Use of APD(Analysis Process and Designer)
Everyone
who worked with BI 7.0 knows that Analysis Process Designer (APD) is
a workbench for creating, executing, and monitoring analysis
processes. The analysis process is primarily based on data that was
consolidated in the Data Warehouse and that exists in InfoProviders.
One of the applications of APDs from a technical point of view would
be feeding query results into a DataStore object or an attribute of a
characteristic. In this post I review a few examples on how
consultants may use APDs for addressing particular analysis tasks.
Analysis Process Designer allows you to set up a
model where you move data from source to target and do some
transformations on the way. As a source we can use any InfoProvider
in the data model. The following types of data target are available
in the Analysis Process Designer:
● Attributes of a
characteristic
● DataStore objects
● Files
●
CRM attributes
● Target groups for SAP CRM
● Data
mining models:
○ Training the decision tree
○
Training the clustering model
○ Training the scoring model
(regression)
○ Training data mining models from third
parties
○ Creating association analysis models
1.
Examples of business applications
1.1. ABC classification for
customers
In ABC classification we assign customers to certain
categories based on business rules. For example, you can classify
your customers into three classes A, B and C according to the sales
revenue or profit they generate. When you choose ABC classification
in APD you have to specify the characteristic for which the
classification is to be performed, its attribute, key figure,
appropriate query, and threshold values for the individual ABC
classes.
1.2. Scoring (traffic light) model
In a number of
BI scenarios we may have a requirement for generating scoring or
traffic light indicators for a certain set of KPIs. We may want to
know, for example, how close the actual value is to the budgeted one.
A range of traffic lights (red/yellow/green) needs to be displayed by
geography, product group, profit center, etc.
Traffic
light indicators need to be assigned to each report line based on a
complex logic. For example, if one or two countries in the region are
underperforming, region’s indicator is set to yellow. If more then
two countries are underperforming region’s indicator for the
analyzed period should be set to red.
As values for
traffic light indicators are not cumulative they have to be
calculated separately for each level of granularity. Knowing
indicators at the lowest level of granularity does not help much in
deriving them for upper levels, as there is a business rule defined
for each level separately. Therefore, we have to build a set of
queries for each level of data model where traffic light indicators
need to be displayed. APD would help us feeding query results into
the cube reporting on scoring results.
2. Example of data
flow for scoring model
The following data flow model can be
used for calculating scoring results. The infocube contains measures
(KPIs) used for scoring, such as sales volume and sales budget. It
also has a set of traffic light KPIs that need to be populated with
indicators for each granularity level.
3. Why using APD in
the scoring model
It is important to note that in the scoring
model instead of APD/Query approach one can use a transformation
(formerly known as an update rule) connecting cube to itself. In the
start/end routine we can build business logic required for scoring
results calculations:
However, this
approach requires complex development in ABAP. Specific scoring
requirements have to be documented by a business user in advance,
which usually makes development cycle longer. Any adjustments to the
scoring logic require ABAP code modifications.
Alternatively,
when we use Query/APD approach, analysts are able to define scoring
requirements in the queries, test and modify them whenever it is
needed. They can also run queries and check preliminary results.
Needless to say, it is usually easier to modify and test queries
rather than transformations with ABAP code.
No comments:
Post a Comment