You are on page 1of 40

Expert Informati on Sessi on

powered by Knowledge on Demand


Evolution of SAP Analytics with
SAP InfiniteInsight
A Partner Community Event
Mike Watschke, Predictive COE
September 9, 2014
2014 SAP AG. All rights reserved. 2
SAP InfiniteInsight short overview
Whats new with SAP InfiniteInsight
Predictive Analytics Roadmap
SAP InfiniteInsight & HANA Integration
Partner Pricing
Tips and tricks for implementation
Agenda
2014 SAP AG. All rights reserved. 3
What is Predictive Analytics from SAP?
SAP Predictive Analytics is:
A comprehensive, leading suite of products from SAP.
An advanced area of analytics designed to provide more intelligent, more
forward-looking (anticipatory), and more actionable analysis.
SAP Predictive Analytics is designed to:
Help companies make better decisions.
Derive better insight and optimize customer relationships.
Make more meaningful and proactive sense of Big Data.
Answer difficult questions that BI is not equipped for.
SAP Predictive Analytics is a suite of
products sold only as a bundled solution!
2014 SAP AG. All rights reserved. 4
Technical
Script-based
Technical
In-memory
Script-based
Semi-Technical
Modeling/Viz
UI-based
Functional
Automated Modeling
UI-based
What could happen?
Whats the best that
could happen?
What happened?
When did it
happen?
What is Predictive Analytics from SAP?
2014 SAP AG. All rights reserved. 5
Business Benefits and Impact
Typical KPIs addressed
Churn rate
Marketing campaign response rate
Click through and conversion rate
Failure rate (maintenance)
Customer satisfaction
Number of website page views
Number of new customers
ARPU (Average Revenue Per User)
Fraud rate
How SAP Predictive Analytics
will solve them
Build predictive models for
every customer interaction
Segment customers
Recommend relevant products
Prevent fraud
Detect early signals
Salvage customers
Increase i n sales
Increase in
conversions
40x
Increase in campaign
conversion rates
260%
8.5%
2014 SAP AG. All rights reserved. 6
SAP Infini teInsight
Provide Business analysts and
Data scientists with a fully
automated process
Data preparation
Create 1000s of derived attributes
Define metadata once
Builds analytic dataset automatically
Predictive modeling / Data mining
Regression / Classification
Segmentation
Forecasting
Association rules
Social Network Analysis
Advanced model deployment and
management
2014 SAP AG. All rights reserved. 7
SAP Predictive Analysis (includes Lumira
codeline)
Provide Data Scientist and
Business Analysts with
sophisticated algorithms to take
the next step in understanding
their business and modeling
outcomes
Perform statistical analysis on your data
to understand trends and detect
outliers in your business
Build models and apply to scenarios to
forecast potential future outcomes
Breadth of connectivity to access almost
any data
Optimized for SAP HANA to support huge
data volumes and in-memory
processing
2014 SAP AG. All rights reserved. 8
Beating the Competiti on
Traditional Predictive Solutions (such as SAS and
IBM)
Time-intensive,
manual
processes.
Linear
relationship
between model
accuracy and
analyst time
Hard to maintain,
share and port
model code or
workflow
diagrams
Stati stical
library where
the right
algorithm must
be selected,
tested and fine-
tuned
Hard Isolated
Slow
Statistici ans pre-
select vari ables,
hence excluding
information
Manual
1
2
3 4
5
2014 SAP AG. All rights reserved. 9
Positi oning against the Competi tion
SAP Predictive Analytics
Reduce decision
latency wi th real-
time insight
Apply actions to
information,
processes, and
apps
Bring Advanced
Analytics to a
broad spectrum
of users
Easy Actionable
Fast
Sel f-service
application of the
right model with
the best fit
Automated
2014 SAP AG. All rights reserved. 10
SAP InfiniteInsight short overview
Whats new with SAP InfiniteInsight
User Interface Enhancement
New Database Support
Security Enhancement
Engi ne Enhancement
Composi te Variables
Geo-referenced Data Integration
more
Agenda
2014 SAP AG. All rights reserved. 11
User Interface Enhancement
Compliance
Crystal Blue User Interface Theme
Accessibility
Keyboard navigation
Mnemonics
Navigation shortcuts
Font Size, High Contrast Mode, report style sheet.
Usability
Ribbon in several panels
Data Manipulation User Interface Enhancement
Merge Panel Redesign
Domain Edition Redesign
Graph Visualization User Interface Enhancement
communities' temporal evolution visualization
Temporal graph duplication
Community browsing enhancement
2014 SAP AG. All rights reserved. 12
New Database Support
Big Data: Hadoop/Hive 11 & 12 support
Hadoop Hive 12 full support
Hadoop Hive 11 limited support due to hive limitations
Connectivity through Data Direct ODBC driver
Explorer/Advanced Data Manipulation (ADM)
Scorer
Greenplum 4.2 support
Connectivity through ODBC driver
Explorer/ADM
Scorer
2014 SAP AG. All rights reserved. 13
Security Enhancement
InfiniteInsight 7.0 now uses Single Sign On (SSO)
Authentication on Linux (only)
The Linux Server must rely on Kerberos for
Authentication.
SSO Client Layers have been deployed on
Windows
Linux
J ava Web Start
Added a new argument in the remote Corba Api
Authentication token is passed to all api calls.
2014 SAP AG. All rights reserved. 14
Engine Enhancement
Ordinal Binning
Ordinal variables statistics representation has changed.
Now Ordinal variables are cut into bins.
The maximum number of bins of a given ordinal variable is 20 (default value).
No Ki/Kr regression compared to previous version representation.
Purpose
Empty bins are now represented on the training dataset in order to handle them when finding unseen
values when scoring.
Non robust ordinal categories are now grouped in single segments.
Debriefing of an ordinal variable is easier
Speed and memory usage enhancement
Ordinal variables with large dictionary are now limited to 20 bins (default value).
It limits the memory usage and speed up the compression process.
Representation of an ordinal variable in previous versions Representation of an ordinal variable in 7.0
2014 SAP AG. All rights reserved. 15
Engine Enhancement
Regression Improvements Input Encoding
For continuous target (regressions) the input variables encoding are no more
based on the target segment frequency (normal profit).
Input variables encoding is now based on the target mean.
Ki/Kr and profit curves are computed based on target mean profit.
Categories and groups with higher target means are more profitable.
Compression groups together categories with similar target means.
Purpose
This target mean based encoding is more natural and consistent for regressions than the
previously used normal profit.
Normal Profit encoding
for a variable in previous versions
Target Mean encoding
for the same variable in 7.0
2014 SAP AG. All rights reserved. 16
Engine Enhancement
Regression Improvements Model Selection
For continuous target (regressions) the model selection is no more based on
the L2.
The model selection is based on the Ki/Kr.
Only possible because now all inputs are encoded based on the target mean (previous slide).
Now the modeling engine (classification and regression) is fully non-parametric.
Significant improvements in modeling quality compared to previous release (based on our
benchmarks)
The 7.0 release encoding is
more consistent for the
modeler estimator.
In the 7.0 release, the Ki/Kr of
the estimator is better than all
individual Ki/Kr of each inputs
which was not the case on
previous releases.
Previ ous Releases
7.0 Release
2014 SAP AG. All rights reserved. 17
Composite Variables
Composite Variables does offer the possibility to
cross existing variables in order to create new
ones.
Pair variables cross.
Multiple variables cross.
Geo-location variables cross (detailed after).
The created crossed variables are named:
CompoVarName_VarName1_VarName2.
o CompoVarName is the prefix name of the composite variable.
o VarName1 is the first variable name.
o VarName2 is the second variable name.
2014 SAP AG. All rights reserved. 18
Geo-Referenced Data Integration
Location-Aware
Natively identifies the location of a person or transaction using latitude and
longitude and translates that data into a new variable type, called geographic
tiles or regions.
Support of a new Storage type
Angle.
Need to use composite variables
to create position variable.
Automatically computes tiles
with additional information
Tile coordinates, Weighted count,
Tile Frequency, Density.
Several export format support
KML, GML, Shape_File
Associated color and opacity can
be set to Frequency, Positive
Target (only on supervised
modelling) and Density.
2014 SAP AG. All rights reserved. 19
Geo-Referenced Data Integration
Co-Location
Detects transactions or events occurring in the same location during the
same period of time.
Proximity Graph Type in Social Module dedicated to Geo-Referenced Data.
o Target Node is a composite position variable
o A size of a tile is mandatory to define the granularity of the processing.
o Needs unique identifiers.
o Can be supervised or not.
Synchronous policy
o This policy will control
how the neighbors of a
neighbor are considered.
o For Synchronous policy,
a neighbor of a neighbor
is only included in the
graph for each dates
couple that verify the
maximum time delta
(under the defined
threshold).
2014 SAP AG. All rights reserved. 20
Geo-Referenced Data Integration
Path Identification
Creates a sequence of geo-localized positions from time-based transactions
to extract patterns of events.
Sequence policy
A sequence of N tiles is
considered if this sequence was
detected M times (no matter the
number of user).
The next transition limit is
the maximum number of next
events that will be considered.
2014 SAP AG. All rights reserved. 21
SAP InfiniteInsight short overview
Whats new with SAP InfiniteInsight
Predi ctive Analytics Roadmap
SAP InfiniteInsight & HANA Integration
Partner Pricing
Tips and tricks for implementation
Agenda
2014 SAP AG. All rights reserved. 22
Planned Innovations Future Direction Today
SAP Infi niteInsight:
Classification, Regression, Clustering
algorithms available in PA
SAP HANA:
Support of DBSCAN Algorithm. (
Density based Clustering )
Support of SVM Algorithm
Support of Silhouette Calculation for
Cluster Results (e.g. the quality of the
cluster )
Improvement of univariate Statistics
information with HANA Online.
Automatic Decimal to double
conversion.
HANA Partition Component
Export model based on HANA PAL as
stored procedure to HANA DB
Visualization Improvements
Improved confusion matrix for
classification algorithms
This is the current state of planning and may be changed by SAP at any time.
(Release 1.17.2)
SAP Predictive Analysis
Product road map overview - key themes and capabilities
Consumption of Models in Lumira
Desktop and Lumira Cloud
Improvements for handling large/Big
Datasets, without requiring
partitioning
BW connectivity & modeling online
and offline
OEM/SDK support for Partners and
customers
Improve HANA PAL support
- Custom online component to call
any HANA PAL from desktop PA
-Charting improvements
General improvements with R,
installation process, bug fixes
Installer improvements
Unified UI for Predictive
Full integration of SAP Predictive
Analysis and SAPII suite
Customization
Run HANA SQL script
Pervasive casual user workflows
Easy model consumption and
maintenance
Simplified predictive workflows
Application consumption
API SDK
Exporting models as text rules &
stub code
Advanced predictive capabilities
Ensemble modeling
Model and algorithm comparison
Auto node and auto modeling
Large data volume visualization
2014 SAP AG. All rights reserved. 23
SAP InfiniteInsight short overview
Whats new with SAP InfiniteInsight
Predictive Analytics Roadmap
SAP InfiniteInsight & HANA Integration
Partner Pricing
Tips and tricks for implementation
Agenda
2014 SAP AG. All rights reserved. 24
In-Memory Predictive Analytics with SAP HANA
SAP HANA
In-memory In-database Predictive Analysis
1
Predictive Analysis Libraries (PAL)
Accelerate predictive analysis and scoring
with native in-database algorithms delivered
out-of-the-box.
Graphical Modeli ng
HANA Studio for app developers. Pre-built commonly
utilized business & predictive algorithms to facilitate a
faster and easier development.
2
R Integrat ion
Execute R scripts via high performing
parallelized connection. Embed R scripts as
part of overall query plan.
SAP Lumira
SAP Predictive
Analysis
SAP
InfiniteInsight
Partner Tools
& Solutions
In-Memory Processing Engine
SQL Engi ne Text Engine
Calculation Engine
PAL
1
R-Scripts
2
Association
Analysis
Cluster Analysis
Classification
Analysis
Time Series
Analysis
Outlier Detection
Link Prediction
Data Preparation
R-Engine
HANA Studi o
Modeler
2014 SAP AG. All rights reserved. 25
Enterprise
Data
CRM
Teradata/Other
Hadoop
F
l
e
x
i
b
l
e
C
o
n
s
u
m
p
t
i
o
n
E
x
t
r
a
c
t
,
T
r
a
n
s
f
o
r
m
,
L
o
a
d
a
n
d
S
t
a
g
e
ERP
Cloud
Social
How i t works Data to Decisions
Agile
Analytics
SAP HANA
SAP BI
HTML 5
Mobile
Apps
API
ByOT
Integration with HANA allows real-
time, high performance predictive
processes and analytics
Automated, results-driven
predictive modeling for data
scientists and business users
alike
Publish results to BI, embed within
apps, generate actionable scores
within DB, more


Leverage any data source
enterprise, departmental, cloud,
3
rd
party, unstructured, more

Infi niteInsight
Predictive
Solution
Any Data
Source
Direct Direct
2014 SAP AG. All rights reserved. 26
SAP InfiniteInsight short overview
Whats new with SAP InfiniteInsight
Predictive Analytics Roadmap
SAP InfiniteInsight & HANA Integration
Partner Pri cing
Tips and tricks for implementation
Agenda
2014 SAP AG. All rights reserved. 27
SAP InfiniteInsight Edge, Modeling Edition
Standard Regional uplifts apply
Standard Partner Edge Discounts apply
Upgradeable to Full Use InfiniteInsight Modeler
Can be sold by direct sales and partners
Restrictions
Maximum of 2 users
Cannot be used with any other predictive components (engine,
social, recommendation)
20K EUR / NUL
SKU: 7017140
SAP Inf. Insight Modeler
(1 NUL)
2014 SAP AG. All rights reserved. 28
SAP InfiniteInsight Edge, Analytics Edition
Standard Regional uplifts apply
Standard Partner Edge Discounts apply
Upgradeable to Full Use InfiniteInsight as long as all the included
options are also purchased
Can be sold by direct sales and partners
Restrictions
Single package purchase
Maximum of 500GB data
Maximum of 8 cores for Sybase IQ
Maximum of 4 cores for Data Integrator
Predictive capabilities cannot be used for data outside of IQ
Can be only used with Edge BI analytics edition
SAP IQ
(8 cores)
100K EUR
SKU: 7017141
Edge Data Integrator
(4 cores)
SAP Inf. Insight Engine
(8 cores)
SAP Inf. Insight Modeler
(2 NUL)
2014 SAP AG. All rights reserved. 29
SAP InfiniteInsight short overview
Whats new with SAP InfiniteInsight
Predictive Analytics Roadmap
SAP InfiniteInsight & HANA Integration
Partner Pricing
Tips and tricks for implementation
Agenda
2014 SAP AG. All rights reserved. 30
Predictive Process Discovery to Deployment
Typically 45 60%
of Project Effort /
Time
Data Loading,
Volume,
Complexity, and
Processing
Performance
Limits:
Load
Exploration
Preparation
New and
Changing Data
Sets
Batch vs. Near-
and Real-time
Data
Typically 15 - 30%
of Project Effort /
Time
Data Volume,
Complexity and
Speed of
Processing Limits:
Model Selection
Test / Validation
Data Volume,
Complexity and
Speed of Processing
Limits:
App Performance
Appropriate Tools
For Users
Typically 15 - 25% of
Project Effort / Time
Move Model Into
Production
One-time? / On-
going?
2014 SAP AG. All rights reserved. 31
Core Predictive (and related) Components
SAP Data
Services
SAP HANA
SAP
InfiniteInsight
CRM/ERP, 3
rd
Party Apps
SAP
BusinessObjects
Files
Databases
Cloud or on-premise
R integration
PAL integration
Big Data and departmental
ready
Use Explorer to create analytical
record even if HANA is used
Zero in on a specific use case
that shows business value and
improvement
Web & Others
Write scores and
logic back to source
or CRM/ERP/custom
Prove uptick from
historical to future if
solution deployed
Publish predictive
results and viz
analysis to
Lumira/PA.mobile,
etc.
2014 SAP AG. All rights reserved. 32
Your Go-To Partner
site supporting
Predictive Analytics.
https://partneredge.sap.com/en/products/infi/about.html
For more i nformati on
Thank You!
Addendum
2014 SAP AG. All rights reserved. 35
SAP Predi ctive Analytics helps compani es make better
decisions, derive better i nsight and optimize customer
relationships by deploying predictive analytics on the
mountai ns of data compani es have in their data warehouse
and CRM systems.
Our end-to-end sol utions unl eashes the power of bi g data to
deliver massive improvements in speed and agility. We
optimize every step in the customer l ifecycle includi ng
acquisition, cross-sel l, up-sel l, retenti on and next best
activity as well as i n fraud, risk and operations.
Elevator Pitch
2014 SAP AG. All rights reserved. 36
SAP ERP/CRM/Cloud opportunities & installed base
Move from BI to Advanced Analytics: Existing BusinessObjects
and Lumira customers
HANA: sell with HANA or up-sell on top of it since HANA is a
predictive facilitator and run-time engine for in-memory, fast
predictions on Big Data
Coexist with SAS/IBM with customers already doing Predictive
Top Selli ng Scenari os
2014 SAP AG. All rights reserved. 37
Predictive Analytics brings business value, so start the use case
with business, not IT
Shift from technology to solution selling an Advanced Analytics
solution from SAP entails HANA, Predictive Analytics, BI, and
possibly analytics applications, Rapid Deployment solution, and
related
Need for more data, hence Predictive Analytics sells more HANA
Helps sell more apps and industry solutions
Expanding the Footpri nt
2014 SAP AG. All rights reserved. 38
Handli ng Objecti ons
Is the quality of the models as good as it was developed using SAS?
Yes. SAP InfiniteInsight aims at answering business issues, and builds accurate models
quickly. A number of model iterations are produced, each with varying degrees of
predictive power and confidence, and customers can choose the one that meets their
requirements.
SAP InfiniteInsight lacks a wi de choice of algorithms i n its statistical library.
The approach is fundamentally different: InfiniteInsight is not a statistical workbench but
a solution who makes predictive analytics usable by business analysts, and not only data
scientists.
SAP has two predi ctive tools Why is that?
SAP acquired a leading predictive solution in 2013. There is a public roadmap where we
have committed to delivering the best of the features and usability between the two tools in
one suite going forward.
2014 SAP AG. All rights reserved. 39
WHAT CUSTOMERS CARE ABOUT
Line of Business
Head of CRM
VP of Sales
Business Analyst,
IT Operations
Customer Service
Head of BI
Data Scientist
Pain Poi nts & Business Drivers
The business is demanding solutions to answer
questions like Why and Whats likely to happen next
Tools on the market today are designed for data
scientists, not analysts nor business users
There is a skills shortage of those can develop models
Inability to provide proactive and actionable predictive
results to business users
Volume and Velocity of the data growing faster than our
ability to drive insights
Can you proactively detect and address customers
who are at risk of leaving?
Are you able to identify and retain your high-value
customers with the right retention offers?
Can you increase your cross-sell and up-sell
effectiveness through cross-channel coordination?
Do you currently use any predictive analytics tools
like SAP Predictive Analysis, SAS, SPSS, R, or
other predictive software solutions?
Where would you consider yourself/your
organization on a PA Maturity scale (1-10)?
Discovery Questions
Executives
CMO
CIO
CFO
Qualifying Opportuniti es
2014 SAP AG. All rights reserved. 40

The model building


process, including data
prep and analysis, now
takes 1 to 2 weeks
100% accuracy
that a signal is
positive at 97%
confidence
The user doesnt have to
really worry about which
algorithm is the best.
Data Scientist Data Analyst Business User
700 models
Built for churn
and X-sell

Viliah Overwater, Snr.


Modeling Analyst,
Vodafone NL
82% accuracy in
sales forecasts, a
10% improvement
"SAP InfiniteInsight has
given us a scalable
approach to create
accurate forecasts
across our business.
Elena Zhukova, Head of
Analytics, Eldorado LLC
Gagandeep Bawa, Manager,
North America FP&A
at eBay Inc.
Predictive Analytics Delivers High Return

You might also like