The Semantic Model Is the Product. The Report Is the Brochure

Executives buy reports. They point at a page and ask when it will be done.

The durable asset is not the page. It is the semantic model underneath: measures, grain, relationships, and a named owner. The report is the brochure. Brochures get redesigned. Products get funded.

The problem

The budget conversation starts with dashboards: how many, for whom, and by when.

That is how you get a beautiful cover on a weak product. Measures disagree, grain is whatever the last extract allowed, and relationships were inferred in a hurry. Nobody can say who is allowed to change “margin.”

Then a new question arrives. The team builds another report, copies the model, and tweaks one measure. Now you have two products and one brochure war.

Power BI did not fail. The funding model failed. You paid for pages and underpaid the thing the pages read.

This is not a DAX tutorial. You do not need to know how a measure is written. You need to know what you are actually buying.

What a semantic model is, in executive language

A semantic model is the agreed meaning of the business, in a form the platform can refresh.

Grain. What one row is: an invoice, a shipment, a day of inventory. If grain is fuzzy, every total is an argument waiting to happen.

Relationships. How those rows connect, such as customer to order, order to cash, and item to plant. If a join is wrong, the brochure will still look confident.

Measures. The definitions leadership argues from, such as revenue, gross margin, fill rate, and working capital days. Each one is named, documented, and owned.

Owners. Who may change those definitions, who signs off on the refresh window, and who answers when the number moves.

That is the product. A report is how one role sees it. Finance may need a P&L layout, ops may need a line view, and the CEO may need five tiles. It is the same product with different covers.

If finance still pastes actuals into Excel because it does not trust the model, that is the split covered in Excel vs Power BI for financial reporting. Keep the pack and fund the actuals layer.

The costs of funding the brochure

1. You rebuild the same logic every quarter. A new page needs a new copy of “revenue,” so analysts re-implement it and definitions drift. The meeting spends twenty minutes on whose brochure is right.

2. Every department ships a private product. Sales, finance, and ops each have a model. Each is locally correct, and together they describe three companies. Conflicting numbers are the symptom in the meeting, covered in Why Power BI Reports Show Different Numbers. The structural cause is unfunded shared meaning.

3. Refresh and performance tax the wrong layer. Bloated, copied models fight for capacity while you tune visuals, and the product is still a snowflake of extracts. Slow pages are often a model problem wearing a chart.

4. AI and “ask a question” fail in public. Natural language on a messy model invents answers from measures nobody agreed on. Chat does not create a definition. It recites whatever you already failed to agree on.

5. Turnover wipes out the asset. The person who “knew the file” leaves and the brochure remains, because the product was tribal. Handing a page to a new controller is not continuity.

6. The CEO cannot tell spend from progress. License counts go up and report counts go up, but the Monday number is still late. You bought an inventory of covers.

Poor governance makes all of this worse: access sprawl, no stewardship, no sunset dates. That framework is in hidden costs of poor Power BI governance. Funding the model is the capital decision inside it.

What the CEO and CFO should fund

Fund one official actuals product for the decisions that hit the P&L. Not fifty. Build one, then a second once the first is trusted.

Fund the steward’s time. A model without hours is a hobby, and someone has to review measures, refresh failures, and requests to change grain.

Fund the fence between the certified model and sandbox models. The board’s brochure reads only the certified one.

Fund Excel as a consumer, not as a second general ledger. Connected workbooks are fine. Parallel actuals are not.

Do not fund a visual redesign as a substitute for grain. Pretty does not fix a fact table that is the wrong shape. And do not fund a platform tour because the model is embarrassing. If one model does not work, another logo will not save it.

If you cannot name the three decisions the model must serve, start with strategy, not a new canvas. The Data & AI Strategy Roadmap is for that conversation. Day-to-day ownership of the product is closer to Managed Data & AI Advisory.

How to fix it

1. Name the product in writing. Use one sentence: “This model is the company’s booked actuals at invoice grain, owned by Finance, refreshed before 7 a.m. on business days.” If you cannot write that sentence, you do not have a product. You have files.

2. Keep the measures on a short list. Pick the ten (or fewer) the ELT will not argue about twice, and write what is in and what is out: returns, intercompany, bill-and-hold. The brochure can show more, but the product’s contract is the short list.

3. Assign a steward with a veto. Analysts propose and the steward publishes. If two teams can publish “the” margin, you funded two products. Deciding who may stamp the number is SSOT. This post is about the asset being stamped.

4. Keep reports thin. Pages should carry filters, layout, and audience, not a private copy of revenue. If a report needs a different grain, that is a product change, not a visual trick.

5. Connect the artifacts that already run the company. Point the board pack, the flash, and the plant huddle at the model. If they still paste, the product is not in production. You shipped a demo.

6. Prove one loop, then stop copying. One domain, one certified model, two or three brochures, and the same numbers in the meeting. Then you have earned the next domain. A Power BI Quickstart can force that loop on one painful report, but only if the report sits on a model you intend to keep.

What this is not

It is not “all finance in the model.” Scenarios, commentary, and pixel-precise statements can stay in Excel. The product supplies actuals, and judgment stays human.

It is not a star-schema lecture. Star schemas matter because they keep grain honest. The CFO funds honesty, not a diagram.

It is not waiting for a perfect warehouse. Many mid-market models start on a clean extract plus disciplined measures. Waiting for perfect is how the brochure factory keeps running.

What good looks like

The CEO asks for a new cut, and the answer is a filter or a thin page, not a new dataset. The CFO changes a policy, such as when revenue is recognized or what counts as scrap. The steward changes one measure and every brochure moves together.

A new plant manager gets the same product, restricted to their rows, not a fork. Capacity goes to refreshing the product instead of twenty cousins.

You still have reports. You just stopped treating them as the thing you own.

Get started

If your last analytics budget was a list of dashboards, you bought brochures.

Want to know whether you have a product or a pile of pages? Book a session with Alluvium. We will name the model, the steward, and the three measures that should be official. For a quick read on the model’s condition, start with a free Model Health check.