Your Dataset Is Certified. The Room Still Asks for the Excel

The workspace shows a certified badge. Someone in the QBR still asks for “the file” anyway, not because they hate Power BI, but because the certified dataset does not answer the question they already trust Excel to answer.

Endorsement is not trust. Trust comes from completeness, ownership, and a measure the room can defend.

Badges do not create belief

Mid-market manufacturers certify datasets to stop the wild west, and that instinct is right. The failure is certifying a partial product: the wrong grain for the decision, missing measures finance needs, and no named steward who will stand up in the room.

So leaders go back to the workbook they believe, and adoption slides while the governance slides stay green. It is the pattern in export to Excel is the real adoption metric, except now the badge makes a product gap look like user stubbornness.

Certified versus wild west describes a promotion path, not a sticker. If the certified model cannot state the sentence for revenue, margin, or on-hand at the grain the meeting uses, the room will invent a twin.

Why certified still loses to the spreadsheet

The grain is incomplete. The model stops at plant when the decision is made by line or customer, and Excel still has that cut.

Measures the room cares about are missing or renamed. “Net revenue” in the badge does not match the controller’s exclusion list. See measures nobody can explain.

There is no steward in the meeting. The author left and the badge stayed, so nobody can change or defend the number. That is the ownership hole from who can change a measure.

Refresh and as-of are fuzzy. The workbook has a timestamp someone typed, and the certified page says nothing. Silence loses to a dated tab.

Late adjustments live only in the file. Ops fixed a shipment in Excel and the certified model never saw it. Certification without a path for corrections teaches people to keep a side ledger.

The costs of a badge without trust

1. Dual systems become permanent. The certified model is “official” and Excel is “real,” so every close reconciles both. The badge sped up the politics, not the clarity.

2. Executives learn to ignore endorsement. After one confident, wrong certified total, the ribbon means nothing and future certifications start underwater.

3. Shadow measures multiply. Analysts export and rebuild the logic offline. The semantic model is the product only if people use it, and forks recreate five Revenues under a certified roof.

4. Stewards become ticket routers. Without authority to demote or complete the model, ownership is just a mailbox, and the room still asks for the file on Slack.

5. Close time shifts to archaeology. Controllers spend hours proving which certified tile matches which Excel column, and month-end turns into a scavenger hunt.

6. Ops stand-ups fragment. Plant managers print or paste from the workbook because the certified app lacks scrap reason or shift grain.

7. Governance theater spreads. Teams chase badge counts while completeness, grain, and steward presence never make the scorecard. The metrics improve and trust does not.

8. Change requests go to the wrong artifact. Leaders ask for “one more column in the Excel,” so the certified model never hears the requirement and debt compounds in the shadow system.

How to fix it: certify a product the room can use

1. Write the decision sentence before you certify. State what question the dataset answers, at what grain, on what calendar. If you cannot say it in one breath, do not endorse it. Certification without a sentence is costume jewelry.

2. Inventory the Excel the room still opens. List the cuts, measures, and as-of stamps people trust and map each one to the certified model. The gaps become the backlog, not a case for “adoption training.”

3. Name a steward who attends the meeting. That means a person, not a distribution list, who will defend the measure, accept change requests, and demote the badge when trust breaks. Put the name in the dataset description and on the close contacts.

4. Complete the grain the decision needs. If the QBR argues about customer margin, ship customer grain, or explicitly certify a coarser number and stop pretending. Partial grain with a full badge is how Excel wins.

5. Publish the measure dictionary next to the badge. Cover inclusions, exclusions, currency, and calendar, and link it from the app. When measures cannot be explained, certification should wait.

6. Put as-of and quality on the page. “Books as of 5:40 a.m. Reconciled to GL control.” That competes with the workbook’s handwritten timestamp. Pair it with the gates in refresh succeeded, data still wrong so the badge is never green over garbage.

7. Define a correction path. Decide how late adjustments enter the model before the next pack. If the only path is “fix it in Excel,” you certified a museum piece.

8. Demote loudly when the contract breaks. A missing steward, broken grain, or repeated Excel overrides should remove the endorsement. Wild west rules need a reverse gear.

9. Retire the competing file on a schedule. After three closes where the certified model matches the decision, archive the old workbook with a redirect note. Do not keep dual truth “for convenience.”

10. Measure trust, not badge count. Track how often the room still asks for the file on certified subjects. Falling requests mean certification worked. Flat requests mean you endorsed a hollow product.

What good looks like

The certified dataset answers the meeting’s first five questions at the right grain. The steward is in the room or on call, the as-of is visible, and the Excel for that subject is archived or clearly labeled unofficial.

People still export for analysis. They just stop treating a side workbook as the books. The badge means a contract kept, not a sticker applied.

Make certification a promotion, not a favor

Authors will ask to certify because a report is popular. Popularity is not readiness.

Require a steward, a sentence, a grain checklist, a measure dictionary, a refresh SLA, and a retirement plan for the Excel twin, with the same bar every time. Exceptions get a date and an owner, not a quiet badge.

Earned certification changes the room. Certification handed out as a favor leaves the file in place.

Start with one certified subject

Pick the subject that still generates the Excel ask, which is often revenue, inventory, or scrap. Complete the grain and measures, put the steward on the invite, and run one close without the twin.

Document what still failed and fix it before you stamp five more datasets.

Leaders should ask: “If we hid the Excel tomorrow, could this certified model run the meeting?” If the honest answer is no, the badge came too early.

Executive takeaway

Certification does not create trust. Completeness, a named steward, and measures the room can defend do. Stop counting badges, close the grain gaps, put ownership in the meeting, and retire the file once the product earns it. Then endorsement means something an executive can believe.

Want a hard look at why certified datasets still lose to Excel in your close or QBR? Book a session. We will map the Excel asks to model gaps, steward coverage, and the promotion bar that makes certification stick. Or start with a free Model Health check.