Copilot Can Build the Report. It Cannot Ask the Right Question.
[the following content is created with artificial intelligence technology]
I have spent twenty-five years watching BI work get faster. Every few years a new release promised to cut the build in half. The projects that failed kept failing for the same reason.
Now the promise is bigger.
In March 2026, Microsoft demonstrated Copilot reviewing a Power BI file. It wrote a 20-page optimisation plan in under 20 minutes. Microsoft put the manual effort for the same review at 16 to 40 hours.
I take that seriously. For that kind of work, the claim is close to true. It will get closer every release.
If you build Power BI reports, you have felt the question behind it. A client asks why they need you when Copilot exists. A manager asks why the next project needs eight weeks.
Look at what the demo did. It read a model that already existed. It found slow measures, oversized columns and a heavy refresh.
Every one of those is a question about how something was built. None of them asks whether it should have been built.
Your hardest weeks on a BI project were never about DAX. They were the weeks spent learning that "revenue" meant three different things in three departments. They were the report delivered exactly to spec that nobody opened after the second month.
Copilot was not in any of those rooms. It answers the question it receives. It has no way to know the question was wrong.
Here is the distinction that matters for anyone funding a BI project. Execution turns a clear requirement into a working model. Discovery finds out what the clear requirement is.
Discovery still runs on people. Someone has to hear what is said in the room and notice what is avoided. Then the organisation has to agree on one version of the truth.
A faster build makes a wrong requirement more expensive, not less. You reach the wrong report sooner. The model is so clean that nobody suspects the question behind it.
We have all seen what happens next. The report is correct, the numbers reconcile, and the business still decides in Excel. That was never a build problem, so no build tool fixes it.
So where does this leave the BI developer? In a stronger position than the headlines suggest, provided you move to where the value moved.
If your work is scoped as a build, every Copilot release shrinks it. If it is scoped as discovery followed by a build, Copilot becomes your fastest junior developer.
It writes the DAX. You decide which numbers deserve DAX at all.
If you fund BI projects, some vendors now quote fewer build hours because Copilot writes the code. Ask how many hours go to finding out what to build. If the answer is "the client provides the requirements," the project is a fast route to the report nobody uses.
At WizEmp, AI works on both sides of that line, and people decide. Wit, WizEmp's smart discovery process powered by AI, structures the conversations with every stakeholder.
People run the room, resolve the conflicts and sign off what the numbers mean. The build then starts from a specification worth building, and AI-assisted development makes it fast.
The moat in BI was never the code. It was knowing what to build. Copilot made that easier to see.
Microsoft says Copilot turns a 16 to 40 hour Power BI optimisation review into under 20 minutes. For execution, that direction is real. For discovery, nothing comparable exists. Understanding what the organisation needs and hearing the requirement nobody says out loud still takes people. If your next BI engagement is scoped as a build project, AI will eat your margin. If it is scoped as a discovery-led engagement, you hold the part no language model reaches. → Scope discovery first
Scope discovery first.
Copilot compresses the build. It does not decide what to build. One structured session establishes what your organisation needs from its next BI project before the first measure is written, so the faster build heads in the right direction.
Book a 30-Minute Discovery CallFrequently asked questions
Can Copilot replace a Power BI consultant?
For execution, Copilot already handles a growing share of the work: writing DAX, suggesting model optimisations, drafting report pages. Microsoft's own demonstration puts an optimisation review at under 20 minutes against 16 to 40 hours by hand. Discovery is different. Agreeing what the organisation needs and which definitions every department accepts depends on people and context. A language model has neither.
What does Copilot in Power BI do well?
It is fastest on a model that already exists or a requirement that is already clear. Strong uses include writing and explaining DAX, spotting slow measures and oversized columns, and summarising a report page. Its output still needs a developer who checks it against what the business meant.
Why does AI make BI discovery more important, not less?
A faster build shortens the time between a wrong requirement and a finished wrong report. When the build took eight weeks, someone often caught the misunderstanding halfway. When it takes days, the error reaches users before anyone questions it. Faster execution raises the value of getting the question right first.
How should a BI developer position their work as Copilot improves?
Scope engagements as discovery followed by a build, never a build alone. Treat Copilot as a fast junior developer that writes the code once you have decided what deserves building. The skills that gain value are challenging requirements and getting departments to agree on definitions. Judging whether AI output matches business intent matters more each year.