There is a particular silence that data people will recognise. A dashboard goes up on the screen, the numbers are all present and correct, and the room pauses not because anything is wrong, but because nobody is quite sure what the figures are saying.
Charts are wonderful at showing shape. Tables are wonderful at showing detail. But neither will look you in the eye and say: here is what actually happened this quarter, here is why it matters, and here is what to watch next. That sentence the one a good analyst would say out loud is the thing dashboards leave out. Closing that gap is what this project was about.
Dashboards that show everything and explain nothing
I was working with a commercial real estate advisory firm running a thorough Power BI setup. Clean data, sound models, well-built reports. By every technical measure it was working.
In practice, the people who needed the insight most brokers, executives, analysts rarely had time to sit with a dashboard and tease out what the numbers meant in context. They wanted summaries. Written, human-readable summaries explaining what was happening in the market, what had shifted since last month, and what deserved a second look.
The obvious route was Microsoft's built-in Smart Narrative. The catch was the bill: unlocking it meant a Fabric capacity licence running to several thousand dollars a month. For a focused advisory firm, that is a heavy line item for one feature especially one that gives you very little say over tone, structure, or depth.
So I built my own.
An AI narrative that lives inside the report
Rather than pay a premium for a feature I could not fully control, I engineered a custom solution that sits entirely inside the existing Power BI environment, needs no extra Microsoft licensing, and honestly produces summaries more useful than the native tool does.
In plain terms, it works like this. The report already holds the numbers: sales volumes, pricing trends, market activity, deal flow, year-on-year comparisons, all modelled and calculated. I wrote a layer of logic that takes the ones that matter and sends them, over a secure connection, to an AI model along with a carefully written brief: what kind of summary to produce, in what tone, at what length, and from whose point of view.
The model writes the narrative to that brief and returns it straight into the report. Seconds, start to finish. No separate tool, no export, no manual step. And the part that makes it genuinely useful: change a filter a different borough, quarter, or property type and the narrative rewrites itself automatically. New context, new data, new summary, in real time.
Why this is different from the built-in version
The native Smart Narrative does one thing: it generates a generic description of whatever is on the page. You cannot meaningfully steer the language, the depth, or the angle. It reads exactly like what it is an automated caption.
This is a different kind of thing. I can have it write as a market analyst, picking apart trends and anomalies in detail. I can switch it to an executive summary three paragraphs, no jargon, focused on decisions. I can ask for a broker-facing version that leans into opportunities. Tone, length, focus, structure all of it is set by the brief, not by whatever Microsoft decided.
It is not a template being filled in. It is a real narrative composed from live data, following editorial direction I define with the client. One report, several perspectives, all generated in seconds.
What it actually delivers
Cost. The licence needed to unlock the built-in feature runs to several thousand dollars a month comfortably forty to fifty thousand a year. This runs on ordinary API usage, in most cases under a hundred dollars a month. That is not a saving at the margin; it is a different order of magnitude.
Control. With the native feature you get what you are given. Here, the client decides exactly how the narrative reads. Want a new perspective next quarter? I adjust the brief. No new licence, no platform upgrade, no waiting for a feature request to be noticed.
Quality. I have run them side by side. The built-in output is serviceable but generic. These read as though written by someone who understands the business because the instructions encode real domain knowledge: what matters in commercial real estate, the language professionals there actually use, and what a useful summary looks like.
Speed. Filter changes, narrative follows. Seconds. No lag, no loading screen, no "generating your report" spinner. It feels native because, in effect, it is.
How it fits together, briefly
For anyone who wants the outline without the deep dive: the foundation is a well-architected Power BI model several sources connected and shaped through a careful ETL process, with a calculation layer producing the KPIs the business relies on. Inside that environment sits a custom script that pulls the relevant numbers, builds a structured request data plus brief and sends it to the AI over a secure API. The finished text renders directly in the report. No external tools, no browser tabs, no copy-paste. Numbers on one side, a professionally written summary on the other, both updating together as the context changes.
What this says about how I work
I could have recommended the premium licence. It is the easier conversation you need this, here is the price, let us move on and it is what a lot of consultancies would do, because it asks nothing of them. Instead I looked at what the firm actually needed, weighed what was available, decided I could deliver something better for a fraction of the cost, and built it.
That is not a one-off. The question I start from is never "what is the standard solution?" but "what is the right solution for this particular context?" Sometimes that is an off-the-shelf tool. Sometimes it is something that does not exist yet and when it is, I build it.
If you are running Power BI reports and quietly wishing they could explain themselves, that is a solved problem now. If you are being quoted thousands a month to unlock a feature that does half of what you need, there is probably a better path. And if you are sitting on a well-built model and wondering what more it could do, the answer is almost certainly more than you would expect. Get in touch if that sounds like your situation.