What an invoice error costs, and which figures survive checking
· Esta página é disponibilizada em inglês.
Published benchmarks put the median share of invoice lines that match first time at 75 per cent, and the median share of supplier invoices that are right first time at 80 per cent. On the real chain behind this service, 41 of 382 invoice lines carried a genuine price difference against the price list.
The email that names them is this service's whole output. Price Double Check takes an invoice and the list it should have been billed on, both forwarded by a buyer, and replies with the lines that disagree. The figures below are its own, taken from the acceptance run, set beside the only two published ones this piece could trace to a stated measure and a sample.
That last clause is most of the point. Search for the cost of invoice errors and the same two numbers come back on page after page: a share of invoices said to contain errors, and a cost per invoice processed. Follow either to its source and the trail runs into a blog post citing another blog post. Neither is printed here.
The two published figures with a method behind them
APQC maintains a library of benchmarking measures, each with a definition, a sample and quartiles. Two of them speak to this directly. The share of invoice line items matched the first time has a median of 75.0 per cent across 4,438 organisations. The share of supplier invoices that are right first time has a median of 80.0 per cent across 476 organisations.
The two are not interchangeable, and the difference between them is the whole subject. One counts documents, the other counts lines. A buyer pays for lines. An invoice of forty lines with one wrong price is a wrong invoice by the first measure and is 97.5 per cent correct by the second, and the cheque that goes out is wrong by whatever that one line is worth.
One real chain, counted line by line
The set behind this service is a sports-nutrition manufacturer selling to a Baltic distributor, and that distributor selling on. Eleven invoices in scope, three price lists, orders and a credit note. Every figure below comes from the acceptance run over those documents rather than from a survey.
| Measure | This chain | Published median |
|---|---|---|
| Invoice lines compared | 382 | — |
| Lines matching the list exactly | 327, or 85.6 per cent | 75.0 per cent of line items matched first time |
| Lines with a genuine price difference | 41, or 10.7 per cent | — |
| Lines that joined no price-list row | 14, or 3.7 per cent: 7 freight and 7 unlisted items | — |
| Lines read incorrectly | 0 | — |
| Invoices reconciling to their own printed total | 11 of 11, to the cent | 80.0 per cent of invoices right first time |
| Item codes recovered from documents further up the chain | 101 | — |
Two of those rows carry most of the weight. The fourteen unjoined lines are not errors and must never be reported as such: freight has no reference price anywhere, and an item that appears in no list is an absence of basis rather than an overcharge. And the eleven reconciliations are what make the other rows worth reading, because a document whose lines do not sum to its printed total has been misread, and a tool that cannot tell those two cases apart is a tool that invents findings.
The cost is not spread across the lines
A rate of one line in nine sounds like a tax on every delivery. It is not how the money behaves. Four of the five findings the owner of these documents had missed by hand were worth a few euros each, and one was not.
- One item billed at €13.23 against €11.235 in the last list received: 17.75 per cent over, €47.76 on a single invoice, and enough on its own to take 91 per cent of the margin on that delivery. The reason it went unseen is that the current year's list had never been sent, so the newest file on hand was a year old.
- One item substituted without notice, from a 550cc pack at €2.07 to an 800ml pack at €2.59, worth €15.60. No arithmetic finds this, because both prices are right for their own codes.
- Four pack sizes reduced at an unchanged price. These appear in no invoice at all, because invoices do not print pack sizes. They are visible only by reading two price lists against each other.
- One invoice number arriving as three files with one date and three different totals, of €3,472.70, €3,482.25 and €3,480.34. The cost of that one is whichever of the three gets paid.
- One price that became normal without being agreed: a bar billed at €1.26 from March against €1.176 in force, with €1.26 entering a list only in October.
So the honest unit is not a cost per invoice. It is a cost per line, and the distribution has a long tail: most differences are worth less than a cup of coffee and one of them was worth most of the delivery. That is also why findings against a price list are ranked by percentage and by novelty rather than by euros — a list carries no volume — while findings on invoices are ranked by euros, where the spending is real. The margin arithmetic is worked through in what a billing error does to a distributor's margin.
What these figures do not establish
One chain, two accounting systems, thirty documents. That is a demonstration, not a study, and it would be a poor argument for any general rate. A supplier with tighter price control would produce fewer differences, and a supplier who revises prices monthly without sending the revisions would produce more.
What it does establish is narrower and more useful: that the differences were real, that they sat in ordinary documents from an ordinary supplier, that a person who knew the business had read those documents and missed five of them, and that nothing about finding them needed more than the documents themselves.
Em resumo
- Why a cost per invoice processed is the wrong figure to track
- It measures handling: keying, routing, approving, filing. All of that can be made cheaper while every price on every invoice stays unchecked. Capture and comparison are different jobs, and the second is the one that moves money.
- What share of differences a buyer can expect to be genuine
- In this run, 41 of the 55 lines that did not match exactly were genuine price differences and 14 were categories with no basis for a verdict. A tool that reports all 55 as problems is reporting a quarter of its output as false, which is roughly the point at which the reports stop being opened.
- How long a buyer has to turn a difference into a claim
- European law on late payment contemplates a procedure of acceptance or verification of the goods and caps it at thirty calendar days unless the contract provides otherwise. A difference found inside that window is a question to the supplier. Found afterwards, it is a request for a credit note.
Forward one invoice and the price list it should have been billed on, and the count comes back by email for your own supplier rather than for somebody else's sample: the lines that matched, the lines that did not, and what each difference is worth. The prices are on the home page, and the controls this check sits inside are in seven accounts-payable controls a small company can run.