Receipt Policy Checker
Build an internal tool that checks every line of a submitted expense report against your written spend policy and hands reviewers a clear, cited flag list to disposition before approval.
Import an expense report, evaluate every line against your real policy, and get a flag list with the exact rule cited - so reviewers disposition each exception and return a cleaned report plus an exceptions log before approval.
Before you start
- A free Supabase account
- A free Vercel account
- A free Resend account
- A sample expense report (CSV or Google Sheet) and your written spend policy
The problem this kills
Expense review is death by judgment calls. One reviewer waves through a $14 airport beer, another rejects it. Someone misses that three "client dinners" all landed on the same Saturday. A $400 charge slips by with no receipt because nobody remembered the threshold was $75. By the time finance notices the pattern, the report is already approved and the money is out the door.
The policy exists - it's written down somewhere - but enforcing it line by line, the same way every time, is tedious work that humans do inconsistently when they're busy. So policy becomes a vibe instead of a rule, and reimbursements drift.
What you'll build
A small web app where a reviewer imports an expense report and instantly sees every line checked against your actual policy: over per-item or daily limits, missing receipt above your threshold, disallowed categories (alcohol, first-class airfare, whatever you forbid), out-of-window dates, weekend or personal-looking spend. Each flag names the exact rule that fired, in plain language, so nobody has to argue from memory.
Crucially, the tool advises - it never auto-denies. The reviewer dispositions each flag (allow with a note, reject, or request more info), and the app returns a cleaned report plus an exceptions log to the submitter and approver. Consistent enforcement, with a human still making every call.
What's inside the Implementation Plan
The plan opens by interviewing you about your business - your actual spend policy, the limits and thresholds, the categories you disallow, the date windows you enforce, and the messy exceptions you live with - so the tool encodes your rules, not a generic template. It reads a short tailored spec back to you and waits for your thumbs-up before building anything.
From there it walks you, prompt by prompt, through:
- Importing an expense report from CSV or a Google Sheet (no integration required to start).
- A rule engine that evaluates each line and cites which policy rule fired and why.
- A reviewer screen to disposition every flag - allow with note, reject, or request info.
- Login, per-organization data isolation, a full audit trail, and a duplicate guard.
- A cleaned-report and exceptions-log export in the exact columns your system expects.
Every build step ends with a ready-to-copy prompt you paste into your AI agent.
The governance it includes (this is the point)
This isn't a toy. The plan builds in the controls finance actually needs:
- Login so only your team can use the tool.
- Row-level security so reviewers only ever see their own organization's reports.
- A complete audit trail - who dispositioned which flag, what they decided, and when.
- A human-in-the-loop gate - the AI flags and cites; a person decides every exception; nothing is auto-denied and nothing commits without review.
- Duplicate guards so the same report-and-line can't be processed twice.
Who it's for
Finance reviewers and approving managers who are tired of case-by-case judgment calls and want consistent, explainable policy enforcement - where the tool catches the violations and the human still owns the decision.
You've got this - paste the first prompt and let the interview tailor it to your policy.