How to automate freight audit and payment with AI

Extract the rate confirmation, bill of lading, and carrier invoice into one typed schema, let software reconcile them line by line, and pay only when they agree. Mismatches and low-confidence fields route to a person, so every invoice is audited before money moves, not a sample after.

What is freight audit and payment?

Freight audit and payment is the settlement step of transportation: checking each carrier invoice against what was agreed before it is paid. The agreement lives in the rate confirmation or the contract tariff. The proof of what actually shipped lives in the bill of lading and the proof of delivery. The audit asks whether the amount billed matches the amount owed for the freight that actually moved.

The money at stake is real. US business logistics costs reached $2.58 trillion in 2024, 8.8 percent of GDP, per the 2025 CSCMP State of Logistics report. Invoices are wrong often enough to matter: Transportation Insight, a freight audit provider, reports that a typical audit finds errors on 3 to 6 percent of invoices, worth 1 to 5 percent of total freight spend in recoverable charges.

The work traditionally goes to a freight audit provider or an in-house accounts payable team with a rate sheet and a queue of PDFs. Either way, the raw material is three documents describing one shipment in three formats, none structured. Freight audit is a document AI problem before it is a payments problem.

Which documents does a freight audit compare?

Every audited shipment rests on a document triangle. The rate confirmation, or the contract tariff for committed freight, is the agreement: linehaul rate, fuel surcharge basis, accessorial schedule, and free time before detention starts. The bill of lading is the operational record: what shipped, when, in what quantity, signed at origin. The carrier invoice is the claim: what the carrier believes it is owed. Proof of delivery and lumper receipts round out the packet.

Disputes concentrate in a few fields:

Disputed field	Where the agreed value lives	Why it gets disputed
Linehaul	Rate confirmation	Wrong rate applied, or the load was re-rated after booking
Fuel surcharge	Surcharge schedule on the rate con, commonly pegged to the weekly EIA diesel price index	Wrong index week or wrong basis applied
Accessorials (liftgate, layover, TONU)	Accessorial schedule on the rate con	Charged without authorization, or billed at an unlisted rate
Detention	Free-time terms on the rate con plus arrival and departure timestamps	Timestamps missing or contested
Lumper fees	Lumper receipt	Receipt missing, or the invoice amount differs from the receipt

Detention deserves its own paragraph because it is expensive for both sides. The American Transportation Research Institute found detention on 39.3 percent of driver stops in 2023, costing the trucking industry $3.6 billion in direct expenses and $11.5 billion in lost productivity. A detention charge without timestamps is an argument. With timestamps, it is arithmetic.

Why does manual freight audit miss money?

Three reasons: thresholds, formats, and time.

Manual audit has a labor cost per invoice, so teams set dollar thresholds and audit above them. Everything below the threshold is paid on trust, and the errors in that long tail are never found. Sampling has the same failure mode with extra steps.

Formats slow everything down. The rate confirmation is a PDF from a broker's TMS, the bill of lading a scan or phone photo from a dock, the invoice an emailed PDF in whatever layout the carrier's billing system produces. An auditor re-keys three layouts into one spreadsheet before the comparison starts, and re-keying is where transpositions creep in.

Time finishes the job. Carriers invoice on payment terms, disputes have windows, and quick-pay discounts expire. When the queue backs up, the options are paying unaudited invoices to keep carriers moving or holding payment and burning goodwill. Most operations quietly pay. A process that exists to catch overcharges ends up approving them.

Why can't OCR or generic intelligent document processing reconcile freight invoices?

OCR reads text. Reconciliation is arithmetic over typed values, and you cannot do arithmetic on text. To audit one shipment, the linehaul on the invoice must be a number, the surcharge schedule must be a lookup table, the free-time window must be a duration, and the timestamps must be timestamps, all matched to the same load number across three documents. OCR gives you none of that, and layouts vary by carrier, so template-based tools break on the next carrier's invoice.

Generic intelligent document processing gets closer but stops at extraction. It hands you fields; it does not join three documents into one settlement record, recompute the fuel surcharge from the agreed schedule, or route a mismatch to a person with the evidence attached.

Then there is the math a technical committee runs. Per-field accuracy of 98 percent compounds, over a ten-field document, to roughly 82 percent document-level accuracy: nearly one in five documents wrong somewhere, and a settlement packet spans three documents. In a workflow that moves money, blended accuracy is the wrong question. The right question is what the system does with fields it is not sure about.

How does schema-enforced, deterministic extraction automate freight audit?

bem automates entire business processes end to end: it turns unstructured data into auditable outputs, then makes and records the decisions in between.

Here is the settlement workflow on bem's functions, behind one V3 API. Documents arrive by email or API. Split breaks a mixed packet into individual documents. Classify labels each: rate confirmation, bill of lading, carrier invoice, lumper receipt. Extract pulls each document into its typed schema and validates the output against it, so a missing field surfaces now, not as a null in your ERP. Every extracted field carries its own confidence score.

Join assembles the three records into one settlement record keyed on load number. From there the audit is rules, not judgment: compare billed linehaul to the agreed rate, recompute the fuel surcharge from the schedule, check each accessorial against the authorized list, compute detention from timestamps and free time. You write those rules as policy, and Decide applies them to every settlement record and records its reasoning.

Clean matches flow through Payload Shaping into your TMS or AP system for payment. A mismatch, or any field below your confidence threshold, routes to a human reviewer with the source document attached. Failures surface as a review, not a wrong payment.

Every determination keeps its evidence: input document, extracted output, function version, confidence, reviewer. When a carrier disputes a short-pay, you send the packet instead of scheduling a call. Functions are versioned and idempotent, same input, same output until you publish a new version, with regression testing and drift detection on every version. That is deterministic guardrails around probabilistic models in practice, and why this workflow can touch money. It is the same engine bem runs across supply chain and logistics settlement and intake.

What results does automated freight audit and payment deliver?

Fleetio is a fleet operations platform. Fleetio cut invoice processing time 65 percent, with 100 percent accuracy on invoice totals. That deployment is fleet invoices, not carrier freight bills, but it is the same money-moving determination: a total that must be right before payment, schema-validated and confidence-routed so uncertain values become review tasks, not wrong payments.

Ply saves 15 hours per team member every week, the same kind of hours a settlement team spends re-keying and eyeballing today. And the average bem customer grows its usage 4x in the first year, growth that only happens when a first workflow survives production.

We do not have a public freight audit logo to name yet and will not invent one. The section above is the pattern we sell into logistics settlement teams. Judge it the way an auditor would: full coverage instead of sampling, a typed record for every invoice, and evidence behind every determination. The economic buyer's packet: SOC 2 Type II, audited annually; zero-retention mode, configurable per workflow on Enterprise; and a clear training policy, because bem never trains on customer data for its own benefit or for other customers; you decide when to fine-tune a function, and that data stays yours.

How do we start automating freight audit and payment?

Start with a single lane or a single carrier, not the whole network. Define three schemas: rate confirmation, bill of lading, carrier invoice. Type the disputed fields strictly: rates as decimals, timestamps as timestamps, accessorials as an enumerated list rather than free text. Backtest before you go live: run the last quarter's paid invoices through the pipeline and measure what full-coverage audit would have recovered, since that number is the business case and it comes from your own freight. Set confidence thresholds conservatively and send every exception to a person at first; the review queue tells you where to tighten. Keep the audit trail on from day one, because the first carrier dispute will arrive before the first month closes.

The goal is not removing people from settlement. It is moving them from re-keying every invoice to deciding the small fraction where the documents genuinely disagree, with the evidence already assembled.