Data Analytics
  • All of Microsoft
  • Python
  • My Role
    • All in My Role
    • Data Analytics
  • Focus
    • All in Focus
    • Microsoft Fabric
    • What's Hot
    • M365 Release
  • Security & Identity
Timespan
explore our new search
Power BI: assess an inherited model before changing it
Power BI

Power BI: assess an inherited model before changing it

by HubSite 365 about Pragmatic Works

Use a decision flow and a table-by-table review to decide whether to leave, tweak, refactor or rebuild an inherited Power BI model.

Key insights

  • Start by asking whether the model answers business questions and refreshes reliably.
  • Choose among leaving the model alone, tweaking it, refactoring it or rebuilding it.
  • Use table names, columns and relationships to identify likely fact and dimension tables.
  • Review many-to-many and bidirectional relationships, and look for better key columns.
  • Rearrange the model view to group dimensions separately from fact tables.
  • A shared date table may already provide a useful connection between purchase and return facts.

An inherited Power BI model is easier to assess when you first decide whether it needs no change, a small tweak, a refactor or a rebuild. This process is useful for Power BI developers and analysts who have to understand an unfamiliar model before changing its logic or relationships. Pragmatic Works demonstrates a practical review of a deliberately messy model, including how to identify likely facts, dimensions and relationships that need investigation.

What's new

This is a review method for an existing model, not a new Power BI feature. Its central idea is to avoid treating every unfamiliar model as a rebuild project. First check whether the model answers the questions the business asks, whether it refreshes reliably and whether its structure is still recognizable.

The decision flow separates four possible responses: leave the model alone, tweak it, refactor it or rebuild it. In the example flow, a model that answers business questions and refreshes reliably can be left alone. If the structure is difficult to understand, consider whether you can still identify facts, dimensions and grain before deciding how much change is needed.

The demonstration gives rough effort and risk levels for the choices: a tweak is measured in hours with low risk, a refactor in days with medium risk, and a rebuild in weeks with high risk. It also recommends shipping a refactor in increments and getting sign-off before a rebuild.

Who it is for and what you need

This approach is aimed at people taking responsibility for a model built by someone else, whether that person left the organization or ownership simply changed. It can help an analyst or developer get oriented before attempting repairs. You need access to the model and enough familiarity with the business questions it is expected to answer to evaluate whether it is doing its job.

Start in the model view and inspect the tables and their relationships. The method does not depend on assuming that table names are accurate: the example includes vague names such as “SKU-ish” and “Who_Bought_This.” Columns and relationships can help reveal what those tables represent. When names are unclear, the demonstration switches to table view to inspect the data.

How it works

  1. Decide the scale of the response. Ask whether the model answers the business’s questions and whether it refreshes reliably. If both are true, leaving it alone may be the right call. If it requires complicated workarounds or its structure cannot be made sense of, assess whether to tweak, refactor or rebuild.
  2. Find a likely fact table. Look for tables that represent events and contain measures or identifiers. In the example, cost, gross and margin columns point toward a sales fact table. The table’s relationships provide another clue, but broken or bidirectional links can also flag areas for review.
  3. Identify dimensions from the fact table outward. Move the likely fact table aside in the model view, then inspect connected tables one by one. Check their contents in table view when the name is unhelpful. For a suspected product table, look for a more suitable key or another useful category column rather than accepting a problematic many-to-many relationship without investigation.
  4. Investigate relationship problems. Review many-to-many relationships and bidirectional filters, and check which columns connect the tables. In the channel example, repeated channel codes explain why the relationship is many-to-many; the presenter suggests adding an ID column as a possible fix. Treat this as a clue to investigate the data structure, not as a universal fix for every model.
  5. Look for facts hidden among other tables. A table with promotion or discount events may represent a fact rather than a simple lookup. The example also identifies a returns table as another fact. Separate these from dimensions such as date, customer, product, channel and territory where the table contents support that classification.
  6. Organize the view and preserve what works. Place dimensions across the top and facts to the side to make the model easier to inspect. Then look for functioning relationships. In the example, a shared date table connects purchase and return facts, giving the reviewer a useful starting point while other relationships remain under review.

This sequence helps build an inventory of what appears to work and what needs more attention. It does not require changing every relationship during the first pass. The demonstration’s emphasis is on understanding the model and identifying a sensible starting point before moving on to changes.

Limitations and alternatives

A model view can suggest a table’s role, but names and relationships alone may not settle the question. The example checks table contents and key columns when the first interpretation is uncertain. It also shows that a table described as a dimension may need further work in Power Query to become a suitable dimension.

Many-to-many relationships and bidirectional filters are warning signs to examine, not proof that every affected table should be replaced. A model may contain useful parts alongside broken links: the shared date table in the demonstration still supports both purchase and return facts. Preserve and understand those working pieces as you choose between a small tweak, incremental refactoring and a higher-risk rebuild.

Use the Power BI topic page for related coverage. The key decision is whether the model remains understandable and dependable for its users, not whether every table can be made to look tidy immediately.

Sources

Frequently asked questions

How should I start reviewing an inherited Power BI model?

Check whether it answers the business’s questions and refreshes reliably. Then identify likely fact and dimension tables and inspect their relationships.

How do I choose between tweaking and rebuilding a Power BI model?

Use the model’s business fit, refresh reliability and recognizable structure to guide the decision. The demonstrated flow treats rebuilding as a higher-effort, higher-risk option than tweaking or refactoring.

How can I identify fact and dimension tables in Power BI?

Look for events, measures and identifiers to find likely facts, then inspect connected tables and their columns to identify dimensions. Table view can help when names are unclear.

What should I check in a messy Power BI model?

Review many-to-many relationships, bidirectional filters, broken links and potential fact tables disguised by unclear names. Also look for working relationships, such as a shared date table.