AI Budget Copilot for FP&A

Inspiration

What it does

How we built it

AI Budget Copilot for FP&A

Finance teams do not need another dashboard that merely reports a variance. They need a fast, defensible answer to: What changed, why did it change, and what should we do next?

AI Budget Copilot turns a budget variance into an evidence-backed investigation workflow for FP&A teams.

What it does

A finance partner selects a cost center and asks a plain-English question, such as: “Why is HKG over budget?”

GPT-5.6 then investigates using focused finance tools rather than guessing:

• get_budget_variance identifies the material variance.
• find_cost_drivers isolates the largest contributing records.
• get_budget_records retrieves the supporting transactions.
• check_budget_controls verifies relevant approval and control context.

The result is a traceable investigation with source evidence, a clear financial narrative, and an Action Packet that can be reviewed before execution.

Why it matters

Variance analysis is often slow because the evidence is spread across budget plans, actuals, transactions, and approval controls. This product makes the AI behave like a disciplined finance partner: it gathers evidence first, explains its reasoning, and proposes an action only when the supporting data is available.

How we built it

We built the dashboard and investigation workflow with Codex. GPT-5.6 is the investigation engine, orchestrating the finance-tool sequence and synthesizing the returned evidence into a concise recommendation.

The experience also guards against a subtle but important real-world failure mode: users may switch cost centers or edit a question while an investigation is still running. Request sequencing and local snapshot validation prevent stale asynchronous responses from overwriting the current analysis.

What we learned

The most useful AI experience for FP&A is not a generic chat response. It is an auditable decision workflow: visible tool calls, evidence tied to a recommendation, and a clear human confirmation step before action.

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for AI Budget Copilot for FP&A

Built With

• codex
• finance
• gpt-5.6
• react
• tailwind-css
• typescript
• vite