Ask the follow-up. See every step it took.
Wisdom is the conversational side of DataGenie: ask it anything about the business in plain English. It understands a multi-stage question, factors in your domain knowledge, and builds an execution plan before it answers.
Text-to-SQL. A coordinator that picks its own tools, in its own order, and shows you the working.
It does not just answer. It plans.
A coordinator that sequences its own services, skills you write once in plain English, and the forecasting and scenario work that most conversational BI stops short of.
Tailored to your industry
Every figure below is drawn from that industry’s own pack — the same numbers its briefs and notebooks carry.
It writes the plan before it answers.
A master coordinator agent selects and sequences these five services dynamically, and the plan branches — so a different result at step two sends it somewhere else entirely.
Most conversational BI runs one fixed pipeline for every question. This writes a plan per question, and the steps it took come back attached to the answer — so you can see why it went where it went.
- A recurring ad-hoc question gets packaged as a Skill, so its plan stops being rewritten
- Every service returns aggregates — no raw rows cross to the model
- Metric Servicetime series for any KPI × dimension, with anomaly flags
- Insights Servicetop stories for a KPI, dimension and period
- Contribution Analysiswhich sub-dimensions drove the deviation
- Forecasting Servicefuture-period forecasts for any metric
- Scenario Planningwhat-if: simulate a change against dependent KPIs
One service’s result chooses the next — it is not a fixed pipeline.
What happens if you do nothing.
Wisdom projects the metric forward, then puts the options side by side — against every dependent KPI, so the second-order cost shows up too.
The forecast is available for any metric, down to a single KPI × dimension slice — not only the headline number somebody once built a dashboard for.
- Scenario planning prices the change you are considering, not a generic one
- The second-order cost is the point: what the fix does to everything downstream
last 5 weeksnow next 4
The next best action, scored.
Every option is priced off the same forecast — so the recommendation and the arithmetic behind it are one object, not two.
Prescriptive work usually arrives as a separate deck: a recommendation here, the arithmetic somewhere else, and a week in between while the two are reconciled.
- Doing nothing is priced too — it is rarely free
- Every figure traces back to the forecast that produced it
- Reprice the four worst lanes Recovers most of the gap, holds volume +$310K
- Renegotiate Northeast freight Slower, but structural+$180K
- Do nothing The gap compounds through the quarter −$420K
Business logic you write once.
Packaged, reusable, defined in plain English and invoked with arguments. The recurring question stops being asked from scratch.
It is how an ad-hoc question becomes a standing one. The Margin Risk Review beside this was written once; it runs every month against a different account, window and threshold.
- Invoked in plain English: “show top 20 margin risks for Costco”
- It replaces the standing scan nobody finishes
Margin Risk Review
Rank customer × material lanes whose contribution margin fell more than the threshold against their baseline, and split the cause across price, freight and rebates.
- Takes
- window · threshold · account
- Returns
- a ranked table, worst lane first
Run monthly with new arguments. It replaces a scan nobody finishes.
It connects the two on its own.
An internal and external timeline, so anomalies arrive explained in context — not merely measured.
Anything orthogonal that could affect the analysis but is not in the onboarded tables belongs here — a supplier outage, a pricing change, a competitor launch. Wisdom joins the event to the anomaly itself, so nobody has to remember it was that fortnight.
- The timeline is yours to keep — internal launches as well as outside shocks
- One entry can explain a story the data alone would call unexplained
- Reefer truck shortage, Northeast Feb 1–14 · external
- Freight cost per case spikes Feb 3 · in the data
- Margin dips in those lanes Feb 6 · the anomaly
Which slices own the move.
Correlation, change and KPI attribution — ranked by how much of the deviation each one owns, not a list of everything that moved.
Contribution analysis keeps going down through sub-dimensions until it reaches the cell that carries the move. The answer is four rows that add up, not forty that do not.
- Correlation, change and KPI attribution run together, not as three reports
- The share is the ranking, so the read is already in priority order
Describe the board. Don’t build it.
Ask for the board and it builds itself — the charts, the layout and the refresh, from the same governed numbers every other answer uses.
There is nothing to place and nothing to wire. You describe the board the way you would describe it to an analyst, and it arrives on a schedule like every other pack output.
- Any board can be described this way — there is no template list to pick from
- The numbers are the ones the briefs and the notebook already use
> build a daily deposit-health board
Net flow · 12 weeks
Outflow by branch
- Branch 0741%
- Branch 1221%
- Branch 0413%
- All others25%
A question in. A plan it wrote itself.
Plain English, no SQL.
Coordinator · 4 of 4
- ✓Metric Service — CM% by customer × material, 13-week baseline
- ✓Insights Service — the lanes that broke the baseline
- ✓Contribution Analysis — decompose price, cost, freight, rebates
- ✓Notebook — nutshell, chart, numbers, meaning
With every step attached.
Twenty lanes cleared the 10% threshold, and they are not evenly weighted. Costco alone owns 38% of the deviation — driven by freight per case, not by price. The top four lanes hold most of the recoverable money.
4 services · 1 notebook cell · every step recorded
Run it again — nothing above was configured. The coordinator chose those services, in that order, for that question — and the plan branches, so a different answer at step two would have sent it somewhere else. Ask the same question again and you get the same steps, and the same numbers.
The insights must find you.
Thirty minutes, your data, a real brief at the end of it.