I had the chance to present 8.4.2021 in Power BI User Group Finland how Helen Ltd. uses Power BI composite models at the enterprise scale. Here is a summary of the presentation, with the example of sales and marketing business intelligence. Helen has over 450,000 customers throughout Finland.
With composite models it is possible to connect to multiple data connections, no matter if using DirectQuery or import mode. When connecting via DirectQuery or Power BI dataset, one can use all the measures of the connected dataset.
One of the key advantages of the composite models is to have small and manageable data models but still have the power to combine all the data and build integrated reports. In addition, it is possible to divide responsibilities and work more effectively as multiple persons do not need to update the same data model, or one person is not a resource constraint due to the maintenance of one huge model. This is great for risk avoidance too.
Furthermore, data update cycles can be different in the different small models, and the failure of one small model does not lead to the failure of all the reporting like with a huge model.
The latest versions of Power BI have the composite modelling feature as a standard. The only thing to do is to add data sources. However, the thing to remember is that currently three layers are supported. If model 1 reads model 2, one can still build model 3 which reads model 2. However, one cannot build model 4 which reads model 3 (chain of 4 steps or layers, 1-2-3-4).
For details, please consult https://docs.microsoft.com/en-us/power-bi/transform-model/desktop-composite-models.
The figures below present the three layers used in the context of sales and marketing reporting in Helen Ltd.
- Layer 1: the key data, concepts, and measures of a business function. In this case, sales and marketing. Layer 1 responds to the shared fundamental needs of multiple reporting and analytics use cases.
- Layer 2: department level reporting. Examples include pricing, customer experience and mobile application development. Different persons are responsible for layer 2 than for layer 1.
- Layer 3: a 360-overview of the selected Layer 1 and Layer 2 measures. Only using links to reports does not work as we want to follow the development of different indicators over time, as an example.
If you are interested on how to improve business with data and analytics, see my post: How to Serve Three Dissimilar Analytics User Profiles.
Source for the icons used in figures: https://vision.cloudera.com/azure-marketplace-features-cloudera-customer-360-offering