Deterministic facts
Figures are computed, never generated.
The analyst decides what to work out; the numbers come from deterministic engines — the KPI engine, queries on the knowledge graph, established statistics libraries. Every figure in an answer must match a result returned for that question, so the same question on the same data gives the same number.
- 14.6%Attrition, Sales, Q3Attrition rate · saved definitionKPI engine
- 31Leavers in the quarterkept with the answerGraph query
- +2.8 ptsAgainst Q2period comparisonKPI engine
- p = 0.004Exit reason × departmentchi-square, Holm-correctedStatistics
Deterministic facts, part by part.
- 01
The KPI engine
KPIs are calculated from their definitions by the platform's KPI engine — the same calculation every time, for every reader.
- 02
Queries on the graph
Counts, rates and totals run as queries on the knowledge graph, and the query is kept with the answer.
- 03
Checked before it is shown
Every figure is checked against what the tools returned — a result, a rounding of one, the reader's own number or a figure code worked out from them; a sentence with one that fails is taken out.
- 04
Library statistics
Tests take their statistics from scipy and statsmodels; forecasts and models run on AutoGluon.
Plan, Compute, Check, Answer.
- Step 1
Plan
The analyst decides what has to be worked out.
- Step 2
Compute
The engines work it out from your data.
- Step 3
Check
Every figure is matched to its result.
- Step 4
Answer
The figures, with the working kept.
An engine behind every figure
- KPI values
- The KPI engine, from the KPI's saved definition.
- Counts, rates, totals
- A query on the knowledge graph, kept with the answer.
- Changes over a period
- The KPI engine's period comparison.
- Tests and associations
- Statistics libraries, with corrections when many tests run at once.
- Forecasts and predictions
- AutoGluon models, tested on data they did not see.
- A new KPI's numbers
- A dry run of the KPI engine — nothing saved until you say so.
Questions it answers, in your words.
- “What is our attrition rate this quarter, by department?”
- “How did headcount change since last month?”
- “Work out this KPI again on today's data.”
On the same model and graph
- AI analystAsk in plain English. It plans the work, runs it on the graph and shows its working.
- Auditability“How we got this?” and “Can I trust this?” — on every answer, chart, forecast and model.
- Statistics & driversTests, associations and contributions from established libraries — corrected when many tests run at once.
Your next people decision, with the why attached.
Request a demo. We'll email you a two-minute walkthrough today, then set up a time that suits you.