Rule of the week: Round Number Claims

The moment this shows up

You're reconciling the director's expense claims for the month and you hit a run of them: $500, $450, $300, $250. No receipts attached, just a description field that says "client entertainment" or "office supplies" or, memorably, "misc." Nothing about any single one of them looks wrong. People do spend round amounts — a $500 flight voucher, a $50 taxi. But five or six round numbers in a row, all unreceipted, from the same person, in the same month, is a different pattern. It's the shape an estimate makes when someone can't be bothered finding the actual figure, or doesn't want to.

This isn't fraud detection in the dramatic sense. Most of the time it's someone reconstructing a month of spending from memory two weeks after the fact and rounding because rounding is faster than digging through their email for a PDF. But "usually innocent, occasionally not, and either way unverifiable" is exactly the kind of thing that should get a second look before it's coded and forgotten.

A worked example

Say a claim batch for August includes:

| Description | Amount | Receipt | |---|---|---| | Client lunch | $150.00 | No | | Taxi to airport | $80.00 | No | | Parking | $45.00 | Yes | | Stationery | $200.00 | No | | Client dinner | $250.00 | No | | Uber | $32.40 | Yes |

Four of six claims are round to the nearest $10 or $50, and none of those four has a receipt. The two claims with actual receipts ($45.00, $32.40) are the ones with realistic, non-round cents. That contrast is the tell: real expenses have odd cents because prices have odd cents. A person reconstructing spend from memory produces round numbers because memory rounds.

None of this proves the $250 dinner didn't happen. It might have. But you can't currently tell, and that's the point worth flagging.

How to find it manually

Export the expense claims for the period (Xero: Business → Expense claims, or your practice management tool's equivalent) with amount, description, receipt-attached flag, and claimant.

1. Filter to claims with no receipt attached. This is your working set — everything else is out of scope for this check.
2. Flag amounts divisible by 50 or 100 with no cents, or cents ending in .00 more generally. A quick way by hand: sort the no-receipt claims by amount and eyeball which ones look "designed" rather than priced.
3. Calculate a round ratio per claimant: (round, unreceipted claims) ÷ (total unreceipted claims) for the period. One round claim in twenty is noise. Six round claims out of eight is a pattern.
4. Cross-check timing. Round-number claims submitted in a batch, days or weeks after the fact, are more suspicious than one submitted same-day.

Honestly: this takes a genuine spreadsheet skill (pivot tables or at minimum sorting and conditional formatting) and real time — expect twenty to forty minutes per claimant per month if you're doing it properly, more if the claims aren't already exported in a clean format. It's also a check that decays fast: do it once at month-end and you'll catch this month's pattern, but you won't see it building across months unless you rerun it every time.

FAQ

Why are round number expense claims a red flag? Because genuine, receipted purchases almost always have odd cents — prices aren't round. A cluster of round-dollar claims with no receipt attached suggests the amount was estimated from memory rather than taken from a document, which means it hasn't been verified against anything.

Does a single round-number claim mean something is wrong? No. One $50 claim is unremarkable. The signal is the ratio — a claimant with a high proportion of round, unreceipted claims relative to their total claims — not any individual transaction.

How the rule does this

The automated version (EXP-002) can't see intent, so it will flag legitimate rounded reimbursements (a flat per diem, a fixed mileage rate) as often as it flags careless estimates, and it will miss a fabricated claim that happens to land on an odd amount. What it does is run the round-ratio calculation above across every claimant, every period, continuously, and surface the ones that cross a threshold, with severity varying by how extreme the ratio and the unreceipted total are.

Finanomaly is currently waitlisted.