Data Insight Writer

Paste the numbers from a dashboard or a board deck and get them written up three ways — a plain objective reading, a narrative version, and a recommended action angle. The three angles are produced by parsing logic that runs inside the page, so you can see the output before deciding whether an AI provider is worth connecting.

What a Dashboard Number Is Missing

A metric line like "SMB churn: 4.2% (+0.6pp vs Q1)" is a fact, not an insight. Turning it into something a reader can act on means deciding what direction the number moved, how big the move is relative to the figure itself, and whether the honest reading is good news, bad news or ambiguous. Doing that by hand for twenty lines is tedious and the result drifts depending on who writes it.

• The input field accepts one metric per line in any of the common shapes — "Q2 revenue: $24M (+18% YoY)", "New customers: 3,200 (+12%)", "Gross margin: 71% (+200bps)" — and the parser reads the metric name, the value and the parenthesised movement separately.
• A sign detector works out from the movement whether each figure rose, fell or held still, and treats basis points, percentage points and plain percentages as different units so a rise of 200bps is not described as a rise of 200%.
• Three output angles are generated from the same parsed set: Objective states what moved and by how much without judging it, Narrative frames the movement as a story with a direction, and Recommended proposes what the numbers suggest you should look at next.
• Context is optional but changes the wording materially. Give it "Q2 board update" and the phrasing leans toward governance and risk; give it "internal Slack post" and it loosens up. The tone selector does the same job at a smaller scale.
• Every figure the parser understood is listed back to you, so a line it misread is visible immediately rather than buried in fluent prose.
• Copy takes the visible angle only, so you can lift the objective reading into a report and leave the recommendation behind for your own eyes.

Getting Useful Wording Out of Thin Input

1. Type or paste your metrics, one per line. Keep the movement in brackets on the same line as the value — splitting a number and its delta across two lines is the most common reason a figure comes back unparsed.
2. Add context if the audience matters. "Q2 board update, internal Slack post" is a legitimate answer and the generator uses both.
3. Pick a tone, then press Generate Insights.
4. Read the Objective column first. It is the least interpretive of the three, and it is the fastest way to spot a number the parser read differently from how you meant it.
5. Switch to Narrative or Recommended for the same figures, then copy out whichever phrasing fits where you are pasting it.
6. If the wording feels generic, the fix is almost always more context rather than a different provider — the angles are derived from what you told it, and a bare list of numbers cannot support a specific recommendation.

Worked Through on a Quarterly Set

Five lines is enough to show where the three angles diverge. Take a quarter that grew but lost margin, with one regional number in the middle of a mix shift.

• Q2 revenue: $24M (+18% YoY) is unambiguously positive and the Objective angle says so without qualification.
• New customers: 3,200 (+12%) grows more slowly than revenue, which the Narrative angle picks up as expansion revenue carrying the quarter rather than acquisition.
• Gross margin: 71% (+200bps) and Europe share: 38% of new signups (+6pp) together read as a mix shift — growth is arriving in a region with different economics, so the margin gain and the regional shift are worth reading as one story rather than two facts.
• SMB churn: 4.2% (+0.6pp vs Q1) is the only genuinely negative line, and because it is the only one that moved the wrong way the Recommended angle centres on it even though it is the smallest number in the set.
• Notice that the sign of the movement, not the size of the number, drives the framing. A 0.6 percentage point drift in churn outweighs an 18% revenue gain in the recommendation because the two figures are not commensurable — a point worth remembering before pasting the output anywhere senior.

Where This Helps and Where It Does Not

• Weekly and monthly reporting where the same handful of metrics is written up repeatedly and the phrasing, not the analysis, is the bottleneck.
• Board and investor updates that need a consistent voice across several quarters — the objective angle in particular keeps the register steady.
• Drafting a first pass for someone else to edit. All three angles are written to be cut down, not published as-is.
• Not a substitute for knowing why a number moved. The generator sees the values you give it and nothing else; it cannot know that churn rose because a pricing change landed, and no amount of context in the text box will supply that.
• Not a forecasting tool. Every angle describes what the figures already show, so a set of numbers with no movement produces flat, correct and unexciting prose.
• Not suitable for anything regulated. The wording is a starting draft, and a compliance review still has to happen.

What Happens Without an API Key

The three angles come from parsing and phrasing rules that ship with the page, not from a language model. That means the default run works with no key, no account and no network request — the metric lines you paste stay in the tab. The parser handles the metric-name-and-bracketed-movement shape it was written for, and keeps the arithmetic of sign and unit straight, which is exactly where a hand-written summary usually slips. What it cannot do is invent a sentence it has no rule for: an unusual metric phrased in a way the parser does not recognise comes back as a value it lists but does not narrate.

• Connecting a provider is optional and changes the phrasing rather than the parsing — the metric list at the top of the results stays the same either way.
• If you do connect one, the settings panel keeps ten provider presets and a Custom base URL option, with the model name editable.
• Nothing about the tool requires sound to be on, a file to be dragged in, or a spreadsheet to be exported first.
• Reloading the page clears the text box. If a metric set is worth keeping, copy the output before you close the tab.

Questions About Reading Numbers This Way

Why is the wording generic when my numbers are specific?

Because the generator only knows what is on the lines you gave it. A value and a delta can support "revenue grew 18% year over year"; it cannot support "revenue grew because the enterprise tier landed", since nothing in the input says so. Filling in the context field with the audience and the occasion is what pushes the output from generic to specific, and even then the causal claims stay yours to add.

What metric formats does the parser actually understand?

The shape it was built for is a name, a separator, a value, and an optional bracketed movement — "Gross margin: 71% (+200bps)" or "Europe share: 38% of new signups (+6pp)". Values with currency symbols, units and thousands separators are read; the movement in brackets is the part that determines whether the angle is positive or negative. A line with no movement and no comparator still produces a value entry, but the narrative angles have nothing to say about direction.

Can I use the output in a report without rewriting it?

The objective angle is the safest to use close to verbatim because it makes no judgement. The narrative and recommended angles take positions, and a position that came from a parser rather than from your analysis is worth a read before it goes into something with your name on it. Treat all three as drafts with different registers.

Is anything about my company's figures sent anywhere?

Without a provider connected, no. The parse and the phrasing are local JavaScript, and the metric lines never leave the page. If you connect an AI provider and press Generate again, the text you pasted is what gets sent to that provider — which is worth knowing before pasting next quarter's unpublished revenue into a tab. The choice is yours at each step, and the tool works either way.

Last verified: October 3, 2026

Launch Data Insight Writer

All Tools · AI Tools · About · How We Test · Contact · Privacy

Maintained by Huang Bin · Last updated October 3, 2026