How to build Power BI reports from Sparx EA

Somebody has asked for a Power BI view of the application portfolio, one that risk, finance and the portfolio office can open without calling the architecture team. There is no applications table in a Sparx EA repository, so Power BI cannot simply be pointed at it. This guide is for the architect or BI developer who has been given that job. It covers how the work is done by hand, how AI changes it, and the decisions that stay with you.

Why the repository will not report on its own

EA’s database is a metamodel, not a vocabulary. Every element is an object in t_object, its stereotype is applied through a cross-reference table, and its business attributes are name and value rows in another. A multi-valued tagged value, such as an application in scope for several regulations, is stored as one joined string, so the obvious count undercounts without any error. And nothing tells you how much of a field is filled in, so a roll-up by business owner looks as authoritative at 12% populated as at 100%.

How it is done by hand, and what it costs

By hand, an architect writes SQL against the repository tables, unpacks the joined strings, builds the tables Power BI needs, defines the relationships and writes the measures that walk from an application to the capabilities it supports. The first report takes weeks. Every change to the modeling language means reworking the queries, and the person who wrote them becomes the only one who can maintain them. Most teams stop at a spreadsheet export, which is stale when it is sent.

How it is done AI-augmented

The architect states what the reporting has to say, and AI does the mechanics. It reads what the repository actually contains, builds a database in your business vocabulary (if your modeling language defines a Business Application with a criticality, an owner and a regulatory scope, you get a business_application table with those columns), generates the Power BI model with the relationships declared and the traversal measures written, and counts the result back against the repository. The field list in Power BI shows your vocabulary and nothing else, so the first report you build is a report, not a modeling exercise.

That moves the skill. The architect no longer writes queries. The architect decides:

• What the business vocabulary is. What counts as an application, and which attributes people would say out loud.
• What the dashboard must answer. The portfolio question your organization asks every quarter, written down before anything is built.
• What counts as complete. How full a field must be before a roll-up is allowed to look authoritative, and what to do about the gaps the build reports.

Those decisions are the difference between a dashboard people trust and one they stop opening.

The decisions, in the order you will make them

Why the Power BI model is generated, not inferred

Hand Power BI a set of tables and let it infer the relationships, and it fails silently. On the Westbrook Bank reference model its auto-detect proposed 63 relationships where 38 were needed, invented 28 from shared values, and missed the traversal relationships entirely. The full finding is its own article. The report looks finished, and nothing in Power BI tells you otherwise, so the model is built deliberately: relationships defined on purpose, measures verified against the source, and EA’s own tables hidden so people see only your vocabulary.

Figures from the Westbrook Bank reference model, a fictitious model we use for teaching; they illustrate the shape of a result, not what you should expect.

What the check finds that a report would hide

The build reports where the repository disagrees with its own modeling language. The one that matters most is a tagged value whose live data falls outside its declared list of permitted values, which you cannot see in EA’s interface until something tries to enforce the rule. Where the data does not say which answer is right, the build asks instead of choosing one and presenting it as fact.

On the Westbrook Bank reference model, the check surfaced three application components with a complete governance tag set: classified Business-Critical, scoped to a financial-privacy regulation, with both business and technical owners named. They have no stereotype at all. Somebody classified them deliberately, and no report would have shown them. The tool reports; you decide what an unstereotyped Business-Critical system should be, and why that decision is yours is its own piece.

Microsoft 365 Copilot, read-only

Microsoft 365 Copilot answers questions from a published Power BI semantic model built from your EA data. Publish the model your reports sit on, and people can ask Copilot the portfolio question in plain language, as of the last refresh and within the access they already have in Power BI. With Kernaro AI Hub there is a second route, through Hub’s read-only API via Copilot Studio or a Microsoft 365 connector. Either way Copilot reads and does not change the model.

Where it stops

• It finds data-quality problems and does not fix them. It reports what is missing, inconsistent or drifted. Your architects decide what each gap means and what to do about it.
• It does not replace Prolaborate, EA’s own reporting or a data warehouse. It gets architecture data into the reporting tool your organization already uses, and the same reporting database can feed Microsoft Fabric or your warehouse when you need it there.

A roll-up looks authoritative at 12% populated and at 100%. The honest report says which one you have.

How we help

We can help you build a reporting database on top of your Sparx EA repository. If you host your repository yourself, we can build the reporting views in the same database and you read from them live. If you don’t have access to the database, or it’s hosted elsewhere, and you need the reporting database in something like Snowflake or Microsoft Fabric, we can help you set it up so you refresh it from your EA model on the schedule you define. If you have Kernaro AI Hub, we can help you use its Power BI connector instead. The mechanics are in the NovoCircle documentation (opens in a new tab).

If you are still choosing a BI tool, read Power BI vs Tableau for Sparx EA analytics. The AI tools for Sparx EA page shows where this sits among the other options.

Frequently asked questions