The First 90 Days of a Power BI Program

Ninety days is enough to prove a reporting loop. It is not enough to boil the company.

A mid-market program that spends the quarter on a roadmap slide has already failed the only test that matters: did one meeting start using one trusted number? That takes one model, one app, and one owner. Everything else is a sequel.

The quarter is a loop, not a museum of pages

Leaders search “Power BI 90 day plan” after a kickoff that felt expensive. They were promised a platform and got a backlog of dashboards and a steering color.

The right first quarter does not ship a catalog. It ships a loop, from source to model to app to meeting to steward, and then it repeats. That is delivery, not a tool upgrade. We already said programs fail in delivery, not in the stack. This piece is the calendar.

This is operating work, not a proof-of-value slide for a later pitch. If day ninety still points at “phase two,” you ran a workshop, not a program.

Three objects must exist by the end.

One semantic model. Grain named, measures named, refresh owned. The model is the product, and the reports can wait to be pretty.

One app. This is the place the ELT actually opens, not twelve workspaces or a SharePoint scavenger hunt.

One owner. A person who can cut scope and freeze a definition. Hosting is not owning, as covered in why the project has no owner.

If you cannot staff those three, do not start a “program.” Start a smaller loop in one domain and tell the truth about it.

What a slide-first ninety days costs

1. The close does not move. You designed a future estate while finance still pastes and month-end still takes a week. The P&L did not notice the quarter.

2. Every VP adds a page. Without a cut line, the backlog is a suggestion box. The team is busy and nothing certified ships. Intake without a no is how you commission a dashboard per question.

3. Steering hears architecture instead of a meeting change. Green on “model in progress” is not control. Control is knowing this measure is signed, this pack is live, and this twin is retired. If Monday still starts with email, the status was decoration.

4. Change gets scheduled after go-live. Users keep the workbook. You declared a deploy, but the operating cadence did not move. Change work belongs in week one, which is the point of data project management and change leadership.

5. You buy platform before product. Capacity, a lake, and another license batch do not create the loop. Capacity is not a strategy. The first quarter should starve the extras until one app is in use.

6. Trust never gets a chance. A demo with sample data teaches the room to ignore the next link, and the second ninety days starts underwater.

What not to do

Do not inventory the whole company before you certify one domain. A catalog without a live pack becomes another report.

Do not staff a center of excellence as a substitute for a steward. A CoE without a ship date is a committee.

Do not measure the quarter by report count. Measure it by one signed measure in one meeting, with the old file gone.

Do not wait for perfect source data. Pick a grain finance will sign with the dirt that exists, then tighten it. Waiting for a pristine lake is how the quarter ends empty.

The ninety-day plan: one model, one app, one owner

1. Days 1 to 15: pick the decision and name the three seats. Decide which meeting and which number. Name the consumer (controller or plant VP), the steward who can freeze grain, and the operations owner for refresh, and write it down. If two domains are “equally urgent,” pick the one that already argues every Monday.

2. Days 1 to 30: freeze grain and a short measure list. Revenue, margin, inventory, on-time, or whatever that meeting actually fights about, but not forty KPIs. Write a sentence per measure covering include, exclude, grain, and owner, because unexplained measures are not KPIs. Connect sources that already exist. Do not rebuild the ERP.

3. Days 20 to 45: stand up the model as a product, not a file. Build relationships that match how the business ships and books, refresh on a clock the close can live with, and set access that matches who may see each plant. A failed 6 a.m. job is a close risk, not a curiosity, so put refresh on the same board as build.

4. Days 30 to 60: one app, one pack, one retire list. The brochure can be ugly, but it must be the pack. Point the meeting at it and name the twins you will turn off. If Excel is still the right place for scenarios, connect it instead of pasting. The split is Excel vs Power BI, not a religion.

5. Days 45 to 75: run the meeting on the app. Sit in the meeting and capture every “that’s not my number.” That list is the ranked backlog. Fix definitions instead of adding pages to dodge the argument. Finance sign-off is a feature, as covered in why finance won’t sign the dashboard, so treat refusal as a spec.

6. Days 60 to 90: lock the loop and cut the rest. The owner is still named, the steward is still answering, and the old file is gone or labeled working paper. Rank next quarter’s one addition, and let everything else wait. If you cannot say no at day ninety, you do not have a program. You have a queue.

Start in one domain, such as finance close or a single plant. Prove the loop, then copy the method, not the pages.

What good looks like on day ninety

The ELT can say where the number lives without a hunt, and the steward can say what the measure includes in one sentence.

Refresh finished before the meeting that used it, at least once during a close week.

Someone lost the old workbook and did not get it back, and the next request hits a backlog instead of a side file.

That is enough. A lakehouse, a CoE handbook, and a second department can all wait until the first loop is boring. If day ninety is a slide of future dashboards, you spent the quarter on hope.

Frequently asked questions

Can we do two domains in ninety days? Usually not. Two domains means two stewards, two grains, and two meetings, which is how both stay half-done. Sequence them.

What if source data is dirty? Dirty data shows up as arguments, not a red X, and you will hit it. Name the duplicates and late files, and do not pause the quarter for a tooling RFP. See data quality shows up as arguments.

Do we need a partner for ninety days? Only if nobody in-house can own sequence, stakeholders, and the first freeze at the same time. A borrowed lead for a quarter is cheaper than a year of status meetings.

Get started

Stop planning a platform. Fund one loop: one model, one app, one owner, ninety days.

Want to know what your first quarter should actually ship? Book a session. We’ll name the meeting, the measure, and the cut, not a roadmap of pages. Or start with a free Model Health check.