AI Agents for Expense
Policy Checks
Rule breaking claims are rare, accidental and cheap. The expensive ones comply with every written rule.
- Rule based expense checking catches the rare, accidental and cheap claims. The expensive population complies with every written rule.
- A claim split into amounts under the approval limit has valid receipts for each part and breaks nothing. It is only visible as a pattern.
- Genuine expense duplicates rarely match on amount and date, because the two claims came from different documents describing the same spend.
- The same meal claimed by two attendees is two honest claims and one meal.
- Expense coding errors all net to the correct total, which is exactly why a review that starts at the total never finds them.
AI agents for expense policy test behaviour across a claimant's history rather than each claim against the rulebook, match duplicates on the underlying event and the card statement, and validate department, branch and account coding. ChatFin reports the claims that comply with policy and should still be questioned.
The framing matters because rule enforcement is close to solved and it was never where the cost was. A claim over the limit gets caught. A claim that sits exactly at the limit every single time, or is split into two amounts beneath it, or appears on a company card as well as in a claim, breaks nothing at all. ChatFin looks for the pattern.
The claims worth questioning are the compliant ones
Expense policy enforcement is built to catch claims that break a rule. Those are rare, usually accidental, and cheap. The expensive population is claims that comply with every written rule and should still have been questioned.
| Claim | Breaks a rule | Should be questioned |
|---|---|---|
| Over the per night limit | Yes | Yes, and it is caught |
| Missing a receipt | Yes | Yes, and it is caught |
| Split into amounts under the limit | No | Yes |
| At the limit, every single time | No | Yes |
| Within policy, far above peer norm | No | Yes |
| Claimed and also on a company card | No | Yes |
Four of the six comply. Every rule based system passes them and every one of them is a pattern rather than an event, which is why they are invisible one claim at a time. ChatFin tests behaviour across claims rather than each claim against the rulebook.
ChatFin tests claim behaviour across a claimant's history rather than each claim against the rulebook, which is where threshold splitting becomes visible.
Duplicate expenses do not look like duplicates
Duplicate detection on expenses usually means matching amount and date. Genuine duplicates in this population rarely present that way, because the two claims came from different documents describing the same spend.
ChatFin matches on the underlying event, the merchant, the date range and the attendees, rather than on amount and date, and it reads the card statement alongside the claims.
ChatFin matches expense duplicates on the underlying event and the card statement rather than on amount and date, which is what catches two attendees claiming once.
Expense coding is wrong in a way that never gets challenged
The general ledger account, department and branch on a claim are chosen by the claimant, who has no reason to care and no way to know. The result is a cost base that is wrong at line level and correct in total, which nothing will surface.
| Coding error | Effect | Why it survives |
|---|---|---|
| Right account, wrong department | Departmental costs are wrong | The total is right |
| Everything to a default category | Analysis is meaningless | Nothing is missing |
| Wrong branch or entity | Entity results are wrong | It consolidates out |
| Project not recorded | Project margin is wrong | The project field is optional |
| Capital treated as expense | Profit is understated | Below any threshold |
Every row nets to the correct total, which is the reason none of it is ever caught by a review that starts at the total. ChatFin validates coding against the claimant's own history, their department and the merchant, and reports the population that disagrees.
ChatFin validates department, branch and account against the claimant's history and the merchant, because every one of these errors nets to a correct total.
Where this underperforms
Three.
Deeper in expense policy
The use cases this hub covers.
Worked on this page
Expense exception workflows · Duplicate expense detection · GL, department and branch validation
Covered as sections here
Compliant but questionable · Event matching · Card reconciliation · Coding validation
Questions finance teams ask first
If a claim is within policy, why question it?
Because policy sets a ceiling, not a norm. A claim at the limit every single time, one split into two amounts under the approval threshold, or one far above what peers spend on the same route all comply completely. None is caught by a rule and each is visible as a pattern across claims, which is what ChatFin tests.
Why does duplicate detection miss expenses?
Because it matches amount and date, and real expense duplicates do not present that way. The same meal claimed by two attendees, a card transaction alongside a cash claim differing by a tip, or a hotel folio claimed in full and itemised. ChatFin matches on the underlying event and reads the card statement alongside the claims.
Does expense coding really matter if the total is right?
It matters everywhere except the total. Right account and wrong department makes departmental costs wrong. Everything defaulted to one category makes analysis meaningless. A missing project code makes project margin wrong. All of them consolidate to a correct figure, which is why nothing surfaces them. ChatFin validates against history and merchant.
Test the pattern, not the claim.
ChatFin tests claim behaviour across each claimant's history, which is where threshold splitting and persistent limit claiming become visible.
Duplicates are matched on the underlying event and against card statements, and coding is validated against history and merchant.
Bring a year of claims and your card statements.
Book a demoAll images are served from the Pexels content network. No third party product screenshots, vendor logos or licensed media are used on this page.
Bring a year of expense claims.
We will test them through ChatFin against your card data and each claimant's own history and show you the compliant claims worth a question.