Your dashboard shows what happened. This tells you why.
Most reporting stops at the chart. The AI insight layer reads your unified data every week, compares it against what's normal, and writes the part a human analyst would spend an hour on — what changed, why it matters, and what to do — ranked by dollar impact.
Example output · illustrative data · runs in your own accounts
You'll get the most out of this if…
- You have dashboards but nobody has time to interpret them.
- Problems get spotted weeks after they start costing money.
- Your team asks "why did that move?" and the answer takes an hour to find.
- You're an agency writing the same commentary for a dozen clients every month.
A chart is a question. Somebody still has to answer it.
Reporting projects usually end at the dashboard, on the assumption that once the numbers are visible the interpretation takes care of itself. It doesn't. Somebody has to notice that CAC moved, work out which campaigns caused it, decide whether it matters, and say what to do — and that somebody is usually busy.
So the dashboard gets checked on Monday, the anomaly gets noticed three weeks later, and the money in between is gone.
This layer does the reading. It compares every metric against its own baseline, finds what genuinely moved rather than what merely wobbled, attaches a dollar figure to each finding, and writes it in plain English. You get the analysis, not just the chart.
Delivered, not described.
A weekly written note
Plain English, no jargon, in your inbox or Slack. Usually three to five findings, never a wall of text.
Ranked by dollar impact
The most expensive thing is first. Not the biggest percentage move — the one costing the most money.
Anomaly detection with baselines
Each metric is compared against its own seasonality, so a normal Monday dip doesn't get flagged as a crisis.
A recommended action per finding
Not "CAC is up" but "pause these two campaigns, they're below break-even". Specific enough to act on.
Every figure traceable
Each number links back to the view it came from, so nobody has to take the AI's word for it.
What actually happens.
Baseline
The model learns what normal looks like for your business, including seasonality.
Configure
We agree which metrics matter and what a material change looks like for each.
Draft
The first few weeks are reviewed together so the tone and threshold are right.
Run
It writes itself every week. You read it, or forward it, or ignore it — it's still there.
Asked on nearly every call.
Is this just ChatGPT reading a spreadsheet?
No. It reads your modelled warehouse data, not exports, and it compares against stored baselines rather than guessing from a single snapshot. The model writes the language; the analysis comes from the data layer underneath it.
What if it says something wrong?
Every figure links to the view it came from, so it's checkable in one click. In practice the failure mode is flagging something unimportant, not inventing something false — and the thresholds get tuned during the first few weeks.
Do we need the warehouse first?
Yes. This layer reads modelled data. Pointed at raw platform exports it would inherit exactly the disagreement the warehouse exists to fix.
Can it write in our own voice?
Yes, and agencies usually want that. The tone is configurable, and the output is yours to edit before it reaches a client.
Want this on your own numbers?
Fifteen minutes, no pitch. Bring the figure you least trust and I'll tell you what's likely behind it.
Book a 15-minute call