Automated freight audit and AI invoice verification for large shippers, by audit stage and selection criterion. Includes rule-based versus AI-native audit, two-way and three-way matching, why every invoice beats a sample, who decides what is disputed, five criteria for choosing a platform, and the path to production in under three months, explained by TetriXX AI's COO.
"Automated freight audit" describes very different things depending on who is selling it. At one end is a rules engine that compares an invoice to a rate table someone keeps up to date by hand. At the other is a system that reads invoices it has never seen before, checks every charge line against the contract that governs it, and hands a person only what needs a decision. Both carry the same label, and a buyer who does not separate them first ends up comparing vendors from different categories.
Table of Contents
- What does automated freight audit actually mean?
- How does an automated audit check a freight invoice?
- Why verify every invoice instead of a sample?
- Who decides which charges get disputed?
- What should you look for in an automated freight audit platform?
- How long does it take to reach production?
- Where does freya fit?
What does automated freight audit actually mean?
In the broadest sense, it is software that checks freight invoices against what was agreed without a person reviewing each invoice. The distinction that matters sits inside that definition.
Rule-based automation compares an invoice to a fixed set of rules: a rate table, an accessorial schedule, contract caps. It works while rate structures are simple and stable. It struggles with formula-driven surcharges, conditional clauses and providers whose invoices arrive in a format nobody configured, and it is only as current as the person maintaining the rules.
AI-native audit is a composition rather than one model with a product wrapped around it. Verification rests on deterministic checks and on heuristics that encode how freight is actually priced. AI carries the work no fixed rule set does well: reading a document nobody configured in advance, recognising a deviation pattern across thousands of lines, and putting the argument behind a provider scorecard or a renegotiation case.
The quickest test between the two is an invoice format the system has never seen. Software built on per-provider templates needs a new template for every carrier you add. A system that structures unseen documents does not, and that is testable in an afternoon.
How does an automated audit check a freight invoice?
Every invoice is read into structured, line-level data, matched to the contract, rate card and shipment record that govern it, and checked against the contracted rate, its formula and the surcharges. Deviations come out with the reason attached.
Two levels of matching decide how much the audit can see:
- Two-way matching compares the invoice to the contract.
- Three-way matching adds the shipment record. That is what catches a charge that is correct for a service you never received, and it is the step where tools in this category diverge.
On real freight the rules cannot be the same everywhere. A parcel lane with a fuel index behaves nothing like an ocean contract with detention terms, and a tolerance that is sensible in one currency is not in another. A serious platform lets you set which deviation categories clear automatically, which always surface, and at what threshold, per provider, mode and country.
Undercharges belong in the same check. Nobody raises a claim asking a provider to invoice more, so they clear automatically, but they are worth recording: an undercharge is usually a rate-card error that gets corrected in the other direction later.
Why verify every invoice instead of a sample?
Because the two answer different questions. Sampling estimates an error rate. Verifying every invoice finds the specific charges that were wrong, which is the only version you can dispute.
The gap is not proportional. A surcharge misapplied on 4% of shipments may never appear in a 5% sample, and if it appears once it reads as an incident rather than a pattern. The errors that cost most are small, repeated and individually unremarkable.
Verifying everything also leaves something behind that a sample cannot: every invoice, charge line and shipment on one record. Procurement, operations and finance can then work from the same data instead of three partial extracts.
Who decides which charges get disputed?
You do. A well-designed audit platform is human-in-the-loop: deviations are raised for review with the reason attached, and a person approves, rejects or corrects each one. Corrections feed back, so classification improves against your own contracts. Ask any vendor whether its platform disputes charges with carriers automatically, and whether that can be turned off.
How much reaches a person is a decision, not a property of the software. Across live freya deployments, the share of charge lines reviewed runs from under one in five hundred to roughly one in three. One operation lets small price differences clear below a set money threshold, and fewer than two charge lines in every thousand reach a person. Another approves no overcharge automatically, because each deviation is how it finds drift in its own rate cards, and about one line in eight is reviewed. Neither is more accurate than the other. They answer the same question differently.
So treat any single straight-through percentage with care. Ask whether the threshold is yours to set, and whether the figure is measured per invoice or per charge line.
What should you look for in an automated freight audit platform?
Five criteria decide the fit, and they are set out with the reasoning behind each in our freight audit comparison:
- Coverage: every invoice, or a sample?
- Carrier independence: does the provider hold carrier APIs or shared pricing data?
- Control over what reaches a person: per charge category, provider, mode and country.
- Tenancy: is compute and storage shared with other clients?
- Return: does recovery cover the platform?
Choose the type of provider first, then compare vendors inside it. A freight audit and payment bureau, a TMS module, a 3PL checking its own invoices and an AI-native platform are four different answers. The comparison page takes each in turn.
How long does it take to reach production?
The objective we plan against is production in under three months. What moves it is circumstance rather than effort: the scope you start with, the modes and markets in play, and how quickly each provider grants access to invoice data and rate cards. The first provider takes longer than the ones after it, which is why scope is phased in waves.
The usual starting point is a calibration on your own data: one or two providers and two to three months of invoices, under NDA. It establishes the error profile against your actual contracts, and it shows whether your invoice data supports continuous verification at all. Messy data is normal. Missing data, such as an invoice that never carries the fields needed to match it to a contract, is what blocks an audit, and the calibration finds that before anything is signed.
Implementations that run long do so for reasons other than the technology: rate cards that do not exist in usable form, a provider slow to grant access, no internal owner for the review queue, or a scope that starts with every provider and mode at once. All four can be spotted in advance.
Where does freya fit?
freya is TetriXX AI's freight audit and cost intelligence platform, in the AI-native category. It verifies every invoice as it arrives, across air, ocean, road and parcel, does two-way and three-way matching, holds zero carrier APIs, and lets you set what clears and what surfaces per charge category, provider, mode and country. It raises everything for review and disputes nothing on your behalf.
It also runs standalone: a first deployment can verify invoices against contracts with no connection into your systems, and integration then adds three-way matching and write-back. The questions buyers ask most are answered on the FAQ.
The fastest way to judge any platform in this category is to see it on your own invoices. Ask for a calibration on two or three months of yours.
