How we know what the charge should have been
Most invoice audits will tell you a charge looks wrong. The harder question is where the right number came from, because that is the one your carrier will ask. This page answers it.
A rate card is two documents, and only one of them is yours
What most sellers call their rate card is a two-page discount letter: percentages off named services, a minimum charge, fuel per the published index. The prices those percentages apply to live in the carrier's published tariff — which you may never have been sent.
So reading a rate card is two jobs with two owners, and we take the harder one.
| Part | Held by | How it is obtained |
|---|---|---|
| Published tariff, surcharge schedule, fuel index, remote-area postcodes, effective dates | Overbill | Maintained centrally, per carrier per period, from public documents. This is the part you cannot assemble yourself, and it is the same for every customer |
| Your discounts, minimums and special terms | You | Read from your own letter at setup, proposed on screen, confirmed by you before anything runs |
Two things we deliberately refuse to do
Both look attractive. Both quietly break an audit.
We do not derive your rate card from your invoices
It sounds efficient: read twelve months of bills and learn what you normally pay. But it infers what should have been charged from what was charged. A carrier that has been overcharging you consistently for six months would be learned as correct, and the audit would find nothing at all. Your rates come from your contract documents, always.
We do not discover rules automatically from your data
Anomaly detection finds outliers, not errors. A uniform overcharge produces no outlier whatsoever, and "statistically unusual" is not something a carrier credits. We use it internally as a prompt to go and look — never as a finding we put in front of you.
AI at the edges, determinism in the middle
This product ends with someone withholding money from a supplier. When the carrier asks why, the answer has to be a contract clause, a published tariff and a shipment record — not a model's impression. So the division is fixed:
AI reads and proposes
Which file is what, what your discount letter says, what an unusual clause should become. Every reading is shown to a person and confirmed before it counts.
The database executes and explains
Versioned rules over effective-dated reference data, identically each time, reproducible months later. A finding from March can be re-run in September and give the same answer.
A frozen, confirmed artefact
A confirmed mapping, a confirmed rate table, an approved rule. That artefact is what the audit stands on, and it is visible to you.
The clause that fits no table
Every real contract has one. Fuel capped at twelve per cent. Free returns under a kilogram. No residential surcharge inside the M25. These do not belong in a rate table, so they become rules — written from a plain description, run against your own history so you can see exactly what they catch before anything is approved, then versioned and kept.
The route is: your clause, in plain English → a rule → run over your last six months → this fires on 41 lines worth £820, read them → you approve → it runs every month from then on, unchanged, until you change it.
Nothing changes under you without being told
Carrier tariffs move. Fuel moves weekly. A surcharge appears with a new code. When any of that changes what you are checked against, it is written down: what changed, why, where it came from, and what it did to your money. You can object to any entry, and you can re-run a closed period to see the difference.
Silent drift in an audit engine ends with numbers from last March that nobody can explain — which is the one failure this product cannot survive, and the reason audit firms get distrusted.
See it on your own invoices
One month of invoices, your rate card and a shipment export. Free, and there is nothing to sign.