Turning Excel Tables Into Narrative: AI for KPI & Management Dashboards

By Christian Matumona, Kyootek AI Financial Controller

A KPI dashboard is built to be scanned, not read — rows of metrics, trend arrows, conditional formatting. That's exactly right for the controller who built it and knows what every number means. It's much less useful for the CFO, the board member, or the department head who needs the three sentences that explain what the dashboard is actually saying this month. Writing those three sentences, every period, for every dashboard that matters, is where AI removes real time.

This is the deep dive on KPI and dashboard narrative referenced in the Complete Guide to AI for Financial Controllers.

The translation problem

Most finance functions maintain more dashboards than they have time to properly narrate. A metrics table gets built once, refreshed monthly, and shared as-is — readers are left to interpret trend lines and conditional-formatting colors themselves. The dashboard is accurate. It's just not communicating.

AI is well suited to this specific translation task: given a structured table with current values, targets, and trend, it can draft plain-language commentary describing what's on track, what's not, and what changed most since last period — consistently, every cycle, without the controller writing it from scratch each time.

What to feed it

The quality of dashboard narrative depends almost entirely on giving AI three things: the metric values, the target or expected range for each metric, and at least one period of trend. Without a target, AI can describe direction but not whether that direction is good.

You are translating a KPI dashboard into management commentary.

KPI table (Metric | This Period | Last Period | Target | YTD Trend):
[PASTE YOUR TABLE]

Write narrative commentary with this structure:
1. Headline — the single most notable movement this period, one sentence
2. On track — bullet list of metrics performing at or above target
3. Needs attention — bullet list of metrics below target or trending negatively, each with the size of the gap
4. Unchanged — one sentence noting any metric flat versus last period, if relevant

Constraints: total length 150-200 words. State only what the data shows — do not infer causes not present in the table. Use plain language a non-finance reader would understand; avoid financial jargon unless the metric name requires it.

The "needs attention" section is usually where the actual value sits — a dashboard scanned quickly can hide a metric that's been drifting below target for three periods in a row. Asking AI to state the size of the gap explicitly, rather than just flagging direction, surfaces that pattern clearly.

Working inside Excel directly

If your KPI dashboard lives in Excel — and most do — Claude in Excel or Copilot can read the live table without an export step. The same prompt structure works; the practical advantage is that the dashboard and the narrative stay in the same file, updating from the same source data each refresh, which removes the risk of the narrative referencing a stale export.

This is one of the cleaner illustrations of why Module 4 of the AI for Controllers course treats Claude-in-Excel as a distinct skill from general AI prompting — the workflow is genuinely faster when the model can see the spreadsheet directly rather than working from a copy-pasted table.

Where this goes wrong

No target, no judgment. A table with values but no target will produce commentary that accurately states direction and confidently misjudges whether that direction matters. Always include the target or expected range.

Too many metrics at once. A dashboard with forty rows produces a narrative nobody reads. Group into a primary set of eight to twelve metrics that actually drive decisions, and narrate those; the rest can stay as reference data in the table without commentary.

Treating the narrative as final. The draft accurately reflects the data. Whether a metric's movement is actually concerning, given context the table doesn't capture — a known seasonal pattern, a planned one-off — is still your call before the commentary goes to its audience.

A small habit that compounds

Once a KPI narrative prompt is calibrated to a specific dashboard, it produces a consistent first draft every period with the same structure and reliability — freeing the controller to spend review time on the metrics that actually need attention rather than re-drafting the ones that don't.

That's a small, recurring example of the broader shift described in What Is an AI-Augmented Financial Controller?: less time assembling the report, more time deciding what it means.