Power BI Quick Measure Suggestions - Where AI-Generated DAX Actually Helps

DAX is the part of Power BI that quietly kills projects. Not because it's impossible, but because it's the place where business analysts who were doing fine in the visuals suddenly hit a wall. They know what they want ("show me sales growth versus last year, only for active customers, in the same store base"), they just can't get the syntax out. So they paste a forum snippet, change a few names, and ship something that's wrong in a way nobody notices for six months.

Quick measure suggestions is Microsoft's attempt at fixing this. You type what you want in natural language, the system writes the DAX for you. We've been using it on client engagements at our Power BI consulting practice since it came out of preview, and the verdict is more nuanced than "AI writes your measures now."

What it actually does

You open the modelling view, hit "New measure", and instead of the formula bar there's a text box. You type something like "year over year sales growth as a percentage". The model thinks for a few seconds and gives you back a measure with a name, the DAX, and a short explanation of what it built. If you like it, you accept. If you don't, you regenerate or refine your prompt.

The DAX it produces is generally clean. It uses CALCULATE, DATEADD, DIVIDE with a fallback for divide-by-zero, and the variable names make sense. For a junior analyst who's been wrestling with time intelligence for the afternoon, this is the difference between giving up and shipping something.

The interesting thing is that it actually reads your model. It knows your tables, your relationships, your existing measures. So when you ask for "sales growth", it figures out which column is the sales amount and which table holds your date dimension. You're not just getting a generic snippet, you're getting something that fits your model.

Where it works really well

Time intelligence is the sweet spot. Year-to-date, month-over-month, rolling 12 months, prior year comparisons, all of that. These patterns are well-documented, well-trodden, and the AI has clearly been trained on a lot of clean examples. If you ask for "rolling 90 day average of orders" you'll get a measure that does exactly that, including the right handling of dates with no transactions.

Ratio measures are another good one. "Percentage of total sales by category" or "share of margin by region" come out clean. The system knows to use DIVIDE rather than the division operator, and it builds the ALLSELECTED pattern correctly.

Filter ratios work too. "Customers who purchased in 2025 but not in 2024" is the kind of thing that takes a careful read of EXCEPT or filtered tables, and it gets it right more often than I expected.

The other place it earns its keep is when you have a measure that's nearly what you want, but you can't figure out how to tweak it. You can describe the change in plain English and let it have a go. Faster than digging through documentation.

Where it falls over

Anything involving custom business logic is hit or miss. We had a client whose definition of "active customer" was specific to their loyalty program, with different rules per region. The quick measure suggestion tried valiantly, but it had no way of knowing what "active" meant in their context. You can spell it out in the prompt, but at that point you're writing pseudo-code in English and translating, which isn't faster than just writing the DAX.

Multi-fact-table models confuse it. If you have a sales fact and a returns fact and you want net sales, it tends to pick one or the other rather than combining them properly. The fix is to make sure you have a clean measure for each and ask it to combine them, but that's a workaround, not a feature.

The other thing to watch is performance. The measures it generates are correct, but they're not always optimal. For a small dataset it doesn't matter. For a 50 million row fact table on a Premium capacity, the difference between SUMX and a properly filtered CALCULATE can be the difference between a 2 second visual and a 30 second one. Don't ship these straight to production without testing them on real data volumes. We've seen this exact thing burn teams during Power BI migrations where the new model looks fine on the dev sample and dies in prod.

It also doesn't know your governance rules. If your organisation has decided that all currency conversions go through a specific exchange rate table with specific filtering, the quick measure won't know that. It'll write something that looks reasonable and is technically wrong by your business rules. This is exactly the kind of subtle error that costs months to catch.

How we use it on client work

The pattern that's emerged from our consulting work is this. Quick measure suggestions is a draft tool, not a delivery tool.

For exploratory work, when an analyst is figuring out what questions are worth answering, it's brilliant. You can prototype a dozen measures in the time it used to take to write two. The conversation with the business becomes "is this the metric you mean?" instead of "give me a week and I'll get back to you".

For production measures, treat the output the same way you'd treat a code review from a clever junior. The logic is usually right, but it needs a second pair of eyes for naming consistency, performance, and business rule alignment. Most of the time we keep about 70% of what it generates and rework the rest.

We've also started using it as a teaching tool. When a junior analyst writes a measure manually and gets it wrong, we'll ask them to describe what they wanted in plain English, run the quick measure suggestion, and then compare the two side by side. They see the pattern they should have used, and the explanation field tells them why. It's faster than walking through it in a session, and it sticks better. This dovetails nicely with the kind of training we offer through our Power BI consulting work.

The thing nobody mentions

Quick measure suggestions only works in Power BI Desktop with the right tenant settings. If your IT team has locked down the AI features (which a lot of Australian enterprises have done while they figure out their AI governance policies), you won't see it at all. The feature shows up greyed out with no real explanation.

If you're at a bank, an insurer, or any government department in Australia, check with your platform team before promising your business stakeholders that this will speed things up. We've had projects where the demo went well on a personal tenant and then the feature was completely unavailable in the client environment. Embarrassing.

The other thing is data sensitivity. The suggestions are generated by sending your model metadata (table names, column names, measure names) to a Microsoft service. The data itself doesn't go, but if your column names contain client identifiers or business secrets, that's a conversation worth having with your security team first.

What I'd change

Two things would make this much better. First, an option to provide context about the business. A small text field where you can describe what "active customer" means or how revenue recognition works in your business. The model is clearly capable of using that context, the feature just doesn't ask for it.

Second, the ability to fine-tune on your existing measures. If I've already written 50 measures that follow a specific style and naming convention, the quick measure should match that style. Right now it ignores everything you've already built and writes in its own house style. That creates inconsistent models, which creates confusion six months later when someone tries to maintain them.

Microsoft will probably solve both of these eventually. The feature is being actively developed and the quality has noticeably improved over the last 12 months.

Where this fits

Quick measure suggestions is a small but useful tool. It's not going to replace good DAX skills, and it's not going to fix a badly designed model. But for the daily grind of "I know what I want, I just can't remember the syntax", it cuts a real amount of time. And for analysts who are learning DAX, it's the best teaching companion that's shipped in years.

If you want help getting your team productive with Power BI, including how to combine these AI features with solid modelling fundamentals, we offer Power BI consulting and broader business intelligence work across Australia. Worth a conversation if your reporting is slower than it should be.

Reference: Quick measure suggestions - Microsoft Learn