Power BI Insights - Using Explain the Increase to Understand Fluctuations

The most common question I hear about any dashboard is not "what is the number", it is "why did the number move". Sales dropped in July, why. Support tickets spiked last week, why. The chart shows the what perfectly well. The why is where people spend their time, and historically the way you answered it was to slice the data by every dimension you could think of, one at a time, until something jumped out. Slow, manual, and easy to give up on before you find the real driver.

Power BI has a feature aimed squarely at this problem. Right-click a data point on a visual, ask it to explain the increase or decrease, and it runs through your data looking for which categories contributed most to the change. Microsoft calls these insights, and the documentation on applying insights to explain fluctuations covers how to trigger them. What I want to give you is the consulting view: what this actually does, where it genuinely saves time, and the places you absolutely still need a person in the loop.

What it actually does

Say you have a line chart of monthly revenue and there is an obvious drop between two months. You right-click the point, choose analyse, and pick "explain the decrease". Power BI then automatically breaks that change down across the other fields in your model and surfaces the categories that moved the most. It might tell you the drop was driven mostly by one region, or one product line, or one customer segment, and it does it as a set of generated visuals you can flick through without building anything yourself.

Under the hood it is running the slice-and-compare work you would otherwise do by hand. It looks at how each category contributed to the overall change and ranks them, so instead of you manually splitting revenue by region, then by product, then by channel hunting for the culprit, it does that sweep automatically and shows you the ones that matter. On a well-built model with clean dimensions, this genuinely saves time. The first time a client watches it instantly point to the one region that tanked their quarter, they get why it is useful.

Where it earns its keep

The real value is speed on the first pass. When something moves and you have no idea why, insights gives you a fast starting point. It narrows a wide-open question down to two or three candidate drivers in seconds, and that is often enough to know where to look next. For a business analyst fielding "why did this change" questions all day, that is a meaningful chunk of grind removed.

It is also good for the questions nobody thought to ask. Because it sweeps across every dimension in the model rather than the two or three a person would check by instinct, it sometimes surfaces a driver you would not have gone looking for. You expected the drop to be about the usual seasonal region, and instead it turns out one product category quietly fell off a cliff. That kind of "huh, I would not have checked that" moment is where automated analysis beats human instinct, because humans check what they expect and skip what they do not.

And it lowers the barrier for people who are not comfortable slicing data themselves. Someone who can read a chart but would not know how to build a breakdown by hand can still right-click and get an explanation. That democratises the "why" question a bit, which is a good thing in an organisation trying to get more people making decisions from data rather than from gut feel.

Where it falls down, and this matters

Here is the part I make sure every client hears, because the feature is seductive enough that people over-trust it. Power BI insights explains correlation, not causation, and it does not know the difference. When it tells you a drop was "driven by" a region, what it actually means is that region accounts for most of the numerical change. Whether that region is the cause, a symptom, or a coincidence is a judgement the tool cannot make. It is doing arithmetic on your model, not understanding your business.

That distinction is not academic. I have seen someone take an insight at face value, report to their leadership that a product line was "the reason" for a revenue drop, and then discover the real cause was a pricing change that happened to land in that product line's numbers. The insight was arithmetically correct and strategically misleading. It pointed at where the change showed up, not why it happened. The tool did its job. The human skipped the step of asking whether the explanation actually made sense.

The other limitation is that insights is only as good as your model. It can only break a change down across the dimensions that exist in your data. If the real driver is something you do not capture, a competitor's promotion, a supply issue, weather, the tool will confidently attribute the change to whatever correlated dimension it can see, and that attribution will be wrong. It cannot tell you about a cause it has no data for. It will just find the best available proxy and present it with the same confidence it presents a genuine driver. On a messy model with badly defined categories, the output ranges from unhelpful to actively misleading.

How I tell people to use it

Treat insights as a hypothesis generator, not an answer. It is brilliant at quickly saying "here is where you should look first". It is not equipped to say "here is why this happened", even though the wording makes it sound like it is. Use it to shortcut the tedious first pass of narrowing down candidates, then bring human judgement to confirm whether the candidate it found is actually the cause or just where the number moved.

Concretely, when the tool points at a driver, ask the next question yourself. It says the drop was mostly one region. Fine, why did that region drop. Now you know where to dig, and you have saved the twenty minutes of manual slicing it would have taken to get to that same starting point. That is the right division of labour: the tool does the fast arithmetic sweep, you do the thinking about what it means. Handing the thinking to the tool is where people get burned.

I would also say do not put it in front of executives as a source of truth. It is an analyst's tool, a speed aid for the person doing the investigation. The output looks authoritative, clean generated visuals with a confident heading, and that polish makes people trust it more than they should. Keep it in the workshop, not the boardroom, until a human has sanity-checked what it found.

The bigger picture

Features like this are a small preview of where analytics is heading, which is toward tools that do more of the investigative legwork so people can spend their time on interpretation and decisions rather than manual slicing. That direction is genuinely good. The risk is the same one that shows up with every AI-flavoured analytics feature: the output looks so finished that people stop questioning it, and a tool that was meant to speed up thinking ends up replacing it. The organisations that get value from this are the ones that use it to ask better questions faster, not the ones that treat its answers as final.

If you want your Power BI estate set up so features like this actually work, meaning clean models with well-defined dimensions where an automated breakdown gives sensible results rather than noise, that is foundational work we do all the time. Have a look at our Power BI consultants page. And if you are thinking more broadly about getting genuine insight out of your data rather than just pretty dashboards, our AI for business intelligence work is aimed exactly at that. Either way, get in touch and we will talk through where you are and what would move the needle.

Insights is a good feature used well and a trap used lazily. It tells you where the number moved, fast. It does not tell you why, no matter what the button says. Keep that distinction in your head and it becomes one of the more useful things in the toolkit.