FP&A After the Reset: Finance on Sovereign Ground

Forecasting, variance analysis, scenario planning, and the long tail of management reporting are exactly the structured, reasoning-heavy work that AI now does well. After the Reset, a finance team can compress days of modelling into minutes. The temptation is to bolt the nearest frontier tool onto the close and move on.

Finance is also the function least able to tolerate an unexplainable answer. Every figure that reaches a board, a regulator, or an auditor must be traceable. The job, then, is to capture the speed without surrendering the lineage, which is a placement question before it is a tooling one.

At the sovereign end: statutory and regulatory reporting, forecasting and scenario planning on actual financials, and management reporting over sensitive internal numbers, run on self-hosted models so the data and the audit trail stay inside the perimeter. At the frontier edge: macro and market research over public data, and first drafts of narrative that contain no sensitive figures, where capability pays and nothing confidential is exposed. The split lands around sixty per cent sovereign.

Databricks' and Snowflake's finance and data reporting both track the same consolidation, financial workloads moving onto governed, access-controlled platforms rather than spreadsheets and external tools. The warehouse, governed, is where AI-assisted finance should read from and write to.

An AI-assisted number is only as good as the lineage behind it: which data, which model, which version, which human signed off. That evidence is straightforward to produce when the model and the data live inside your governed environment, and fragile when the reasoning happened behind a third-country API you cannot inspect. The CSRD sustainability-reporting regime and the EU AI Act both push the same way, toward documented, controllable, auditable systems.

This is finance's version of the sovereignty thesis: the architecture that keeps the auditor happy is the same one that keeps the data sovereign. Control is not a constraint on AI in finance; it is the precondition for using it on anything that matters.

Inventory where AI is already touching financial numbers, including the shadow spreadsheets and personal tools, and classify by sensitivity and reporting impact. Move anything that feeds statutory or board reporting to sovereign infrastructure with full lineage.

Pilot one high-value sovereign workload, forecasting or variance analysis is the cleanest, prove the lineage holds, and keep the frontier model to public-data research. Then make placement and lineage standing requirements, so every AI-assisted number is auditable by default.