faster,
Faster,more
moreaccurate
accuratemodeling
scores
“Analyze”:
Faster, more accurate modeling
SPSS
Faster, more accurate modeling
SPSS
• Build efficiently: write
model once in SPSS, R,
Python or SAS
• Deploy fast: eliminate
manual process & errors



Faster transactions
CICS, IMS
WebSphere, SAP, …
• Generate insights quickly: enable
rapid time-to-insight from
predictive analytics, integrated
directly into business processes,
using native Spark
• Drive economic value: scalable,
OpEx- & CapEx-efficient, high ROI
Unique approach
Utilizes
the Predictive
to driving
Modeltimely,
Markup
accurate
Language
business
(PMML)
decisions
+ proprietary
based on
Zementis
predictive
technology
analytics
Streamlines
The
approach?
andIn-line,
accelerates
integrated
predictive
analytics
model
functionality
development and deployment
Compatible
Z4z
directly embeds
with IBMpredictive
SPSS dataanalytics
mining tools
functionality
+ many other
withinIBM
thetechnologies
core activities
and
that
platforms
define the business process
© 2015 IBM Corporation
16
What about SPSS Modeler?
What it Is
•
•
•
Predictive analytics platform
Designed to bring predictive
intelligence to decisions made
by individuals, groups, systems
and the enterprise
Provides a range of advanced
algorithms and techniques
• Text analytics
• Entity analytics
• Decision management
• Decision optimization
How it Helps
Faster, more accurate
modeling
Faster
transactions
Analyze
Transact
The IBM z Systems Data Lifecycle
Ecosystem supports both
Zementis and
SPSS Modeler
• More options for data scientists
to:
• Develop models
• Deploy models
• Operate models directly
within business processes at
scale
• Analyze
• Build efficiently
• Deploy fast
• Transact
• Generate insights quickly
• Drive economic value
© 2015 IBM Corporation
17
Agenda
 Real time analytics: what is it?
 Predictive analytics: why it is important
 The z Systems solution: Zementis
 Why z Systems: architecture advantages
 Use cases: Risk Management, Customer Analytics
© 2015 IBM Corporation
18
…but, outdated views of infrastructure can impede progress
Significant complexity
Separated data warehouses
Analytics latency
Transactional data is not readily available
Source Data
Lack of synchronization
Data is not easily aggregated and fresh
Data duplication
Multiple copies of the same data
Predictive Scoring
Excessive costs
Of moving data around
Predictive Modeling
© 2015 IBM Corporation
19
The architecture you want: leverage core capabilities to automate, optimize, govern operational business
decisions
Process Orchestration
Operational Systems of Record
Move the data to the analytics
Predictive Models
Network
Customer
Accounts
Payments
Purchases
Card
Claims
etc.
Network
Data
Accessed
Better approach?
Operational Systems of Insight
Data Moved
Network
Business Rules
Data
Accessed
Considerations






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














Customer
Accounts
Payments
Purchases
Card
Claims
etc.

Process
Orchestration
Predictive Models

© 2015 IBM Corporation
20
Sample performance findings: full architecture
Orchestration
3 Predictive SPSS Models
Demographic
Anomaly
(95 inputs)
Session Anomaly
(31 inputs)
Fraud Propensity
10%
zHubAdapter
8%
35%
9%
(157
inputs)
•
•
16%
22%
FraudRule
Business Rule
z/OS Transaction Environment
doTransfer
•
Three advanced SPSS
models executed during
each transaction with many
inputs
Performance measures
show favorable results (26.5
msec CPU time, end to end)
More optimizations possible:
z196 used, interface
optimizations, etc.
Sample Transaction: CICS
© 2015 IBM Corporation
21
Sample performance findings: real-time scoring alone
Add Real-Time Scoring to Transaction
1.0 millisecond difference
Response Time (msec)
Sample Workload Specifics (Lab Measurements):
• IBM z13
• Transactions > 320,000 in 5 min.
• Predictive Model: Logistic Regression
• Inputs to Model : 12
0.06 millisecond difference
CPU msec/tran
0
5
Without Scoring (Baseline Transaction)
Some models may require data preparations to
generate model inputs, these will vary, one
example follows
10
15
Added Real-Time Scoring
** Internal testing shows scoring to be a fairly fixed consistent cost, so relative impact will improve with transactions
heavier than 10ms baseline
Add Data Preparations to Transaction
Minimal SLA and
CPU impact!
Response Time (msec)
Data Preparations, 12 fields total:
• 1 required no data preparation
• 5 required database access
• 6 required calculations
Significant opportunity to optimize data
preparation using pre-aggregated data and IBM
DB2 Analytics Accelerator
CPU msec/tran
0
2
4
6
Without Scoring or Data Prep (Baseline Transaction)
8
10
12
Added Data Preparation
© 2015 IBM Corporation
22
Agenda
 Real time analytics: what is it?
 Predictive analytics: why it is important
 The z Systems solution: Zementis
 Why z Systems: architecture advantages
 Use cases: Risk Management, Customer Analytics,
© 2015 IBM Corporation
23
Use Case 1:
Countering payment fraud, waste, abuse and financial crimes
(banking example)
© 2015 IBM Corporation
24
Business view - bank: incrementally enhance existing fraud detect with predictive approach
Fraud rule engine
Transaction
 Determine fraud
risk
 Continue
 Pend/investigate,
etc.
Banking
Transaction
Initiated
Business Goal: reduce loss due to card fraud, reduce card
 In place today
 Extend with additional insight
from predictive analysis
Transaction
Data
scores
Historical
Transaction
evaluat
Data
es
Results of
large,
complex
queries
Model
Historical
IBM DB2
Analytics
Accelerato
r
Account
Merchant
Account Data
deactivation, grow revenue associated with card purchases,
Data
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
© 2015 IBM Corporation
25
Architecture view - bank: incrementally enhance existing fraud detect with predictive approach
z Systems
Benefits
Fraud detection rates improved through
integrated infrastructure that delivers new
capabilities without sacrificing SLAs or
compromising security
New detect function
Existing banking
z/OS
Core
systems of
record:
ODM
Decision
Mobile Device:
Initiate
Payment
Transfer Funds
Check Balance
• Internet
IBM
SPSS
Banking
Modeler
scoring
Adapter for DB2
• Etc.
• Flexibility to invoke processes real-time
or scheduled, tight or loose coupled to
invoking application
• Prioritize detect operations based on
inputs (tran, LOB, amounts, etc)
• Leverage same infrastructure across
multiple functions with both
standardized processes & variations for
specific LOBs as needed
• Can provide as a service across
multiple clients, differentiate through
context
IBM Operational
Decision Manager
Server
• Payments
Orchestration
Fraud check request
WMB
MDM DB2
CDH DB2
Response
DB2 z/OS
z/OS
Zementis
Scoring
Adapter
Acct
Data
Cust
Data
Fraud Detection
Workflows
Trx
Data
Models
z/OS
DB2 Analytics
Accelerator
© 2015 IBM Corporation
26
Use Case 2:
Predictive customer intelligence / executing the next best action
(Banking up-sell example)
© 2015 IBM Corporation
27
Business view - bank: optimized next best action for customer-initiated loan
Customer call: need a new business car loan
Advise: eligible for new car loan up to $35,000 with $1,000
down on vehicle
Short tenure small
business LOC
requesting car loan
Eligible to leverage current LOC for car loan with improved
overall rate and LOC credit increase. If LOC used for loan,
then down payment not required
Offer decision
considerations
Offer accepted by customer
Capturing
Customer
Activity
•
•
•
•
Master Information & Systems of
Record Transactions
Real-Time Analytics &
Decisions
History
Transactions
Customer
Model
Loan Payments
Banking clients want to integrate information across all product lines in order to make real-time, targeted decisions
Why real-time? Bank may risk losing customer business or loyalty for other products
Need to incorporate high value, predictive advanced analytics as part of transactional systems
Why do we need all the data to score? Reduce bank’s risk in approving loan
© 2015 IBM Corporation
28
Zementis Screenshots
© 2015 IBM Corporation
29
Zementis Screenshots
© 2015 IBM Corporation
30
Thank You
© 2015 IBM Corporation
31