By Jeffrey Shear, J Shear Consulting

A question I'm hearing from clients more and more: "Claude can build dashboards now. Do we still need Power BI?"

It's a fair question. In October 2026, Anthropic released Claude Dashboards in beta, letting users build live dashboards by asking questions in plain English. At the same time, Microsoft has opened a preview of its Power BI MCP server, which lets AI assistants like Claude query Power BI data directly. The lines between "AI assistant" and "BI platform" are blurring quickly.

My short answer: they're good at different things, they're better together, and neither one will save you from bad data. Let's start with that last point, because it matters more than any tool decision you'll make.

First, the part that matters most: clean data

Every dashboard — whether it's built by a Power BI developer or generated by AI in thirty seconds — is only as trustworthy as the data underneath it. If your customer records are duplicated, your product codes are inconsistent, and three departments define "revenue" three different ways, no tool will fix that. It will simply display the problem.

AI actually raises the stakes. A clunky spreadsheet with a wrong number at least looks suspicious. An AI-generated dashboard with a clean design and a confident, well-written explanation looks authoritative, even when it's built on contradictory data. Polish makes bad numbers more believable, not less.

That's why, in Are Your Dashboards Lying to You?, I recommended a simple sequence: agree on the 10 to 15 metrics that run the business, write down what each one means, name a single source of truth for each, and clean the data behind them at the source. That work comes first, regardless of which tool you choose. Skip it, and the Claude-versus-Power-BI debate doesn't matter.

With that foundation in place, here's how I think about the tools.

What Power BI does best

Power BI is a mature, governed reporting platform, and for many mid-sized companies already using Microsoft 365, it's the natural backbone for business intelligence. Its real strengths are:

A single, governed definition of each metric. Power BI's semantic model lets you define "revenue" or "gross margin" once and use that definition everywhere.

Security at scale. Row-level security means each regional manager sees only their region, from the same report.

Reliable, scheduled refresh and distribution. Reports that dozens or hundreds of people rely on every week.

Long-term maintainability. Version control, deployment processes, and a large pool of people who know how to build and support it.

Its limitations are equally familiar: building good models takes skill, new questions often wait in an analyst's backlog, and reports tend to multiply faster than anyone uses them.

What Claude does best

Claude's strength is speed and flexibility — the distance from a business question to a useful answer:

Answering ad-hoc questions in plain English. "Why did margin dip in the Northeast last quarter?" without waiting for someone to build a report.

Prototyping dashboards in minutes. Test whether a view is useful before investing development time.

Explaining the numbers. Writing the narrative and commentary leaders actually read, and showing the query behind any figure.

Accelerating the Power BI team. Drafting and debugging DAX formulas, documenting data models, and reviewing measures for errors.

Spotting data problems. Flagging duplicates, gaps, and anomalies that need to be fixed at the source.

Its limitations matter too. Claude Dashboards is new and in beta, and its governance, distribution, and version-control capabilities are not yet at the level of a mature BI platform. Left unguided, an AI assistant may also calculate its own version of a metric rather than using your official definition. And as with any AI tool, you need to review how your data is handled under your plan and policies.

Which tool fits which job

Better together: the pattern I recommend

For most mid-sized companies, the strongest approach isn't choosing one. It's using each for what it does best, in the right order:

1. Clean the data and define the metrics. The metric dictionary and source-of-truth work described above. Non-negotiable.
2. Make Power BI the home of your official definitions. Build the semantic model so that "revenue" means one thing, everywhere.
3. Connect Claude to that model. Using Microsoft's Power BI MCP server, Claude can query your governed measures directly, and it runs under each user's own permissions. People get fast, plain-English answers that match the official numbers.
4. Prototype in Claude, promote to Power BI. When an ad-hoc dashboard proves valuable and many people need it, rebuild it as a governed Power BI report.
5. Use Claude to speed up the BI team itself. Faster DAX, better documentation, quicker reviews.

If your organization is deeply invested in Microsoft, Copilot in Power BI is also worth evaluating alongside this approach. The principle is the same: the AI layer should sit on top of governed, clean data, never replace it.

Practical cautions before you start

Check the status. Claude Dashboards is in beta and the Power BI MCP server is in preview. Features and terms will change; pilot before you standardize.

Involve IT early. Connecting Claude to Power BI requires an administrator to enable a tenant setting and register an application in Microsoft Entra ID.

Point AI at governed measures, not raw tables. This is how you keep AI answers consistent with your official numbers.

Verify before you trust. Click through to the query behind any important number, especially early on.

Review data policies. Confirm how data is processed and retained under your Claude plan and your own compliance requirements.

A composite example

Consider a 120-person distribution company — a composite of situations I've seen. They already used Power BI, but leadership was frustrated: new questions sat in the analyst's queue for weeks. Someone connected an AI assistant directly to the raw sales tables as a quick fix.

Within a week, two executives received different answers to the same revenue question, and trust in the experiment collapsed. The tool wasn't the problem. Duplicate customer records and inconsistent product codes were.

They stepped back, spent about two months cleaning customer and product data and agreeing on core metric definitions, and rebuilt their Power BI model around those definitions. Then they connected Claude to the governed model instead of the raw tables. Sales leaders now ask their own questions and get answers that match the monthly reports. The analyst backlog shrank, and the analyst's time shifted to building the few dashboards worth making permanent.

The bottom line

Claude or Power BI? For most companies, the answer is both, used in the right order: clean data first, clear definitions second, Power BI as the governed foundation, and Claude as the fast, conversational layer on top. Get the order wrong, and you'll simply produce inconsistent numbers faster — and more convincingly — than ever before.

If you're deciding how AI fits into your reporting, or your dashboards aren't earning your team's trust, I help companies build the clean data foundation first and then put the right tools on top of it.

Book a consultation →