You are on page 1of 47

hybris Marketing

Predictive Model Management


May 2015

Objectives

1.

2.

3.

Explain Predictive
Model Management
(PMM) as part of
hybris Marketing

Explain the Predictive


Scenario Buying
Propensity for
Business Users in ADT.

Explain the different


Predictive
Scenarios, i.e.
predictive scoring;
heuristic scoring;
integration of PAL;
integration of SAP
Infinite Insight;
integration of R;

Customer

Agenda
Consumption

of Predictive Models

Buying Propensity in ADT


Scores in Contact Factsheet

Overview

of Predictive Model Management


Features of Predictive Model Management in Detail
Heuristic Scoring
Integration of SAP Infinite Insight
Integration of R package (nu-SVM)
Predictive Success Measurement
Predictive Scenarios and Shipped Examples
CRM Business Partner: Buying Propensity; Loyalty Score (HRF-rule-based)
Demo Banking
Demo Insurance
Customer

Overview

Consumption of Predictive Models

Management of Predictive Models

hybris Marketing

Customer

Consumption of Predictive Models in ADT

Predictive KPIs choose a


KPI to calculate the scores for
the members in the selected
segment.

Target Objects (e.g. Products)


choose an object for which the
prediction shall be calculated

Customer

Consumption of Predictive Models in ADT

Gain chart shows the gain


achieved by selecting only a
share of the top-scoring
members.

Thresholds to select the


share of top scoring members
of the segment. Or use slider
of x-axis.

Customer

Consumption of Predictive Models in Contact Fact Sheet

Tiles with predictive KPI


each based on a different
predictive model and calculated
on the fly for this single contact.

Personalization of tiles flag


those tiles, which shall be visible
on the screen.

Customer

Predictive Model Management (PMM)


Key Messages

Usability
Simple UI; Overview List; Details on Fact Sheet
Easily identify and publish the best model fit

Flexibility
Any scoring scenario can be configured
Extensibility is well supported for custom use cases and data

Integration
Many statistical modules can be integrated (PAL; R; Infinite Insight)
Runs on HANA Data Views

Performance
High speed performance on big data
Calculation of scores on the fly
Training on integrated Target Groups

Optimization
Optimization of business processes in various places: ADT; Contact
Factsheet; Contact/Sentiment Engagement;
Customer

Overview
Predictive Model Management (PMM)

Usability

Scenarios
Get an overview on or search for
existing predictive models.
Create new predictive models to
calculate scores for specific purpose
based on preconfigured predictive
scenarios

Usability

Usability

Models

Training

Maintain predictive models:

Create and adjust model fits

Create based on preconfigured


predictive scenarios
Change and enhance predictive
models (Model Fits & Scope)
Identify the best fit and publish to
the business scenario

Select predictors or use Infinite


Insight for automatic selection
Assess fitting quality by crossvalidation, Lorenz curve,

Identify the right predictive model and identify the best model fit for
publication to the business process to optimize user decisions.
Customer

Features of PMM

Online-Help

PMM = Predictive Model Management

Features
List of Predictive Models
List of Predictive Models allows you to search and display existing
predictive models.
Filter Icons. Allows
to filter Predictive
Models by status.

Filter Icons.
display only the
models you own or
all models.

Search field. Allows


to search for predictive
models.

Workset Predictive
Models. The user need to
have the PFCG-role
SAP_CEI_PBA assigned to
be able to enter this workset
.

Create button. create


new predictive models.

Result List Area. shows information as the predictive scenario, target object, status and administrative data (name, ID, owner,
trained date). You can delete existing predictive models using the button at the end of a row in the table.

Customer

Features
Details of a Predictive Model (Predictive Scoring)
Navigation area
Details or Notes
can be selected

Header

Area
contains attributes
of the predictive
model and
administrative
information

Applicable Scope
data context for
which the
predictive model is
available

List of Model Fits


contains all
created model fits
with relevant
attributes

Lorenz Curves
show the Lorenz
curve for each
flagged model fit in
the table above.
For heuristic
scores (or other)
the Box-Plots
would be shown.

Button Bar

Customer

Features
Predictive Model: Header Area
Attribute

Explanation

ID

consecutive number automatically given by the system.

Predictive Model

name of the Predictive model as entered by the user. Can be changed later
on.

Predictive Scenario

pre-defined in customizing. It covers the data source (HANA view) and the
allowed implementation methods. This concept will be explained in more detail
later.

Target Object

object for which a prediction shall be calculated, e.g. a specific product or a


product bundle or category. This concept will be explained in more detail later.

Start / End Date

refers to transactional data in the data source. In customizing it can be


maintained, which date-field to consider and if start/end date is relevant for the
predictive scenario.

Status

a predictive model can have four different status: New, In Preparation,


Released and Closed. This concept will be explained in more detail later.

Owner

can be maintained. This is the user who serves as the contact for this
predictive model for questions or feedback from the business users. He is
responsible for the validity and quality of the model and receives further
requirements from the business.

Customer

Features
Predictive Model: Header Area continued

Attribute

Explanation

Training Set

is the set of data, which contain enough and relevant historical information to
train a predictive model, which can be used for prediction. Technically, a
training set is an ADT-generated target group.

Target Variable

is related to the Target Object and is the dependent variable in the predictive
model equation. For scoring it is binary containing 0 and 1, which indicates if
the product was purchased (1) or not (0). Alternatively, could be Interest
shown.

No. of Hits

relates to the target variable, which is dichotomous (i.e. contains only 0 and
1) and counts all positives (i.e. cases where value is 1) in the whole training
set.

No. of Members

is the count of all members in the training set (target group for training).

Description

Is a free text field, which appears as column in the overview list. Some short
notes to characterize the predictive model can be entered here.

Administrative Data

Contain these attributes: Created by; Created (date); Changed by; Changed
(date).

Customer

Predictive Model: Navigation Area

Facet Details
contains the details of a predictive model, which is the Applicable Scope and the list of Model Fits
together with the related visualization (curves showing the distribution of the fitted model execution on
the training set).

Facet Notes
provides a simple text editor to take notes on the predictive
models purpose, configuration, etc.

Customer

Predictive Model: Applicable Scope

Applicable Scope
You can optionally restrict the validity of a predictive model by defining an applicable scope. For
example, you define a set of countries for which the predictive model is valid. Choose the button Add
Scope, select an Attribute, such as COUNTRY, and specify the attribute values. The available
attributes for an applicable scope are defined in Customizing.
This information is processed and checked in the business application and the predictive model is only
offered there, if the validity criteria are met, i.e. the attribute values match the data context in the
application. For example, if in the ADT-segmentation model the selected segment is filtered by country
Canada then the predictive model would not be available, when the applicable scope is reduced to
countries Germany, Austria and Switzerland.
By the bin icon at the end of each row the attribute and is values can be deleted from the applicable
scope.

Customer

Predictive Model: List of Models

List of Model Fits


Gives an overview on all created model fits for this predictive model. The list shows information as name,
description, implementation method, quality coefficient (Gini value), best fit indicator (need to be chosen
manually), published indicator, last trained on (date) and changed by. With the first column a color is assigned to
each fit which is used for the curves below the table. The curve can be hidden, when the color in the list is
unchecked. You can delete existing predictive models using the button at the end of each row in the list.
You can create new predictive models by pressing the button Add Model Fit below the list. Check the status of
the predictive model. You can only create model fits for models having the status In Preparation, or Released. If
necessary, choose In Preparation (at the bottom of the workset) to change the status.
To edit an existing model fit click the name of the fit in the list of Model Fits. As a result, the Model Fit details are
displayed.
Customer

Features
Predictive Model: Lorenz curves

Lorenz curves
Each Lorenz curve in the chart represents one model fit. It displays the relation between the calculated
scores, such as the buying propensity per customer, and the actual hits in the training set, such as the
buying decision. For the Lorenz curve all customers are sorted in a descending order based on the
calculated scores (x-axis). The actual hits are cumulated along this sorted list of customers and displayed
(y-axis) as a percentage of all hits in the training set.
Lorenz curves are only available for models with Target Object/Variable, which are most predictive scores.
Heuristic scores do not have a Target Object and the distributions are compared with Box-Plots.
Customer

Features
Details of a Predictive Model (Heuristic [rule-based] Scoring)

List of Model Fits


contains all
created model fits
with relevant
attributes

Box Plots
showing the value
distribution for
each flagged
model fit in the
table above.

Box-Plots
Each Box-Plot in the chart represents one model fit. It displays the distribution of the score values
generated by the model fit. The box plot shows the range of the values, the Q2 and Q3 quartiles, the
median and the mean. The purpose of this chart type is to compare the distribution of different rulebased score calculations.
Customer

Features
Predictive Model: Box-Plots

Box-Plots
Each Box-Plot in the chart represents one model fit. It displays the distribution of the score values
generated by the model fit. The box plot shows the range of the values, the Q2 and Q3 quartiles, the
median and the mean. The purpose of this chart type is to compare the distribution of different rulebased score calculations.

Customer

Features
Details of a Model Fit (example: Logistic Regression)
Navigation Area

Header Area
contains attributes
of the model fit and
administrative
information

Content Area
layout depends on
the implementation
method, here it is
logistic regression
(PAL)

For logistic
regression (PAL)
the weight (=
regression
coefficient) and the
strength (p-value)
is shown.

Button Bar
Customer

Features
Model fit: Header Area
Attribute

Explanation

Model Fit

name of the model fit as entered by the user. Can be changed later on.

Implementation Method

By creating a model fit a popup appears, where the name, the implementation
method and the description can be maintained. The implementation method can
be chosen from a dropdown list, which is maintainable in customizing for each
predictive scenario.
The Implementation Method can be
a statistical procedure, e.g. logistic regression from PAL, a Decision Tree
from R or a Infinite Insight algorithm, or anything similar
a HANA Rule Framework (HRF) based scores
or a HANA procedure for heuristic scoring.

Quality Coefficient

indicates the goodness of fit in a range from 1 to -1. In the chart is is the area
between the Lorenz curve and the bisecting line.
It is only shown for predictive models with Target Object.

Best Fit

is a check box to indicate the best fitting model, which is maintained manually
by the user on the predictive model screen. Here, in the model fit screen it is in
display mode.

Description

is a free text field, which appears as column in the overview list. Some short
notes to characterize the model fit can be entered here.

Administrative Data

contain these attributes: Created by; Created (date); Changed by; Changed
(date).
Customer

Features
Details of a Model Fit (SAP PAL Logistic Regression)

Cross-Validation
Chose the no. of
subsamples and press
validate. The system
calculates a
corresponding no. Of
validation curves based
on the chosen predictors.

Predictors
List of chosen predictors included in
the model fit with the estimated
weights. Predictors can be removed or
added from a repository.

Data Parameters
extra data parameters can be
entered. For example, revenue
generated by the customers formerly
with a certain product.

Customer

Features
Details of a Model Fit (SAP Infinite Insight binary classification)

Training Results
Statistics on the model fit result from Infinite Insight.

Contributing Predictors
List of automatically detected
predictors by Infinite Insight with
percentage expressing the contribution
strength. Predictors can be removed or
added from a repository.

Customer

Features
Details of a Model Fit (SAP PAL Decision Tree C4.5)

Cross-Validation
Chose the no. of
subsamples and press
validate. The system
calculates a
corresponding no. Of
validation curves based
on the chosen predictors.

Predictors
List of chosen predictors included in
the model fit with the estimated
weights. Predictors can be removed or
added from a repository.

Model Parameters
are influencing the fitting algorithm.
Here, max. tree depth and min. node
size for the split.

Customer

Features
Model Fit (SAP PAL Decision Tree C4.5) Visualization

Decision Tree
Visualization
Expand and collapse
nodes; fit tree to screen;
zoom in tree; thickness
of lines show share of
members in sub-nodes;
blue bars show
percentage of
responders in node.

Customer

Features
Details of a Model Fit (R nu-Support Vector Machine(SVM))
Additional Parameters
showing the number of support vectors.

Cross-Validation
Chose the no. of
subsamples and press
validate. The system
calculates a
corresponding no. Of
validation curves based
on the chosen predictors.

Predictors
List of chosen predictors included in
the model fit with the estimated
weights. Predictors can be removed or
added from a repository.

Model Parameters
are influencing the fitting algorithm.
Here, the nu parameter.

Customer

Features
Details of a Model Fit (Heuristic scoring by SQL procedure)

Data parameters

Histogram
L.

Customer

Features
Details of a Model fit (SAP HANA Rule Framework (HRF) D.-table)

Decision Table
to define conditions (rules) based on a HRF
vocabulary and assign a score value for each of the
condition criteria.

Histogram
shows the distribution of the score
values as defined in the decision table.

Customer

Features
Details of a Model fit (SAP HANA Rule Framework (HRF) Text)

Text Editor
Only one condition expression and one output field
(score) are possible. In the output field also a
formula is possible.

Histogram
shows the distribution of the score
values as defined in the output field.

Customer

Features
Predictive Success Reporting (eligible) Campaigns
Quick filter and search
Quick filter for favorites and target groups based on prediction

List of campaigns
which are based on the same data source as the
predictive scenario and match the applicable
scope of the present predictive model.

Customer

Features
Predictive Success Reporting Success Measurement
Conversion rates
for target and control group and the Lift.

Key figures and metrics


to compare the quality of the model by
varying threshold parameters. Results are
also visualized by gain and lift chart.

Tiles
representing the
underlying campaign,
target & control group
and seg. model.
Customer

Features
Instance Authorization / Marketing Area for Predictive Models

Marketing Area
Predictive Models need to be assigned to a
marketing area. The authorization to see and/or
change a predictive model can now be controlled
on a user role level (user authorization can be
restricted to specific marketing areas).
Customer

Predictive Scenarios

Predictive Scenario
The core instance to create and run Predictive Models
Key Messages

Before a predictive model can be created in the CEI


UI at first the predictive scenario has to be defined in
IMG customizing.

A Predictive Scenario defines the context of the


predictive model, which includes the data source, the
use case, and the applicable algorithms (=
Implementation Method) and others.

Step 1: Define Predictive Scenario with the related


data source (HANA view)

Step 2: Define Predictive Data Source

Step 3: Define Implementation Method, e.g. PAL


algorithm, Infinite Insight, R algorithm, HRF
vocabulary,

More detailed information can be found in the


Customizing for SAP Customer Engagement
Intelligence -> Predictive Scenario

Customer

Predictive Scenario
Define Attributes for Applicable Scope

You can optionally restrict the validity


of a predictive model defining an
applicable scope.

In the IMG for SAP Customer


Engagement Intelligence the
attributes can be defined which are
available and analyzed to ensure that
the Applicable Scope is met.

The attributes defined in IMG are


visible in the Predictive Model based
on this Predictive Scenario and can
be used to restrict the usage of the
model.

Customer

Predictive Scenario
Define Application Anchors and Settings

Applications use the application


anchor to identify the predictive
scenarios that are relevant for the
predictive score.

Customer

Predictive Scenario
Settings for Display of Predictive Scenarios on User

Define the settings for the


appearance of those predictive
scenarios that are to be displayed as
scores in the Account or Contact fact
sheet on the user interface.

Step 1: define the icon, which shall


appear on the tile in the fact sheet

Step 2: define ranks with thresholds,


colors and descriptions

Customer

Shipped Examples

Predictive Examples
CRM Business Partner: Buying Propensity
The ADT-Profile CRM Business Partner: Buying Propensity contains BP master data as
well as some transactional data from presales (Activity; Lead; Opportunity) and sales
(Sales Order). These are based on replicated data from CRM One-Order structures.
Customers, who fill these data tables can use this ADT-Profile for segmentation out of
the box.
With a simple ADT customizing entry the Predictive KPI Buying Propensity can be
made visible in this ADT-Profile. By activating BC-Set
CEI_ADT_CRM_BUYING_PROPENSITY with transaction SCPR3 the ADT-profile is
visible.
After these steps the CEI user with role SAP_CEI_PBA can create predictive models
based on this predictive scenario. The assigned implementation method here is Logistic
Regression based on PAL.

Customer

Predictive Examples
Demo Banking

Demo Banking has two Predictive Use Cases:

Buying Propensity (with SP03)

Cancellation Propensity (with SP04)

The ADT-Profile Demo Banking is available via BC-Set CEI_ADT_CRM_DEMO_BANKING


(transaction SCPR3)

The 7862 customer records can be generated by executing the ABAP-report


CUAN_DEMO_BANKING_GEN (transaction SE38).

The two Predictive Scenarios are also available via BC-Set

CEI_CPRED_DEMO_BANKING for Predictive Scenario Demo Buying Propensity Banking

CEI_CPRED_DEMO_CHURN for Predictive Scenario Demo Cancellation Propensity Banking

With transaction SCPR3 the BC-sets activate the predictive scenarios, i.e. they are made
visible (the visibility flag in IMG customizing is set).

After these steps the CEI user with role SAP_CEI_PBA can create predictive models based on
this predictive scenario. The assigned implementation method here is Logistic Regression
based on PAL as well as SAP Infinite Insight.
Customer

Predictive Examples
Demo CRM Customer Loyalty (SP04)
Use Case: Calculate a Customer Loyalty Score based on rules, which are defined with
Hana Rules Framework (HRF).
The ADT-Profile Demo CRM Customer Loyalty and the Predictive Scenario Demo
CRM Loyalty Score is available via BC-Set CEI_ADT_CRM_DEMO_HRF (transaction
SCPR3)
The demo is based on CRM BP master data as well as some transactional sales data
(Sales Order). These are based on replicated data from CRM One-Order structures.
The HRF rules have to be created manually in the system as Predictive Model Fit.
After these steps the CEI user with role SAP_CEI_PBA can create predictive models
based on this predictive scenario. The assigned implementation method here is SAP
HANA Rule Framework (HRF) with the corresponding vocabulary and object models.

Customer

Predictive Examples
Demo Insurance (SP05)
Demo Insurance has the predictive use case Contract Cancellation Propensity for life
insurance
The ADT-Profile Demo Insurance is available via BC-Set
CEI_ADT_CRM_DEMO_INSURANCE (transaction SCPR3)
Data can be generated with an ABAP report SEPMH_DG_CEI, which is not shipped
before NW 7.40 SP10! In the meantime a csv-file with some generated data could be
made available. The data set contains about 20 contract attributes and 30 master data
attributes. The program generates data distribution and correlation in the data, which are
realistic and can well serve for predictive demos.
The Predictive Scenario is also available via BC-Set
CEI_CPRED_DEMO_INSURANCE for Predictive Scenario Demo Insurance
After these steps the CEI user with role SAP_CEI_PBA can create predictive models
based on this predictive scenario. The assigned implementation method here is Logistic
Regression based on PAL as well as SAP Infinite Insight.
Customer

Summary

Summary

You should now be able to


Explain Predictive Model Management
Use the features of Predictive Model
Management Marketing Permission
Publish Predictive Models into Business
Scenarios for real-time calculation of scores (e.g.
Buying Propensity)
Understand Predictive Success Reporting
Explain Predictive Scenarios and Use Cases

Customer

2015 SAP SE or an SAP affiliate company. All rights reserved.


No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company.
SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate
company) in Germany and other countries. Please see http://global12.sap.com/corporate-en/legal/copyright/index.epx for additional trademark information and notices.
Some software products marketed by SAP SE and its distributors contain proprietary software components of other software vendors.
National product specifications may vary.
These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP SE or its
affiliated companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP SE or SAP affiliate company products and
services are those that are set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as
constituting an additional warranty.
In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop
or release any functionality mentioned therein. This document, or any related presentation, and SAP SEs or its affiliated companies strategy and possible future
developments, products, and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time
for any reason without notice. The information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forwardlooking statements are subject to various risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place
undue reliance on these forward-looking statements, which speak only as of their dates, and they should not be relied upon in making purchasing decisions.

Customer

You might also like