An analyst disputes a 400 dollar limited access fee on your Columbus to Memphis lane in March, sends the carrier the gate photographs, and recovers it. In April the identical fee lands on the same lane from the same carrier. She wins that one too. By September the dispute log holds seven clean recoveries, an analyst who is plainly good at her job, and a charge that has never once stopped arriving.
Nobody in that story did anything wrong. The audit behaved exactly as specified: it compared an invoice to a contract, found a variance, and got the money back. What it never did was ask why the variance keeps happening, because no single invoice contains that answer.
A dispute queue has no memory
Invoice audit is stateless by design. Each bill is scored on its own evidence against its own contract terms, and that independence is what makes the finding defensible when the carrier pushes back. It is also what makes the same result land in the queue every month looking brand new.
The scorecard compounds it. Freight audit programs report recovery dollars and dispute win rate, and both of those numbers improve when a recurring overcharge keeps recurring. Seven wins reads as a strong quarter. Seven wins against one cause nobody fixed reads as seven months of an analyst re-proving a point the carrier conceded in March.
Size finishes the job of hiding it. A repeating charge tends to be small enough per occurrence to clear review in a few minutes and large enough across a year to be worth a meeting. Four hundred dollars against a 2,800 dollar invoice is an exception an analyst works and closes. The same four hundred, twelve times over, on three lanes into the same consignee, is a figure that has never appeared on a report, because no report your team runs groups spend that way.
What a repeating charge is telling you
A charge that arrives on schedule is not an anomaly. It is a rule executing correctly somewhere, and the only useful question is whose rule. There are four answers, and they belong to four different people.
Your master data is stale. Old Dominion's shipper guidance on the 2025 NMFC changes states that Docket 2025-1 took effect on 19 July 2025, that a revised 13 subprovision density scale replaced the previous 11 tier system, and that roughly 2,000 commodity listings were consolidated into single items. If a SKU in your product master still carries its pre-docket class, every shipment of that SKU rates wrong at tender, the carrier reclassifies it on arrival, and the correction reaches you as a charge. The bill is right. The input is wrong. A claim against it fails on the merits, and filing the claim twelve times produces twelve denials.
The contract never covered it. An accessorial outside the negotiated package rates off the carrier's public tariff by default. Nothing is in breach, so nothing is recoverable, and the fix is a pricing conversation rather than an audit one. That one goes to Miles at the next bid, with the occurrence count and the annualised exposure attached to the line so it gets priced instead of overlooked again.
The facility earns the charge. Some consignees have no dock, a trailer length limit at the gate, security screening, or an appointment desk open two mornings a week. The fee is owed on every delivery because the condition holds on every delivery. The response is operational: change the delivery profile, negotiate a flat, or accept the fee as part of the true lane cost and stop treating it as a variance.
A carrier rating rule misfires systematically. A location code sitting in the carrier's own system with a limited access flag on it, a fuel table reading the wrong week, an origin zip band set during onboarding and never revisited. One call to the pricing desk corrects every future occurrence. Twelve claims correct none of them.
Three of those four are not disputes at all. That is why a process whose only output is a dispute can run for years without closing any of them.
| Invoice by invoice | Across the group | |
|---|---|---|
| What it compares | One bill against one contract | Every shipment on the lane, billed and clean |
| What it produces | A dispute with evidence | A cause with a named owner |
| Next month | The charge is back | The charge stops, or it is priced in |
| Cost of the fix | Paid again every cycle | Paid once |
The clock a pattern loses against
TForce Freight publishes its rule plainly: an overcharge claim shall not be paid unless filed in writing with the carrier within 180 days of receipt of TForce Freight's bill, and for the purpose of that rule receipt means ten days after the shipment was tendered. The same page restricts filing to the payer of the original freight charges and points at the claim procedures in NMFC series items 300201 through 300209. Windows differ by carrier, and some of them count from pickup rather than from the invoice, so the bar falls in a different place for every carrier in your routing guide.
Run the arithmetic on a pattern caught late. A charge that started fourteen months ago and surfaces today is not a fourteen month recovery at that carrier. Roughly the first eight occurrences sit outside the window already, and they were outside it before anyone went looking. Detection latency is not a reporting inconvenience here. It decides whether a confirmed overcharge is money you collect or a number you write down.
How Penny groups the evidence
Penny runs attribution as a second pass over invoices she has already audited, and the unit of analysis is the group rather than the bill.
- Normalise the charge first. One fee carries a different code and a different description at every carrier, so a grouping keyed on raw text splits a single pattern into six fragments, each too small to look like anything. The naming problem is the same one set out in eight ways to say detention.
- Group on the fields that could plausibly cause it. Carrier, normalised charge type, origin and destination facility, SKU, service level, and the shipment attributes that feed rating.
- Test whether the group is stable. A fee on eleven of twelve moves into one consignee is a rule. The same fee on two of forty is an event. Amount behaviour carries as much signal as frequency: a figure identical every time points at a flat tariff item, a figure that tracks weight or dwell points at something computed.
- Isolate the discriminating field. The cause is whatever is true of the billed shipments and false of the clean ones on the same lane. If every billed move went to one consignee door and none of the unbilled ones did, the attribution is in hand and it is testable on next week's shipments.
- Attach a value and an owner. Occurrence count, mean amount, annualised exposure, the share still inside the carrier's filing window, and which of the four causes above it is.
Where a group is too thin, or where two fields separate the billed shipments equally well, Penny returns the candidates and names the field that would settle it rather than declaring a cause. A confident attribution to the wrong driver is expensive in a specific way: somebody spends a quarter renegotiating a tariff item that was never the problem, and the charge carries on arriving the whole time the fix is in flight.
What to do next
Pull twelve months of accessorial lines and group them by carrier, charge code and destination facility. Sort that result by occurrence count rather than by dollar value, which is the inverse of how every freight spend report you currently own is sorted. Repeating causes live at the top of that list, and most of them are too small individually to have ever reached a dashboard.
Take the top twenty groups and put the four way question to each one: stale input, missing contract term, genuine operating condition, or carrier rating rule. Then count how many occurrences in each group already fall outside the filing window of the carrier that billed them. In most programs that second number is the one that gets the work funded.
Bring the twelve months to a walkthrough and we will run the grouping live, with the cause, the discriminating field and the recoverable share shown for every pattern Penny finds.