Seven accounts-payable controls, ranked by what they catch

16 September 2026 · Esta página é disponibilizada em inglês.

Anna Lindqvist

Rank them by what they find, not by the order a textbook gives. Reconcile every invoice to its own printed total, keep one identity per invoice number, compare each unit price with the price list you were given, and read a supplier's months together rather than one at a time.

Price Double Check does the third of those by email, on documents a buyer forwards. The other six stay in the office, and the order below is drawn from one acceptance run over eleven invoices and 382 lines, where the controls that found things and the controls that felt thorough turned out not to be the same set.

The order matters more than the list

Every finance team knows the full list of controls. What a small one lacks is the hours, so the list gets performed from the top until the day runs out, and the controls at the bottom are never run at all. Sorting the list by what it catches per hour is therefore the whole exercise.

One measured chain is not a statistical claim about all suppliers. It is, though, a run in which the arithmetic was checked line by line afterwards, so the ordering below rests on what was actually found rather than on what a control is supposed to be worth.

The seven, in the order that pays

1. Reconcile each invoice to its own printed total. Add the lines you have read and refuse to go further until the sum equals the total printed on the document. In the run every one of the eleven invoices agreed to the cent, and that is what made the other six controls trustworthy: a comparison of lines you have misread is worse than no comparison, because it is confident.
2. Give each invoice one identity: the supplier and the number, together. Three files in the run carried one number and one date with three different totals, and two more shared a number of which a credit note cancelled only the second. Without this control the same goods are paid for twice or the correction is lost.
3. Compare each unit price with the price list the supplier sent you, keyed on the item code. This is where the money was: 41 real price differences in 382 lines, one of them 17.75 per cent over the last list held and worth €47.76 on a single line.
4. Check the item code rather than the description. Of 442 lines compared in the wider set, 335 carried a description that did not match the price list word for word, so matching on names produces both false alarms and silent misses. Codes carry undocumented suffixes and repeat at different pack sizes, which is work, but it is the only key that holds.
5. Read a supplier's months together. One item was billed at €1.26 from March against a list price of €1.176, and €1.26 only appeared in a price list in October. Nothing in any single invoice showed that; it is visible only when the months are laid side by side, and it is the control a monthly routine skips first.
6. Compare each new price list with the one it replaces. Four items in the run had their pack size reduced at an unchanged price, which is a rise that no invoice can show, because invoices do not print pack sizes at all.
7. List what you could not judge, and send that list to a person. Fourteen lines in the run joined to nothing: seven of them freight, which has no reference price anywhere, and seven with no row in any price list. Left inside an automated check they become noise; written down as an open list they are a short conversation with the supplier.

What the run actually found

Two of those rows deserve a second look. The 41 differences are the case for running control three at all. The 101 recovered codes are the case for keeping everything a supplier ever sent you: those lines printed no code, and the code was recovered from the supplier's own documents higher up the chain, which only exists if nothing was thrown away. What that means for a team with no matching module is set out in the note on three-way matching without an ERP.

Three controls that are usually run and rarely pay

• Approval by an authorised signature. Somebody senior signs off a total. Nobody reads the lines, and every finding above lives in the lines.
• A percentage tolerance on the invoice total. A line 17.75 per cent over the list disappears inside a large order, because as a share of the total it is small.
• Rounding checks at the cent. Lists carry four decimals and invoices two, so a cent either way is arithmetic rather than a finding. Treat it as a finding and the team learns within a month to ignore the report.

None of the three is wrong in itself. They are simply cheap to run and, in a chain where somebody is quietly billing above the list, close to silent. A control that never finds anything is not a reassurance; it is an untested control, and it costs the same hour as one that works. What a running comparison of that kind costs, against the hour it replaces, is set out on the home page.

Forward one invoice and the price list it should have been billed on, and the third control on this list runs itself: the reply comes back by email with the lines that do not agree, and with the lines it refused to judge kept separate from them.