Provenance is the product.
Every vendor now claims accuracy. Very few can show their work. In high-stakes data, the difference between a demo and a system is whether a result can be traced back to its source — and defended a year later.
The market is shifting from “how accurate is your model?” to “can you prove this result?” The firms that can answer the second question will win the work that matters: the close, the regulatory submission, the resource statement. Accuracy is necessary. Provability is the moat.
The close is a controls problem before it is an AI problem.
Automation promises to compress the month-end close. But the close is not slow because matching is hard; it is slow because controls are manual and evidence is assembled at the end.
A model that reconciles faster but cannot explain a match does not shorten the close — it moves the work to audit. The gain only appears when provenance is designed in from the first day: source lineage, model rationale, human sign-off. Controls first; speed follows.
Explainability is not a feature you add at the end.
In regulated domains — clinical data, genomics, resource estimates — a result is only useful if it survives review. Explainability bolted on after the fact is a validation burden, not a benefit.
We design traceability into the model, the data and the workflow, so the explanation is a by-product of how the system works. That is a different engineering decision, made at the start — and it is where most AI projects in regulated industries quietly fail.