Every BI builder knows the feeling. You connect a fresh datasource, open a new dashboard, and stare at an empty canvas. The data is right there — hundreds of columns, thousands of rows — but the first chart is the hardest one to draw. Which metric leads? What belongs together on a page? Is anything in here even interesting? The blank canvas is not a design problem so much as a starting-momentum problem, and it is where most dashboards quietly stall.
AI Insights 365 — On-Prem Edition takes a different approach to that first step. Instead of asking you to invent a layout from nothing, it can read your datasource and build a complete starting dashboard for you — charts chosen, fields mapped, pages laid out. You go from connection to a working draft in one move, then spend your time refining instead of bootstrapping.
The real cost of the blank canvas
The empty dashboard is expensive in ways that do not show up on any timesheet. It front-loads all the decisions: what to measure, how to visualize it, how to arrange it. Those decisions require you to already understand the dataset, which is exactly the understanding a dashboard is supposed to help you build. It is circular, and the circle is where momentum dies.
The result is familiar. Datasources get connected and never visualized. "I'll build that dashboard" becomes a backlog item that ages for weeks. And when something finally does get built, it is often a cautious first guess — a couple of bar charts — that nobody trusts enough to act on.
The fix is not a faster chart editor. It is removing the cold start altogether.
There is a second, quieter cost too: inconsistency. When every dashboard begins as a blank canvas, every builder makes different first choices, and a team ends up with a patchwork of one-off layouts that are hard to compare and harder to maintain. A consistent, automated starting point does not just save the first hour — it nudges a team toward a shared baseline that individual dashboards can then diverge from deliberately rather than by accident.
How AI auto-report generation works
AI auto-report generation produces a full dashboard or report from a datasource. You pick the datasource, optionally add a prompt describing what you care about, and the AI builds out a dashboard with appropriate charts for that data.
The mechanics are straightforward from where you sit:
- Pick a datasource. Choose the connected source you want a dashboard for.
- Optionally add a prompt. Describe the angle you want — for example, the metrics, time frame, or audience you have in mind. Skip it and the AI works from the data alone.
- Generate. The AI analyzes the datasource and assembles a dashboard with charts suited to what it found.
- Review the draft. You land on a populated canvas rather than an empty one.
Crucially, this is not one generic routine bolted onto every source. AI Insights 365 uses per-datasource builders tuned to the kind of source you connected:
- SQL databases
- Power BI
- REST APIs
- File / URL sources
Each builder understands the shape of its source, so the charts it proposes fit how that data is actually structured rather than forcing every source through the same mold.
The difference is concrete. A SQL source exposes tables, columns, and types the builder can reason about directly. A Power BI source comes with model structure that informs how measures and dimensions should be visualized. A REST API returns nested, semi-structured responses that need different handling than a flat result set. A File or URL source might be a spreadsheet or a feed with its own quirks. A single generic generator would have to treat all of these as the lowest common denominator; per-datasource builders instead start from what each source actually offers, which is why the drafts feel like a sensible first pass rather than a random scattering of charts.
The role of the prompt
The optional prompt is where you steer. Left blank, the AI works purely from the data and gives you a broad, balanced starting dashboard. Add a sentence — name the metrics that matter, the period you care about, or the audience you are building for — and the output bends toward that intent. It is worth treating the prompt as a quick way to encode context the data alone cannot convey, such as which of several plausible measures is the one your stakeholders actually watch.
From auto-draft to your dashboard
An auto-generated dashboard is a starting point, not a final answer — and that is the point. The hard part, getting from zero to something concrete, is done. What remains is the work BI builders are good at and enjoy: shaping a real draft into the dashboard you actually want.
It also helps to set expectations about what the AI can and cannot know. It can read structure, types, distributions, and the patterns in your data, and it can act on the intent you express in a prompt. It cannot know the unwritten context in your head — that one segment is being wound down, that a particular metric is vanity, that last quarter's spike was a one-off you would rather not feature. That gap is not a flaw; it is the division of labor. The AI handles the mechanical, time-consuming first pass, and you supply the judgment that turns a competent draft into the dashboard your team will actually rely on.
Because the output lands on the same drag-and-drop canvas you would have used anyway, refining it is ordinary editing:
- Re-arrange or resize components across the page.
- Swap a chart type where a different visualization tells the story better.
- Adjust which fields drive a chart, or tighten a query.
- Add the KPI card the AI didn't know you cared about, or remove ones you don't.
- Split content across multiple pages as the dashboard grows.
You are editing forward from a working draft instead of building backward from nothing. That single change in starting position is what turns "I'll get to it" into "it's done."
Explain by AI: insight, not just charts
Generating the dashboard is half the value. The other half is understanding what it shows. For any chart, Explain by AI produces insight bullets — a concise read of what the visualization is actually telling you. Instead of squinting at a trend line and forming your own hypothesis, you get a quick, plain-language summary of the pattern in front of you.
This matters most for the people a dashboard is built for. A chart communicates to whoever already knows how to read it; insight bullets communicate to everyone else. They turn a visualization into something a stakeholder can act on without a follow-up meeting to ask what they are looking at.
Explaining the anomalies
Trends are useful; the exceptions are often where the decisions are. AI Insights 365 includes anomaly explanation for outliers — when a value breaks from the pattern, the AI explains the outlier rather than leaving you to notice it and investigate from scratch.
That changes the rhythm of analysis. The classic loop is: spot something odd, wonder if it is real, dig through the data to find out. Anomaly explanation compresses the first steps — the unusual point is surfaced and characterized, so your effort goes toward deciding what to do about it instead of confirming it exists.
Together, Explain by AI and anomaly explanation address the two questions a dashboard reader almost always has: "what is this telling me?" and "should I be worried about that spike?" Answering both in place, on the chart, is what keeps a dashboard from becoming a thing people glance at and then escalate to an analyst for interpretation. The interpretation travels with the visualization.
Why this accelerates time-to-insight
Put the pieces together and the throughline is speed to a usable answer:
- No cold start. A populated dashboard appears from a datasource and an optional prompt, so the first hour goes to refining instead of bootstrapping.
- Fit-for-source charts. Per-datasource builders propose visualizations that match how SQL, Power BI, REST, or File/URL data is structured.
- Built-in interpretation. Explain by AI turns charts into insight bullets, so understanding ships with the visualization.
- Faster investigation. Anomaly explanation surfaces and characterizes outliers, shortening the path from "something's off" to "here's why."
The cumulative effect is that more datasources actually become dashboards, those dashboards get built sooner, and the people reading them spend less time decoding and more time deciding. Time-to-insight is not improved by any single feature here — it is improved by removing friction at every step from connection to comprehension.
A typical first session
It helps to picture how this plays out end to end. You have just connected a SQL datasource for a system nobody has visualized yet. Instead of opening a blank dashboard, you point AI auto-report generation at that source and add a one-line prompt naming the two metrics leadership keeps asking about. A few moments later you are looking at a populated dashboard: a KPI card for each headline metric, a trend chart over time, a comparison across your main categories, and a detail table.
It is not perfect, and it is not supposed to be. You notice the trend chart would read better as a line than as the area chart it picked, so you switch it. You drop a composition chart that is too granular to be useful. You add a second page for the drill-down a stakeholder always requests. Then you open Explain by AI on the trend chart and paste its insight bullets into the narrative, and you let anomaly explanation characterize the odd dip in last month's figures so you can note the cause inline.
What would once have been a half-day of staring and second-guessing is now a focused half-hour of editing — and the result is something you would actually publish as a report, schedule as a subscription, or hand off with confidence. The starting momentum problem simply does not arise.
It runs on your engine, on your terms
A frequent and fair question about AI features is where the analysis runs. In AI Insights 365 — On-Prem Edition, the AI engine runs on your chosen engine: connect it to Azure, or run it fully air-gapped with Ollama. The auto-report generation, Explain by AI insight bullets, and anomaly explanation all operate within that choice.
For teams with strict data-handling requirements, the air-gapped Ollama option means the AI assistance described here can work without data leaving your environment at all. That is a meaningful distinction. Plenty of AI features are effectively off-limits to regulated or security-conscious organizations because they route data to an external service. Here, the same auto-generation and explanation capabilities are available whether you choose the convenience of Azure or the isolation of a fully air-gapped deployment — the feature set does not shrink when you choose the stricter posture.
The capability and the control are not a trade-off — you get AI that builds and explains dashboards, hosted the way your organization needs it, alongside the rest of the platform's on-prem foundation of connectors that reach your in-network databases directly, RBAC, and AD/Entra SSO.
Where to start
If you have a connected datasource sitting unused because nobody has had time to visualize it, that is the ideal first candidate. Point AI auto-report generation at it, add a one-line prompt if you have an angle in mind, and let it draft. Then do what you do best: refine the result, lean on Explain by AI to narrate the charts, and let anomaly explanation flag what deserves a closer look.
The blank canvas was never the interesting part of the job. Let the AI handle the cold start, and spend your time on the analysis that actually moves decisions.