Self-Serve Finance

The AI doesn’t do the math. It decides what math to do.

Anyone can ask, in plain language. The AI works out which calculation answers the question; a deterministic engine runs it against your agreed definitions and your guardrails. Ask the same thing next year and you get the same number.

Question
AI interprets
Engine runs math
Answer

The engine computes, not the model

No number in an answer was written by a language model. The AI selects and parameterises the calculation; the engine runs it and returns the result.

Your rules are part of the math

Agreed definitions, guardrails and what a given person may see are applied when the calculation is built, not reviewed afterwards. No phrasing gets around them.

Every answer shows its working

Which definition, which inputs, which assumptions, as of when. One click from the number, and it travels with the answer when someone forwards it.

Reproducible and auditable

Re-run the question and get the same figure. Every question, answer and derivation is kept, attributable to a named person. Nothing rests on trusting the AI.

Defensible in the meeting where it gets quoted

A number is only useful if it survives being challenged. Every answer carries the definition it used, the inputs it read and the moment it was computed, so the challenge is answered from the answer itself.

−1.2pton group gross margin
Definition
Gross margin v4, owned by group control
Inputs
List price, 5% discount, committed volume, cost to serve
Computed
26 August 09:14, against model v19
Self-Serve Finance

Everyone asks the same governed model.

Bring the question your teams argue about most — from any of them, in their own words — then open the working. That is the whole evaluation.