Worked example. This is an illustrative worked example, not a client result. The services firm and every number in it are assumptions. It shows AI expense report review: software that reads every submitted report against your written t&e policy and flags the ones that break it before an approver signs off. A person still approves every payment.
Picture a professional services firm asking one question: how much of what we reimburse each month breaks our own policy? The design runs AI expense review across every submitted report, so out-of-policy spend gets caught before approval, not clawed back after payment. The share that breaks policy is the number to measure, and this post does not claim one.
#Why catch out-of-policy spend before approval instead of after?
Clawback is the expensive way to enforce a policy. Once a reimbursement is paid, getting it back means an awkward conversation with a colleague, a manual reversal in the ledger, and a note in next month's close. Catching the same expense before approval costs one click. The approver sees the flag, the reason, and the exact policy line it broke, then decides.
Does AI for expense report review replace the approver?
No. The AI reads every expense report against your t&e policy and flags the ones that break it. An approver still reviews and signs off on each flag, and every decision lands in an audit trail. The AI does the first read; a person makes the call.
#What does the design consist of, and on top of what?
Nothing is replaced. Assume the firm already runs Expensify for submission and NetSuite for the ledger. The design adds a review step between them. Every report gets read against the written t&e policy, the ones that break it get flagged with a reason, and a human approval queue holds each flag until someone signs off. Nothing auto-approves. This is the kind of internal step our AI consulting team scopes, and the review tool itself is the kind of thing our AI implementation work ships.
We are
Stennir builds AI that flags out-of-policy expenses for a human to approve, on top of the Expensify and NetSuite you already run.
We aren't
It is not an auto-approval bot that pays out reimbursements without a person reading the flag first.
#What breaks policy most often?
- Meals booked over the per-diem cap
- Missing receipts above the receipt threshold
- Personal items mixed into a hotel folio
- Travel booked outside the approved window
- The same expense submitted twice across two reports
None of these are fraud. Most are honest mistakes. The point of AI expense review is to catch them quietly, at submission, before they become a clawback or an audit finding.
#What would the finance team measure?
Assume two approvers read every line of every report by hand today. The AI does the first read, so they only look closely at the flagged reports. The things to measure are approver hours each month, the share of reports flagged, the share of flags that were real, and time from submission to sign-off, each against a baseline taken before launch. No result is claimed here.
There is no client quote here, because there is no client. The aim in one line: find out about broken policy at submission instead of at close.
One disclosure. This is an illustrative worked example. There is no client, no engagement and no measured result behind it, and nothing in it is rounded from real data. We would rather show you a clearly labelled assumption than a made-up precise number.
If reimbursement leakage and slow approver review are quietly costing your finance team, that is a use case we design. We scope it against the stack you already run, keep a person on every approval, and hand you the audit trail. More worked examples like this one live in the journal. When you're ready, tell us what you're building.
