To evaluate a freight audit provider without sitting through a sales cycle, run a pilot on your own invoices: pick 40 to 60 recent freight bills with their rate confirmations and PODs, write down what a pass looks like before you send anything, hand the same sample to every vendor you are considering, and score the findings on whether they are real and collectible rather than on how many there are. Two weeks, start to finish. No demo required.
That is the whole method. The rest of this post is the detail: how to build the sample, what to seed into it, what to measure, and the ways a pilot quietly produces a misleading answer.
A demo shows you a vendor’s best case on the vendor’s data. A pilot shows you your case on your data. There is no version of a product walkthrough that tells you what a freight audit tool will find in your specific mess of carriers, accessorial schedules and rate cons. Only your invoices do that.
What a freight audit pilot is actually testing
Three questions, in order. Most buyers only ask the first one.
1. Does it find anything? The detection question. Given your bills, does the provider surface exceptions you did not already know about?
2. Are the findings real? The precision question. A tool that flags 40 exceptions of which 12 hold up is worse than one that flags 14 of which 13 hold up, because your team pays the review cost on every flag. Ask this one hard.
3. Can you collect? The recovery question. A finding is not money. It becomes money when a carrier issues a credit. A pilot short enough to be worth running usually will not close this loop, which is a real limitation you should plan around rather than pretend away.
If you only measure the first question, every vendor passes and you will pick on price or on who was friendliest. Score all three.
Build the sample before you talk to anyone
The sample is the pilot. Get it wrong and nothing downstream matters.
Use 40 to 60 invoices, not 10 and not 500. Ten is too few to distinguish signal from luck. Five hundred is a project. Somewhere in the 40 to 60 range you get enough carriers and enough charge types to be representative without spending a week assembling it.
Pull from the last 60 to 90 days. Recency matters for a reason that has nothing to do with convenience. Under 49 U.S.C. 13710, a shipper must contest a bill within 180 days of receiving it to preserve the right to challenge it. Run your pilot on bills from 14 months ago and every finding is a history lesson. The full set of clocks is laid out in our freight billing deadlines reference, and they are the reason a pilot on stale invoices is worth less than it looks.
Cover your actual carrier mix, not your best carriers. If 30 percent of your spend is LTL, roughly 30 percent of the sample should be LTL. Include the regional carrier nobody likes dealing with. Include the drayage bills if you have them. A sample of clean truckload invoices from your two best partners tests nothing.
Send the supporting documents. An invoice on its own can only be checked against itself. Reconciliation needs the rate confirmation (contract rate, accessorial schedule, free time trigger) and the POD or BOL (timestamps, delivery type, piece count). If a provider tells you it can audit invoices with no supporting documents, ask precisely which error types that catches. The honest answer is: math errors, duplicates within the sample, and tariff-level checks. Not much else.
Include at least one invoice you already know is wrong. This is the single highest-value thing you can do, and almost nobody does it.
Seed the sample with known errors
Before you send anything, find two or three bills where you already know the answer. A detention charge you disputed and won. A duplicate invoice your AP clerk caught. An accessorial that was never on the rate con. Do not tell the vendor which ones they are.
These are your controls. If a provider misses a duplicate you found by eye, that is diagnostic in a way no reference call will ever be. If it catches all three plus two you missed, that is diagnostic too.
One caution on the arithmetic: a seeded control tells you whether the tool catches that specific error type. It does not give you a hit rate. Three controls out of 50 invoices is not a statistically meaningful measurement of anything, and you should resist the temptation to write “detected 3 of 3, so 100 percent accurate” in a memo to your CFO.
Write the scorecard before you see any results
Decide what passing means while you are still neutral. Findings are seductive. A vendor that hands you a report with a big dollar number at the top will make you want to say yes, and the number at the top of a findings report is not money, it is a hypothesis.
| Dimension | What to measure | A reasonable bar |
|---|---|---|
| Detection | Exceptions surfaced that you had not already flagged | More than zero net-new, and at least one you consider material |
| Seeded controls | Did it catch the errors you planted | All of them, or a clear explanation of why not |
| Precision | Of the flagged exceptions, how many survive your own review | The large majority. Track the false positives explicitly |
| Evidence quality | Does each finding cite the document and show the recomputation | You can send it to a carrier without rebuilding it |
| Explainability | Can you follow the arithmetic on a fuel surcharge or detention finding | Yes, line by line, without a call |
| Turnaround | Calendar time from sending documents to receiving findings | Inside your carrier credit period. See below |
| Effort | Hours your team spent on setup and document handoff | Under half a day for a pilot this size |
| Handling of small findings | Are sub-$100 exceptions reported or silently dropped | Reported, with a threshold you control |
Fill in your own bars. The point is that they exist in writing before the report arrives.
Turnaround is not a vanity metric
Freight bills come with a payment clock. 49 CFR 377.203 sets the standard credit period at 15 days beginning the day after the freight bill is presented, extendable by published tariff to no more than 30 calendar days. That is the window in which catching an error is cheap, because you have not paid yet.
After you pay, the error is still recoverable, but the path is different and slower: an overcharge claim, which under 49 CFR 378.8 the processing carrier must pay, decline or settle within 60 days of receiving it in writing. Recovering money you already sent is strictly worse than not sending it. So when you measure a provider’s turnaround in a pilot, measure it against your credit period, not against a vague sense of “fast.”
The two-week pilot, day by day
This is a schedule that works for a 40 to 60 invoice sample. Compress it if you can, but do not skip step four.
| Days | What happens |
|---|---|
| 1 to 2 | Assemble the sample. Pull invoices, rate cons, PODs. Note your seeded controls in a private file |
| 2 | Write the scorecard. Set your bars. Circulate to whoever will make the decision |
| 3 | Send the identical package to every provider on your list, same day, same documents |
| 4 to 8 | Providers work. You do nothing. Note when each one comes back |
| 9 to 11 | Review every finding yourself. Mark each one valid, invalid, or already known |
| 12 | Score against the card. Compare providers on precision first, volume second |
| 13 to 14 | Send two or three real disputes from the strongest findings and see what the carrier says |
Step four, sending the same package to everyone, is what makes the comparison mean anything. Different samples produce different findings and you will learn nothing about the providers, only about the invoices.
Days 13 and 14 are optional and valuable. You will not get carrier responses back inside the pilot window, but starting the disputes tells you whether the evidence packet is actually sendable. If you have to rebuild a finding before you can put it in front of a carrier, the finding was half-finished.
What to read in the findings report
Skim the total, then ignore it. Go line by line.
Does each finding name the document it contradicts? “Detention charge appears unsupported” is not a finding. “4.0 hours detention billed on invoice 48213; POD carries no in or out times; rate con free time is 2 hours from arrival” is a finding. Our post on the six documents that win a detention dispute is the standard to hold the report to.
Is the arithmetic shown? Fuel surcharge is the clearest test, because it is pure computation. A finding should show the index value used, the week applied, the peg, the scale step and the resulting per-mile or percentage figure. If it just says “fuel surcharge appears high,” you cannot dispute it and neither can they. The mechanism behind the most common version of this error is in our post on which DOE week applies to your fuel surcharge.
Are the small findings there? A $42 residential surcharge on a commercial address is a real error. Whether a provider reports it tells you something about its economics: under a contingency model, a $42 finding costs more to chase than it returns, so the incentive is to drop it. That is rational, not dishonest, and we walk through how each freight audit pricing model shapes what gets reported.
How many findings did you reject? Count them. This is the number most buyers never write down, and it is the best single predictor of what the tool will cost your team in review time every month for the next three years.
What a pilot cannot tell you
Be honest about the limits, or the pilot will overstate its own conclusions.
- It does not measure your recovery rate. Carrier responses take weeks. A pilot measures findings, and findings convert to dollars at a rate you will not know for a quarter.
- It does not test scale. Fifty invoices handled well says nothing about 1,500 a month, exception queue management, or what happens during a peak week.
- Small samples are noisy. If Darrigo Consulting’s benchmarks are in the right neighborhood (recovery of 1 to 3 percent of spend on truckload, 3 to 8 percent on LTL, 2 to 5 percent on parcel, consultant-authored, so treat it as a reference point and not a promise), then a 50-invoice sample can easily land above or below your true rate by chance. Do not extrapolate a pilot’s dollar total to an annual figure and put it in a business case. It will be wrong in whichever direction flatters the decision you already made.
- It does not test the relationship. How a vendor behaves when a finding turns out to be wrong, or when your CFO asks an awkward question in month nine, is not observable in two weeks.
- A zero-finding result is information, not failure. If a provider reads 50 of your bills and finds nothing, either your carriers bill unusually cleanly or the sample was too easy. Both are worth knowing. Rerun with your messiest carriers before you conclude either way.
If you are the shipper
Your pilot is about the credit period. You want findings before the bill is paid, evidence you can forward to the carrier without rework, and a threshold setting that does not silently discard the long tail of $40 and $80 errors, because at volume that tail is most of the money.
If you are the broker
Your pilot is about the gap between the rate confirmation and the carrier invoice, and it is worth building the sample around loads where you already suspect margin moved. Include consolidated invoices covering multiple loads, because that is where naive matching breaks, and include at least one settlement where an accessorial was billed to you and not passed through. Our rate con versus carrier invoice reconciliation procedure is the workflow a broker-side pilot should be testing against.
The checklist
- Pull 40 to 60 invoices from the last 60 to 90 days, matching your real carrier and mode mix.
- Include the rate confirmations and PODs. Note what you could not find; that gap is your finding too.
- Seed two or three known errors. Do not disclose them.
- Write the scorecard first: detection, precision, evidence quality, turnaround against your credit period, effort.
- Send the identical package to every provider on the same day.
- Review every finding yourself and count the ones you reject.
- Send two or three real disputes to test whether the evidence packet is sendable as-is.
- Score, then decide. Precision before volume, evidence before dollars.
- Do not annualize the pilot total. Use it to choose a provider, not to forecast a return.
If a provider will not accept a sample of your invoices without a scoping call first, that is itself a data point about how the next three years will go. The evaluation you can run yourself, on your own bills, in two weeks, is more informative than any demo, and it costs you nothing but the time to assemble the documents. Before you commit, it is also worth knowing exactly what a freight audit provider should show you before you sign.
Sources
- 49 U.S.C. 13710 (180-day window to contest a freight bill; 180-day window for balance-due billing)
- 49 CFR 377.203 (15-day standard credit period, extendable by published tariff to no more than 30 calendar days)
- 49 CFR 378.8 (60-day clock to pay, decline or settle a written overcharge claim)
- Darrigo Consulting, shipping and freight audit guide (recovery benchmarks of 1 to 3 percent truckload, 3 to 8 percent LTL, 2 to 5 percent parcel; consultant-authored)