The last mile of reporting
Most organizations have invested heavily in producing financial numbers: ledgers are reconciled, cubes are built, dashboards are published. Yet every month, analysts still spend days writing the same kind of narrative — why revenue moved, which costs drove the variance, what changed against budget.
That narrative is where decisions are actually made. A variance of +6.7% means little until someone explains that personnel and technology costs were the primary drivers, and whether that was planned.
Why generic AI falls short
Pointing a general-purpose language model at a spreadsheet produces fluent text, but not reliable explanation. Without an understanding of account hierarchies, allocation rules and the difference between actual, budget and forecast, the model is guessing.
- It cannot distinguish a timing difference from a structural one.
- It has no view of which dimensions matter to the business.
- It cannot show where a statement came from.
Explanation built on a financial model
The approach we take is to let AI reason over a governed financial model rather than raw data. Drivers are calculated first — deterministically — and the language layer describes what the model has already established.
The result is commentary that is consistent from month to month, grounded in definitions finance already trusts, and traceable back to the underlying figures.
Where to start
Pick one recurring report with a well-understood structure — a monthly management pack or a cost-centre review. Automate the preparation, then introduce AI commentary alongside the analyst rather than instead of them. Measure the time saved and, just as importantly, how often the commentary is accepted without edits.