Why thresholds invite splitting
Approval thresholds exist to put a second set of eyes on larger spend. A common structure has managers approve up to $500, directors up to $2,500, and anything above that go to a VP with procurement involved. Every threshold also creates an incentive to stay just under it.
Most splitting is not fraud. It is an employee who needs a laptop dock before a client visit and knows the director is on leave. The result is the same, though: the control you designed never fires. With over 80% of front-line managers spending under 30 seconds reviewing a report, a split that looks like two ordinary purchases will usually sail through.
Four common split patterns
Splits fall into a small number of shapes. Knowing the shapes makes them easier to spot in a card feed, because the signal is in the relationship between transactions rather than in any single one.
Each shape points to a different review question. Same-card splits ask why the employee avoided the next approver. Cross-card splits ask whether the attendee lists overlap. Recurring charges ask whether procurement should have issued a purchase order months ago.
- Same card, same merchant, same day: two charges that each sit just under a limit
- Different cards, same merchant, minutes apart: two cardholders each paying part of one bill
- Same merchant, consecutive days: an invoice paid in pieces across a weekend
- Recurring sub-threshold charges: weekly payments to a vendor for work that should be one purchase order
Reading MCC codes and descriptors
Merchant category codes tell you what kind of business took the payment. 5812 is restaurants, 5813 is bars, 7011 is lodging, 4121 is taxis and limousines, 4511 is airlines and 5734 is computer software stores. A pair of 5734 charges on the same day from the same descriptor is a different story than a pair of 5812 charges in two cities.
Descriptors and timestamps add the rest. Two charges at the same restaurant two minutes apart on two different cards almost always mean one bill split between colleagues. Two charges at a software store forty minutes apart on one card usually mean one order paid twice.
A worked example at Teague Outdoor Co.
Teague Outdoor Co. found two splits in a single month. In the first, a product manager charged $2,488.00 and $1,964.40 to a design software vendor 38 minutes apart, both under MCC 5734. Together they total $4,452.40, which required VP approval. Each charge alone needed only a director.
In the second, two regional sales managers paid $487.30 and $462.15 at the same steakhouse at 21:14 and 21:16 on the same evening. The combined $949.45 should have gone to a director, but each half fell under the $500 manager limit. The attendee list on both reports named the same nine people.
Neither case involved personal spend. Both were people avoiding a slow approval. The fix in the first case was a faster approval path for software renewals, which removed the reason to split.
Detection rules that do not drown you in false positives
A naive rule that flags every pair of charges at the same merchant will bury AP in hotel incidentals and airline tickets. Good rules combine a proximity test, a threshold test and a short list of exclusions.
Start conservative and widen the net only after you have reviewed a month of hits. A rule that produces three real findings a week will be trusted; one that produces forty, mostly noise, will be muted within a quarter.
- Group charges by normalized merchant within a 24 to 72 hour window
- Flag when each charge is within about 15% below a threshold and the sum exceeds it
- Check across cardholders who share a cost center or approver
- Exclude hotel pre-authorization and final settlement pairs
- Exclude airline tickets with different passenger names under 4511
- Exclude contracted recurring subscriptions with a matching purchase order
Fixing the policy, not just the report
When the same team splits repeatedly, the threshold is probably wrong for their work, or the approver is a bottleneck. Look at approval cycle times before tightening rules, because a tighter rule on a broken process just moves the workaround somewhere harder to see.
Phyto Nex AI groups related card charges across employees and shows finance the combined amount, the threshold it crossed and the approver it would have needed. Finance decides whether to route it for retroactive approval, ask the employees for context or close it as legitimate.
Rohan Mehta, Staff Engineer, Duplicate Detection
Rohan builds the matching systems that pair card transactions, receipts and reimbursement claims across employees and payment methods. Before Phyto Nex AI he worked on payments reconciliation at a card issuer.