runbookify
← All plans
Accounting & Finance / Audit, Controls & Compliance

Journal Entry Anomaly Detector: Score the Riskiest JEs for Review

Import a journal-entry export, let AI score every entry for red flags with plain-English reasons, rank the riskiest, and have a reviewer disposition each one — without ever touching the GL.

IntermediateA weekendBuilds onNext.jsSupabaseResend
What you'll build

A web tool where you import a journal-entry export, AI scores each entry for anomaly red flags and explains why, the riskiest are ranked for review, your reviewer records a conclusion (cleared / needs explanation / escalate) on each, and it exports a JE-testing workpaper plus an exceptions log — a detective tool that never alters the ledger.

Gated download

Enter your email — the plan downloads instantly and a copy lands in your inbox.

By submitting your email you'll also receive the weekly runbookify newsletter. You can unsubscribe at any time.

Before you start

  • A Supabase account (free)
  • A Vercel account (free)
  • A Resend account (free)
  • A CSV export of posted journal entries (entry, accounts, amount, preparer, post date, period)
  • Claude Code or any AI coding agent

The problem this kills

Somewhere in this period's general ledger sits a handful of journal entries that deserve a second look. A round-dollar top-side entry booked on a Saturday. A reclass to revenue posted in the final hours of the quarter. A manual entry from a preparer who almost never touches that account. A rare debit-credit pairing that nobody can explain off the top of their head. Each one might be perfectly fine — or it might be the thread you pull on during a SOX walkthrough or an audit.

The trouble is volume. A real GL has thousands of entries a period, and the risky ones don't announce themselves. Eyeballing them all is impossible, so most teams sample a few, hope they picked the right ones, and move on. Meanwhile the textbook fraud and error patterns — round numbers, odd timing, unusual preparers, entries to cash and revenue near period close — are sitting right there in the export, fully detectable. You just need something that scores every entry against those patterns and floats the riskiest to the top. You do not need to be a developer to build it.

What you'll build

A simple internal web tool for controllers and internal audit. You import a journal-entry export (entry ID, accounts, amount, preparer, post date, period). The tool scores every entry against a set of configurable anomaly rules — round-dollar amounts, entries posted on weekends or holidays, unusual or rare preparers for an account, rare account pairings, and entries hitting cash or revenue close to period end — and gives each a risk score with a plain-English reason for every flag. The riskiest entries rank at the top of a review queue. Your reviewer opens each flagged entry and records a conclusion: cleared, needs explanation, or escalate — nothing in the ledger is ever changed. When you're done, the tool exports a JE-testing workpaper (every tested entry with its flags, score, and disposition) and a separate exceptions log of everything that needs explanation or escalation.

What's inside the Implementation Plan

The downloadable plan is a step-by-step file you paste into an AI coding agent. It opens by interviewing you about your business — your current JE-testing process, the system your GL lives in, the exact columns and naming in your export, your chart-of-accounts conventions for cash and revenue, your period-close calendar, your typical and peak entry volumes, and which red flags actually matter in your environment. It reads a short spec back to you for a thumbs-up, then builds the tool around your answers instead of a generic template. From there it walks the agent through the data model, the CSV import with a duplicate guard, the configurable scoring engine (with rule thresholds you control), the ranked review queue, the disposition gate, and the workpaper and exceptions-log exports. Every step ends with a ready-to-copy prompt.

The governance it includes (this is the point)

This isn't a toy, and it isn't a black box. The plan builds in the controls a real finance and audit function needs: login so only your team can use it, row-level security so people only see their own organization's entries, a complete audit trail of every disposition and reviewer note (who, what, when, and why), a hard human-review gate so every flagged entry carries a person's documented conclusion, and duplicate guards so the same export can't be imported and scored twice. Just as important: this is a detective tool by design — it reads the ledger and scores it, and it never writes back to or alters the GL. The AI raises the suspicion and explains it; a person reaches the conclusion and signs it.

Who it's for

Controllers, internal auditors, and accounting leads who run JE testing or continuous monitoring and are tired of sampling blind. If you can describe what makes a journal entry "worth a look" in your world — your round-dollar threshold, your close calendar, your sensitive accounts — you can build this.

You've got this — open the plan, paste the first prompt, and you'll be scoring your first real GL export this weekend.

Gated download

Enter your email — the plan downloads instantly and a copy lands in your inbox.

By submitting your email you'll also receive the weekly runbookify newsletter. You can unsubscribe at any time.