In this article
Key takeaways
- A freight audit checks every carrier invoice against the contracted rate and the shipment record before it is paid. Sampling and spot checks miss most small errors, and small errors repeated across thousands of invoices are where the money goes.
- The errors are predictable: rate drift, unsupported accessorials, fuel surcharge on the wrong index week, duplicate bills, weight or class changes without proof, and rates billed after they expired.
- An AI agent runs the three-way match on every line, resolves routine exceptions against the contract, drafts disputes from the evidence, and sends only ambiguous cases to a person, with every decision logged.
- For a national 3PL processing more than 12,000 carrier invoices a month, this recovered $1.4M of overbilling in the first year, cut the invoice exception rate from 22% to 6% and the AP cycle from 40+ days to 12.
- Results depend on data more than models. Digitized rate agreements with effective dates, clean shipment records and an agreed exception taxonomy matter more than which AI model you use.
What is a freight audit, and why does it matter?
A freight audit is the check of every carrier invoice against what was agreed and what actually happened: the contracted rate, the shipment record and the proof of delivery. It catches overcharges, duplicate bills, unsupported accessorial charges and wrong weights before the invoice is paid, and recovers money when an error is found afterwards.
Freight invoices are unusually error-prone because the final price is built after the shipment, from many moving parts. The linehaul rate depends on the lane and the contract version. The fuel surcharge depends on a diesel price index for a particular week. Accessorials such as detention, lumper fees, liftgate or inside delivery depend on what happened at the dock. Weight and freight class can change after a reweigh. Each part is a chance for the bill to differ from what you owe.
Any single error is usually small. That is exactly why it gets paid. A $40 fuel difference or one unsupported detention charge is not worth an analyst's afternoon, but the same pattern across thousands of invoices a month becomes a real line on the P&L. A freight bill audit exists to catch that pattern systematically rather than by luck.
Freight audit vs freight paymentFreight audit decides whether an invoice is right. Freight payment pays the approved invoices, codes them to the general ledger and reports on spend. Providers often sell freight audit and payment as one service, but an invoice should only reach payment after it has passed audit or its exception has been resolved.
Why freight invoice processing is harder for a 3PL
Shippers audit what they pay. A third-party logistics provider sits in the middle: it pays carriers and bills its own customers, often on different terms. Every overcharge it fails to catch comes straight out of its margin, and every slow payment strains the carrier relationships its service depends on.
Volume that scales with growth
Invoice count grows with every new customer and lane. If each exception needs a person, the AP team has to grow at the same rate.
Hundreds of rate agreements
Each carrier has its own contract, amendments, accessorial schedule and fuel table, and each one changes on its own date.
Evidence spread across systems
The rate lives in one place, the shipment and gate times in the TMS, the invoice in a portal or inbox, and the payment in the ERP.
Carriers waiting to be paid
A long AP cycle hurts capacity and goodwill with carriers. Paying fast and paying right usually pull in opposite directions.
This is the gap that AI in logistics is best suited to fill in finance operations: high volume, clear rules, evidence that already exists in systems, and a long tail of cases that still need human judgment.
Common freight billing errors: rates, accessorials and duplicates
Most freight invoice discrepancies fall into six groups. Each one has a clear test, which is what makes them suitable for automation.
Rate mismatch
The linehaul or per-mile rate does not match the contracted lane rate, or a minimum charge is applied when it should not be.
TestBilled rate vs the rate agreement version in force on the ship dateUnsupported accessorials
Detention, lumper, liftgate, inside delivery or redelivery billed without evidence that the service happened or was ordered.
TestCharge vs the order, gate times and delivery notesFuel surcharge errors
The surcharge is calculated on the wrong week's diesel index, the wrong table, or on accessorials it should not apply to.
TestRecalculate from the contract's fuel table and the index for the right weekDuplicate invoices
The same load billed twice, under a new invoice number, by a broker and a carrier, or once by EDI and once by email.
TestLoad, PRO or BOL number, dates and amount vs invoices already received or paidWeight and class changes
Billed weight, dimensional weight or freight class differs from the shipment record, with no reweigh or inspection certificate.
TestBilled weight and class vs the BOL and any inspection evidenceExpired or superseded terms
The carrier bills an old rate after an amendment, or applies an annual increase before its effective date.
TestContract effective dates vs the ship date on each invoiceNone of these needs advanced AI to detect in principle. The difficulty is doing all six checks on every line of every invoice, with the right contract version, while invoices keep arriving.
Manual vs AI-powered freight audit
Manual auditing is not careless; it is rationed. With limited hours, teams check the largest invoices, sample the rest and pay anything that looks roughly right. The trade-off between speed and accuracy is built into the process.
Manual audit
- Spot checks and samples, usually on the largest invoices
- Rate lookups in spreadsheets or PDFs, often an out-of-date version
- Accessorials approved because proving them wrong takes too long
- Disputes written one at a time and tracked in email
- Headcount has to grow with invoice volume
- No consolidated view of where margin is leaking
AI agent audit
- Every line of every invoice checked against contract and shipment
- Rate tables kept current, with each amendment's effective date
- Accessorials tested against gate times, orders and delivery notes
- Disputes drafted from evidence, tracked and followed up
- Throughput scales with compute, not people
- Leakage reported by carrier, lane and charge type
| Dimension | Manual or rules-only | AI agent with human review |
|---|---|---|
| Coverage | Sampled or high-value invoices | All invoices, all charge lines |
| Unstructured invoices | Keyed in by hand from PDFs and emails | Read and normalized automatically, low-confidence fields flagged |
| Exception handling | Every exception goes to a person | Routine exceptions resolved against contract terms; ambiguous ones routed |
| Disputes | Written and chased manually | Drafted with evidence, tracked, stale cases followed up |
| Audit trail | Scattered across email and spreadsheets | Every match, exception and decision logged with its evidence |
| Control | Depends on who is checking | Configurable value and confidence thresholds for human approval |
Rules engines in traditional freight audit software already handle the clean cases well. What AI agents add is the messy middle: reading invoices that arrive as PDFs or emails, matching a charge to evidence spread across systems, deciding whether an exception is routine, and writing the dispute. If you are weighing this against a generic AI tool, our explainer on agentic AI vs generative AI covers why an agent that takes actions is different from a model that only answers questions.
How AI agents audit freight invoices, step by step
An audit agent is not one model guessing at invoices. It is a workflow that combines document reading, deterministic checks against your own data, and a small number of judgment calls, each of which is logged.
- Ingest
Invoices arrive by EDI, carrier portal, email or PDF.
- Extract
Charges, references, weights and dates are read and normalized.
- Match
Invoice vs rate agreement vs shipment record, line by line.
- Classify
Each mismatch is typed: rate, accessorial, fuel, duplicate, weight.
- Resolve
Routine cases settled on contract terms; disputes drafted.
- Reconcile
GL coding, credits matched, leakage reported.
1. Ingest and extract
Structured EDI invoices are easy. The hard part is the long tail of PDFs, scanned bills and email attachments from smaller carriers. The agent reads each one, pulls out the load or PRO number, dates, weights and every charge line, and normalizes charge names so that "DET", "Detention" and "Wait time" become the same thing. Fields it is not confident about are flagged rather than guessed.
2. Match against contracts, rate cards and shipment records
For each invoice the agent finds the shipment in the TMS and the rate agreement version that was in force on the ship date. It then compares every charge: linehaul against the lane rate, fuel against the contract's fuel table and the correct week's index, accessorials against the order and the gate times, weight and class against the bill of lading. This is the three-way match, and it is mostly deterministic: the agent's value is in finding the right evidence, not in inventing a rule.
3. Flag discrepancies and suspicious charges
Differences inside an agreed tolerance pass. Anything outside it becomes an exception with a type, an amount and the evidence that produced it. Patterns also count: a carrier whose detention charges keep rising, or invoices that repeat a reference number with small changes, are flagged even when each invoice alone looks plausible.
4. Classify, resolve and dispute
This is where an agent differs most from a rules engine. A routine exception, such as a fuel surcharge on the wrong week, can be recalculated and short-paid with the working attached. A clear overcharge becomes a dispute letter built from the evidence. An ambiguous case, such as a detention charge where the gate log is missing, goes to an analyst with everything already gathered. The next section shows these paths.
5. Reconcile and report
Approved invoices are coded to the general ledger and released for payment. Carrier credits are matched back to the original invoice, so a dispute is not closed until the money arrives. Finance gets a live view of open disputes and leakage by carrier, lane and charge, which is often the first time anyone has seen where margin actually goes.
We help logistics finance teams list the checks, the data each one needs and the exceptions that should always reach a person, before any build starts.
The three-way match, line by line
Here is what an audit looks like on a single load. The carrier sent two invoices. Each charge line is compared with the rate agreement and the shipment record, and each gets a verdict and an action.
$2,840.00
$2,840.00 contracted lane rate
1 load, 42,000 lb, delivered on time
Match
Approved. Rate, lane and service level agree.
$610.60
20.7% at the index week of pickup
Pickup in the week the index was 20.7%
Adjusted
Carrier used the rate for the following week, 21.5%. Recalculated to $587.88; short-paid with the calculation attached.
$225.00
$75/hr after 2 free hours
Arrived 08:05, released 09:50: 1 h 45 min on site
Dispute
Inside the 2 free hours, so nothing is billable. Dispute drafted with gate times from the shipment record.
$95.00
$95 when requested on the order
Not on the order; dock delivery
Dispute
Service not ordered and no delivery note showing use. Disputed with the order attached.
$2,840.00
One payment per load
Load already invoiced on INV-48213
Duplicate
Blocked before payment. Linked to the original invoice for the AP record.
Illustrative example. The carrier, load, rates and amounts are fictitious and do not come from the case study below.
Two things stand out. First, most of the money in this example is not in the clever cases: it is the duplicate, which a person checking invoices one by one can easily miss when the second copy arrives a week later through a different channel. Second, the agent does not decide disputes by itself. It assembles the evidence and the calculation, and the approval thresholds decide whether a person signs off before anything is sent.
How freight invoice exceptions are classified, investigated and resolved
An exception is not a problem by itself. The cost comes from how long it sits and how many people touch it. The goal of exception handling is to send each one down the shortest correct path.
Released to payment and coded to the GL. No one touches it.
Recalculated or short-paid with the working attached, then logged.
An analyst decides, with the invoice, contract clause and shipment data side by side.
- Classification. Every exception gets a type (rate, accessorial, fuel, duplicate, weight or class, expired terms), an amount and a confidence. The type decides the path; the amount and confidence decide whether a person must approve.
- Investigation. The agent gathers what an analyst would look for: the contract clause, the amendment history, gate times, the order, proof of delivery, and earlier invoices for the same load. When evidence is missing, it says so instead of guessing.
- Disputes. Dispute correspondence is drafted from the matched evidence, sent according to your approval rules, tracked, and followed up when the carrier goes quiet. Credits and adjustments are reconciled to the original invoice.
- Human approval thresholds. Value and confidence limits keep unusual charges under review. A new carrier, a charge type the agent has rarely seen, or a dispute above a set amount can always require sign-off.
Explainability is a requirement, not a featureFinance leaders sign off on payments and auditors review them. Every automated approval, adjustment and dispute should be traceable to the contract term and the shipment data it relied on. If an agent cannot show why it paid or withheld a dollar, it should not be paying anything.
Case study: how a national 3PL recovered $1.4M and cut invoice exceptions from 22% to 6%
A national third-party logistics provider was processing more than 12,000 carrier invoices a month against hundreds of contracted carrier rate agreements. Bonami built an AI invoice exception and reconciliation agent, part of Bonami X AI, connected to the provider's ERP and TMS. The client is anonymised in the published case study, so it is here too.
within the first year, from overcharges and duplicate billings that had previously been paid without challenge
Results as published in Bonami's 3PL invoice reconciliation case study. Client anonymised.
Roughly 22% of the 12,000+ monthly invoices were flagged as exceptions, and each needed a manual three-way match against the contracted rate, the shipment record and the invoice. Overbilling and duplicate charges routinely slipped through. There was no consolidated view of open disputes, and the AP cycle had stretched past 40 days, straining carrier relationships and the monthly close.
- Three-way match of each invoice against the rate agreement and shipment, with fuel surcharge and accessorial logic
- Classifies rate, accessorial, duplicate and weight mismatches; resolves routine ones on contract terms
- Files and tracks disputes, follows up stale cases, reconciles credits
- Keeps rate tables current, applying increases from their effective date and flagging expired terms
- GL coding, a margin-leakage dashboard by carrier, lane and charge, and a full audit trail
The agent absorbed the routine share of exceptions, so the AP team only saw cases that genuinely needed a person. Systematic matching against contracted rates turned invisible leakage into recovered cash. Clean invoices cleared quickly and disputes moved in parallel instead of in a manual queue, which compressed the payables cycle and repaired carrier relationships.
One design choice mattered as much as the matching itself: configurable value and confidence thresholds. They kept unusual charges in front of people and made finance leadership comfortable letting the agent approve and dispute the rest, because every decision came with its evidence.
Business benefits of AI-powered freight audit
Recovered cash and avoided overpayment
Overcharges and duplicates are stopped before payment or recovered through disputes, instead of being written off as the cost of doing business.
A shorter AP cycle
Clean invoices clear without waiting in a queue, which helps carrier relationships and makes the monthly close easier.
Time back for the AP team
People spend their time on the cases that need judgment and on carrier conversations, not on lookups and copy-paste.
Visibility into margin leakage
Leakage by carrier, lane and charge type gives procurement real data for the next rate negotiation.
An audit trail by default
Every match, exception and dispute is logged with its evidence, which simplifies internal controls and external audits.
Growth without a matching backlog
Peak season and new customers add invoices, not hiring plans. Throughput scales with compute rather than headcount.
For shippers, the same approach applies to their own freight spend. For 3PLs and freight forwarders it has an extra effect: the margin between what you pay carriers and what you bill customers becomes something you can measure every week instead of every quarter. Our guide to AI in transportation covers the operational side of logistics AI; freight audit is its finance-side counterpart.
Before you automate freight audit: data, rules and KPIs
In our experience, audit automation succeeds or stalls on data, not on the choice of model. Get these in place first.
Digitize rate agreements with effective dates
Lane rates, minimums, accessorial schedules and fuel tables for every carrier, including amendments and the date each one takes effect.
Connect the TMS and ERP
Read access to shipments, gate times, orders and proof of delivery in the TMS, and to vendors, invoices and payments in the ERP. Our ERP integration work usually starts here.
Agree an exception taxonomy and tolerances
Name the exception types, the tolerance for each, and who owns each type. This becomes the agent's routing table.
Set approval thresholds
Decide which values, carriers and charge types always need a person, and which routine fixes the agent may apply on its own.
Define the dispute process
Who signs off, how disputes are sent, how long to wait before following up, and how credits are matched back.
Take a baseline, then run in shadow mode
Measure today's numbers, let the agent audit alongside the team without acting, compare its decisions, then switch on actions one type at a time.
KPIs to track from day one
Share of invoices that fail the match. The case study moved this from 22% to 6%.
Share of invoices approved without anyone opening them.
Dollars stopped before payment plus credits received from disputes.
Days from invoice receipt to payment, for clean and disputed invoices separately.
Open disputes by age and carrier, and the share resolved in your favour.
Staff minutes per exception, before and after. The case study reported 80% less.
How Bonami helps with freight audit automation
Bonami builds finance-operations agents as part of Bonami X AI, our AI agent development practice. We build them into the systems you already run rather than asking you to move to a new platform, and we design the controls (thresholds, evidence, audit trail) alongside the automation.
Audit workflow assessment
We map your invoice sources, rate data, exception types and dispute process, and identify which checks can be automated first.
Invoice exception agent
Three-way match, exception classification, dispute drafting and tracking, built on the AI invoice exception handler.
Reconciliation and integration
ERP and TMS connections, GL coding and close-ready reconciliation with our autonomous reconciliation agent.
Frequently asked questions about freight audit and AI agents
What is a freight audit?
A freight audit is the check of every carrier invoice against what was agreed and what actually happened: the contracted rate, the shipment record and the proof of delivery. Its job is to catch overcharges, duplicate bills, unsupported accessorial charges and wrong weights or classes before the invoice is paid, and to recover money when an error is found after payment.
What is the difference between a freight audit and freight payment?
Freight audit decides whether an invoice is correct. Freight payment is the process of paying approved invoices, often with consolidated payment runs, GL coding and reporting. Many providers sell the two together as freight audit and payment, but they are separate steps: an invoice should only reach payment once it has passed audit or its exception has been resolved.
What are the most common freight billing errors?
The usual ones are rates that do not match the contract, accessorial charges such as detention, lumper or liftgate billed without supporting evidence, fuel surcharges calculated on the wrong index or week, duplicate invoices for the same load, weight or freight class changes without a reweigh certificate, and invoices billed against an expired or superseded rate agreement.
How does an AI agent audit a freight invoice?
The agent reads the invoice, pulls the matching rate agreement and shipment record from the TMS and ERP, and checks each charge line against them. Clean invoices go to approval. Mismatches are classified by type, routine ones are resolved against the contract terms, disputes are drafted from the evidence, and only ambiguous cases go to an AP analyst with the reason and documents attached. Every decision is logged.
What is a three-way match for freight invoices?
In freight, a three-way match compares the carrier invoice with the contracted rate (the rate agreement or tariff) and the shipment record (what moved, when, how heavy, with which services, and the proof of delivery). If all three agree within tolerance, the invoice can be approved. If not, the difference becomes an exception to resolve or dispute.
Can AI replace the freight audit team?
No, and it should not try to. In a well-designed setup the agent handles the volume of routine checks and routine exceptions, and people handle ambiguous cases, carrier conversations and policy decisions. Configurable value and confidence thresholds decide which charges always go to a person. The aim is to point the team at the invoices that need judgment, not to remove it.
What data do you need before automating freight audit?
Digitized rate agreements and amendments with effective dates, access to shipment records in the TMS, invoice data from EDI, portals or PDFs, proof of delivery and accessorial evidence where it exists, and AP data in the ERP. You also need an agreed exception taxonomy, approval thresholds and a dispute process. In our experience, gaps in rate data are the most common reason audit automation underperforms.
How do you measure the ROI of freight audit automation?
Track the invoice exception rate, the share of invoices approved without a touch, overbilling recovered or avoided, the AP cycle in days, dispute ageing and the time staff spend per exception. Take a baseline for each before you start, and report the same measures monthly afterwards. Recovered dollars matter, but cycle time and staff time are often the larger long-term gain.
The takeaway: audit every invoice, and let people judge the hard ones
Freight overbilling rarely looks like fraud. It looks like a fuel surcharge on the wrong week, a detention charge nobody can disprove, or a duplicate that arrives through a different inbox. Sampling cannot catch a pattern made of small amounts; checking every line against the contract and the shipment can.
That is the job AI agents are good at: doing the lookups and the matching on every invoice, resolving the routine cases, building the evidence for disputes, and handing the genuinely ambiguous ones to your team. For one national 3PL it meant $1.4M recovered in the first year and an exception rate that fell from 22% to 6%. Start with your rate data and your exception rules, and the rest becomes engineering.
Want to see what an audit agent would find in your invoices?
Tell us how many carrier invoices you process a month, how they arrive, and where your rate agreements live. We will outline which checks to automate first and what data the agent would need.
This article is general guidance. The hero illustration and the sample invoice are fictitious and illustrative. Case-study results are quoted from Bonami's published 3PL invoice reconciliation case study; the client is anonymised and results will differ for other organizations.