Samples and Profile Selection 

What are samples?

Samples are a representative selection of business partners on a database, for example ten per cent of the total number. The term "representative" means here that the business partners the sample contains are typical for the database as a whole, that is the distribution of attribute values is roughly the same for the business partners in the sample as for all the business partners on the database.  In sampling terms, this means that the pattern of distribution of attributes among the business partners within the sample is the same.

Samples are used to support profile modeling. The figures for distribution of attributes calculated during sampling are estimates based on the sample.  When you use samples, counting operations that you trigger in the Segment Builder are then performed only on this reduced number of business partners. The number of business partners displayed on the interface is a projected figure arrived at by multiplying the interim result by the relevant factor. The accuracy of the projected figure will depend on the size of the sample and on how representative it is. Generally speaking, the smaller the sample, the lower the accuracy, but the faster the count.

Your sample consists of 5,000 business partners out of a total of half a million. When, during creation of profiles in the Segment Builder, you select an attribute like "Male" the number of males appears as 250,000 (assuming 50-50 distribution between the sexes).  Only the business partners in the sample (5,000) have been checked: the figure 250,000 is arrived at by multiplying the figure 2,500 by the appropriate multiplier (in this case 100).

Technically speaking, each sample is a target group of category "sample" and therefore belongs to a profile set that is used to group several samples together. This means that you first create a profile set, then assign the relevant samples to this set. You could for example create samples of different sizes (and therefore each offering a different performance/accuracy trade-off) all based on the same algorithm and stored within the same profile set. 

The sample type is used to control authorizations as to which users can access which samples in the Segment Builder.

When should you use samples?

No absolute figures can be given, but generally samples should be used whenever a database is reasonably large. During profile modeling with the Segment Builder, the distribution of attributes among business partners can be displayed online using the "Count" function.  Therefore, if the database is very large and samples are not used, performance could suffer: waiting for this information on attribute distribution could take an inordinate length of time, thus slowing down the modeling process.  Accurate samples enable you to forecast the final size of a target group with a certain degree of accuracy. 

Samples are used during the modeling phase only: when you generate the relevant target group(s), all the customers on the database are selected.

What sampling methods are available?

SAP supplies as standard a total of eight rules for generating samples. You can also implement your own algorithms by writing your own function modules. For more information, see the Implementation Guide for Customer Relationship Management under Marketing ® Segment Builder ® Define Rules for Building Samples and Processing Target Groups.

Using samples in the Segment Builder

Sampling in Segment Builder is activated by selecting a sample group and a specific sample within this group. You do this by choosing Extras ® Settings when you first start the Segment Builder. For more information, see User Settings in the Segment Builder.

If you do not select a sample, sampling is automatically inactive.

If, while working in the Segment Builder, you select a different sample, the system recalculates the distribution of attributes based on this sample. This means that during the modeling process, you can start with a small, fast sample and then move on to a larger, more accurate one.

Samples and BW queries

To use BW queries during target group selection, you must extract your sample(s) to BW.  To do so, in the transaction Maintain Samples for Segment Builder choose Export Sample to BW. This function is available from CRM Release 3.0 Support Package 05 and from BW Release 2.1 Support Package 08.

Samples and InfoSets / InfoSet queries

If you use InfoSets based on table joins, you require a separate InfoSet for sampling.  This InfoSet should be a copy of the original InfoSet that, however, also contains the additional table CRMD_MKTTG_TG_I. This is needed because the result of a join without a sample filter would be too broad (i.e., each business partner would appear in the sample as often as it appears in the target group. This separate InfoSet is required to incorporate the representative selection of business partners (that is, the sample) into the generated database query. The first InfoSet (without sampling) is still required both for target group selection and for counting of business partners without sampling in the Segment Builder.

SAP supplies a number of InfoSets as standard, for example the InfoSet CRM_MKTTG_BP_ADDR with corresponding sampling InfoSet CRM_MKTTG_BP_ADDR_SMP. For more information, see Standard InfoSets for Use in the Segment Builder.