Image: Shipment warehouse in action by U.S. Department of Agriculture, licensed under CC0 1.0.
An LTL invoice arrives $340 above the rate confirmation. The bill shows class 85; the quote was built on class 92.5. Your three-way match flags the variance, an analyst opens an exception, and four days later the charge is approved because the carrier attached a reweigh certificate and nobody has grounds to argue with it.
Nothing in that sequence tested whether the pallet measured what the carrier says it measured. The match established that two documents disagree. It cannot tell you which of them is wrong.
What the three legs were designed to test
Three-way matching is a purchasing control, and it works well on what it was built for. A purchase order records what you agreed to buy. A goods receipt confirms what showed up on the dock. The invoice asks for money. All three describe the same static object, a quantity of a thing at an agreed price, and the control holds because counting cartons settles the question.
Freight is not a thing. It is a service performed over hours or weeks, across facilities you do not control, by equipment you do not own, under conditions that keep changing after the price is agreed. Every leg of the match weakens when you point it at a freight bill.
- The order leg becomes a rate confirmation or a contract rate table. A contract rate is not an order. It is a conditional price list, and it does not say what this shipment should cost until you know how the shipment actually ran.
- The receipt leg becomes a proof of delivery. A POD establishes that something arrived and someone signed. It is silent on arrival time, dwell, a refused appointment, a second delivery attempt, or a trailer the facility could not work.
- The invoice leg carries charges that did not exist when the other two documents were written: detention, redelivery, reconsignment, reweigh, layover, limited access, driver assist.
The three documents describe three different moments. Reconciling them is not really a matching problem, because the fact that would settle the disagreement appears in none of them.
A variance is not an error
Matching produces two outcomes: the numbers agree, or they do not. Everything interesting happens in the second case, and the standard answer there is a tolerance. Anything under the threshold clears automatically, and anything over it goes to a person.
That threshold is doing work it cannot do. Set it tight and the exception queue fills with charges that are perfectly legitimate. Set it loose and consistent overbilling lives comfortably underneath it. Neither position separates a carrier billing four hours of detention that never happened from a carrier billing four hours your receiving dock genuinely caused. Both land as the same gap against the same rate confirmation.
Volume turns that setting into a load-bearing decision rather than a configuration detail. Cass Information Systems reports processing 35 million freight invoices worth $37 billion in 2025. That is one provider's book, not an industry total, but it gives a sense of the scale at which a threshold decision repeats.
Reclassification made the gap obvious
The clearest recent example sits in LTL. The NMFTA's Docket 2025-1 took effect on July 19, 2025, moving classification toward density as the primary factor. FreightWaves reported that 40 percent of all NMFC items fell within the scope of the review, amounting to more than 5,000 potential changes, and that the density scale expanded from 11 subprovisions to 13.
Consider what that does to a document comparison. A class on the invoice that differs from the class on the quote is now roughly as likely to reflect a correct reclassification as a carrier error. The match reports a mismatch in both cases, with identical confidence and identical usefulness. What resolves it is the reading the dimensioner and the scale took when the freight was picked up, which is to say a measurement rather than a document.
The fourth leg
The missing leg is the record of what actually happened to the load. Not the carrier's account of it, and not a signature at the end, but an independent trace: geofence crossings at origin and destination, trip stop times, the appointment that was booked against the one that was kept, EDI 214 status events, captured dimensions and weight, and the exceptions logged while the shipment was still moving.
Chase produces that trace as a by-product of running execution. Track and trace, exception detection and root cause analysis generate exactly what an invoice check needs three weeks later, provided somebody retains it and ties it to the load.
This is the distinction the three-legged version cannot draw. A rate that is wrong and a load that genuinely changed show up as the same number on the invoice. Only the execution record tells them apart, and without it an AP team is choosing between paying everything plausible and disputing everything unfamiliar.
| Question | Three-way match | With execution evidence |
|---|---|---|
| Does the line price the contract correctly | Yes | Yes |
| Did the delivery happen | POD signature | Geofence and trip stops |
| Did the billed accessorial event occur | No | Yes |
| Who caused the event | No | Attributed to a party |
| Is a class or weight change legitimate | Flagged as mismatch | Checked against capture at pickup |
| What happens to a small difference | Cleared by threshold | Explained or disputed |
What Penny does with the fourth leg
Penny audits every invoice before payment rather than sampling after it, and she treats a variance as the start of the question rather than the answer. She recomputes the charge from the clause governing that lane and equipment, pulls the execution record for the load, and attributes each difference to something specific: a reclassification supported by captured dimensions, a detention event supported by observed dwell, a redelivery nobody can account for.
Where the evidence backs the charge, it clears without a human touching it, including charges a threshold would have escalated for no reason. Where the evidence contradicts it, she drafts the dispute with the timestamps and the contract clause already attached. Where the evidence is absent, she says so and asks for it, which is a more honest outcome than an approval dressed up as a match. The same reasoning runs underneath what a detention charge has to prove before you pay it.
Where to start
Pull last month's freight invoices that cleared on tolerance rather than on an exact match. For each one, write down the evidence that was used to approve it. If the answer is the invoice itself, you have found the gap, and its size is the sum of that column.
Fixing it does not start with tighter thresholds. It starts with connecting the execution data you already collect to the bill that arrives afterwards, so the fourth leg exists at the moment the question is asked.
Book a walkthrough to see Penny run the fourth leg against your own freight invoices.