The rubber stamp

It’s the third of the month and the accounts clerk has four thousand carrier invoices to approve by Friday. She has one pair of eyes, a spreadsheet, and a rate card she hasn’t had a clear hour to open since the contract was signed.

She checks the first few properly. Fuel levy looks right, the freight lines look about right, the accessorials are the usual mess. They pass. So she does the only thing the workload allows: she approves the batch and moves on. Checking every line on every invoice against the contract, by hand, is not a job a person can do in the time the month gives her.

The leak you can’t see

Here is the uncomfortable thing about that batch. If one carrier overbilled you four hundred dollars on a single invoice, she’d catch it, because a single wrong number stands out. But nobody overcharges you that way. It comes in as a rounding-up here, a wrong zone there, a fuel levy calculated on last quarter’s rate, a re-delivery fee that shouldn’t apply, spread thin across thousands of lines where each one looks plausible on its own. You can’t overpay on one invoice and notice. You overpay a little on thousands, and you never see the individual dollar leave.

Freight-audit benchmarks and operator interviews put the typical carrier-invoice overpayment somewhere between three and eight per cent of total freight spend. Take an operator we’ll call Downs Freight, a roughly sixty-truck interstate carrier and 3PL with a $22 million annual freight bill. Downs Freight is an illustrative composite of the businesses this is built for, not a real RMAI client. On that bill, three to eight per cent is somewhere around $660,000 to $1.7 million a year, leaking out through invoices nobody has the hours to check line by line. That range is illustrative arithmetic on their bill, not a figure anyone has measured for them. But you can see how a number that large hides so well. It never arrives as a shock. It arrives as a rubber stamp.

Four leaks, one Friday

The invoice is only the leak with a dollar sign you can name. Walk into Downs Freight at 4:40 on a Friday and there are four running at once.

The 1,400-line sub-contractor invoice nobody can check is one. Beside it, about two hundred proof-of-delivery dockets are stacked as PDFs and phone photos, waiting for someone on Monday to re-key them into the TMS by hand. A shipment throws off seven to ten documents, and re-keying them swallows two to four hours a day per person on the desk. The phone won’t stop, because a quarter to half of all inbound customer contact is somebody asking where their pallet is, and it spikes exactly when everyone is busiest. And in the yard, a driver is about to be rostered onto a Monday run that would push him past his legal fatigue hours, which matters because a Category 1 breach of the Heavy Vehicle National Law carries a maximum corporate penalty of $3.55 million and exposes directors personally.

Four threads, four different anxieties, four different people carrying them. They feel unrelated. They are the same problem wearing four hats: work that is too high-volume and too detailed for a human to check every instance of, so it gets stamped, stacked, or answered on the run.

Which leak you can actually plug

The instinct at this point is to ask “should we get some AI in here?” That’s the wrong question, and it will send you shopping for the wrong thing. The right question is narrower and far more useful: which of these four leaks lives in a system an outside agent can actually read and write? Because that, not the cleverness of the model, decides which one pays back first.

We mapped the Australian logistics stack for exactly this. The pattern is clear once you see it. The systems that hold an operator’s own data and open the door to it are the Australian-built layer. CartonCloud exposes a full REST API with OAuth2 and a public Postman collection, read and write, across customers, orders, consignments and documents. The multi-carrier shipping platforms (MachShip, Shippit, Starshipit, SmartFreight, and Australia Post) let an agent create consignments, pull labels and track. Geotab opens your own telematics and fatigue data through an open API and SDK. Xero gives you OAuth2, native webhooks and an official integration server over the ledger. TransVirtual, Datapel, EROAD and the other telematics engines sit just behind them, reachable with a little more friction. This is the layer where the invoice, the PODs, the shipment status and the driver-hours data all actually live.

Then there’s the closed side of the map, and honesty about it is the whole point. The government rails, the NHVR compliance portal and the Australian Border Force customs system, don’t expose your data at all. You lodge into them through accredited software; you don’t read out of them. The tier-1 enterprise systems (Manhattan Active, CargoWise, WiseTech’s eHub) are genuinely capable but reachable only through your own enterprise tenancy or partner-certified developers, and often through old XML and EDIFACT feeds rather than a modern REST API you can just switch on. Two vendor API claims we checked didn’t survive verification at all: Microlistics and Infios turned out to be integrator-mediated EDI only, so we won’t tell you we integrate with them, because we can’t. And the “AI” your own vendors already sell you, the in-product assistants like Geotab Ace or Xero’s, is the one surface an outside agent cannot reach or reconfigure. You can buy it. You cannot build your reconciler on it.

So the freight-invoice leak is not just the biggest. It sits in Xero and your TMS, which are wide open. That’s why it’s the first build, not the cleverest one.

What actually changes

Put the four demos next to the four Friday threads and the shape of the work becomes plain.

An invoice reconciler reads every one of the four thousand invoices against the actual rate card. It matches each line, and where a charge doesn’t fit the contract it quotes the exact clause it used, drafts the dispute, and hands it to a person. A POD checker reads the two hundred dockets, confirms the signature is present, compares the GPS stamp to the delivery address and the docket to the manifest, writes the clean structured fields into the TMS, and surfaces only the exceptions. A shipment-status responder answers the “where’s my pallet?” calls from the TMS and carrier feeds, cites the milestone it’s quoting, and when a feed is stale it says so and flags a human rather than inventing a location. A Chain-of-Responsibility monitor runs deterministic rules against the legislated HVNL fatigue windows, reads the driver logs and telematics, and flags the breach before the shift is rostered, not after the logbook lands.

In every one of those, the machine reconciles and flags. A named person decides. Nothing is auto-disputed, nothing is auto-paid, no compliance flag is auto-lodged. That isn’t a courtesy line, it’s the architecture, and it’s load-bearing for the failure mode that actually destroys value: not the overcharge the tool flagged wrongly, but the one it quietly passed as clean. So low-confidence lines route to a person by default rather than auto-passing, a sample of the auto-passed set is checked, and there’s an audit log of what was not flagged, so a silent miss is reviewable. For fatigue in particular, the check is deterministic rules against the standard, BFM and AFM windows in the legislation, never an LLM’s guess.

The honest ledger

Two things I won’t dress up. RMAI is new to this sector, and there is no first-party logistics result to show you yet. Every number in the proof column is third-party or vendor-reported, and labelled that way. Ofload, on AWS, cut its average cost of transport by fifteen per cent. Polaris reports clearing eighty-five per cent of customs documents automatically and dropping order creation from fifteen or twenty minutes to about two. Schaeffler ran an OCR and AI three-way match across three thousand-plus freight invoices a month and eliminated the carrier-overbilling leakage. A document pipeline at n8n processed thirty thousand PDFs in two weeks at ninety-four per cent accuracy, human review only on the flagged ones. Hunter Express, on TransVirtual, lifted revenue per consignment double digits in the first month and redeployed eight back-office staff. These are what the pattern does elsewhere. They are not promises about your numbers, and the free 30-minute discovery call is exactly where we walk the pattern on your own carrier contracts and rate card first.

The second thing: where your data goes. The build runs inside your own tenancy, your TMS, your document store, your driver and compliance records. It stays onshore in Australia, access is role-based, and it’s never used to train any third-party or public model. Where a frontier model is in the loop we’ll name plainly what reaches it and under what terms, and a DPA is on the table. As for cost, a focused build ships in three to six weeks in the $10,000 to $60,000 band, quoted fixed-scope after the discovery call, on your numbers. One tool, say the invoice reconciler, starts from about $10,000. On recovered admin time alone the payback runs roughly nine to fifty-six months; the freight-overpayment lever sits on top of that and typically shortens it sharply. That last part is illustrative and re-baselined on the call, not a quote.

Where to start

You don’t need a transformation programme to find out whether this is real for you. You need one honest look at where your freight data actually lives, and which of your four leaks sits in a system an agent can reach.

Two ways in, both low-risk:

• Read the map first. We’ve written a plain-English brief on the Australian logistics software landscape, every major system rated for what an agent can actually read and write, and what “closed” really costs you. → https://realmindsai.com.au/guides/logistics/
• Book a free 30-minute discovery call. We’ll take one real path, your carrier invoices or your POD flow, name the system underneath it, and show you the highest-value place to close the gap, before anyone promises a build. → https://outlook.office.com/book/[email protected]/?ismsaljsauthenabled

There’s a batch of invoices being approved right now, somewhere in your operation, that nobody has the hours to check line by line. The only question worth asking is whether the few per cent leaking out of it is ever going to become visible enough for a person to decide what to dispute.