Deselected

Start with one invoice line. A fuel delivery is priced from the wrong grid: $572 of overcharge. On its own, that line is beneath any recovery audit’s claim threshold. The labour of documenting and pursuing it costs more than it returns. Now run the same configuration error across a year of deliveries: 2,000 occurrences, $1,143,366. Nobody looked for it. Not because it was hidden, but because at $572 a time, it was deselected.

The classic recovery audit does not miss small recurring errors. It deselects them, rationally, because its business model prices them out.

The deselection has a second-order cost. When the audit reports, its findings are read as the size of the problem: the engagement returned so much, therefore that is what was there. A clean-looking audit can coexist with a recurring seven-figure leak, and in our experience of the categories, it usually does.

The pricing decides what gets checked

Recovery audit is paid on contingency: a share of what it finds. A finding takes days to document, substantiate, and negotiate, and those days must be paid for by a percentage of the amount recovered, so a finding is only worth pursuing if it is large enough to carry its own labour. The audit therefore samples: the largest suppliers, the categories with a history of findings, the deviations large enough to see without recomputing anything. Coverage of 5 to 15% of invoices is not a methodology choice but a labour-allocation decision, and under contingency pricing it is the correct one.

This is competent work, done by people who know the categories well. It is also, by construction, blind to the error patterns that live below the threshold: wrong grid selections, stale index bases, surcharges that outlived their clause. Which are precisely the patterns that recur, because nothing about finding one instance stops the next. And a firm paid a share of what it finds has no revenue interest in the error not recurring.

And the calendar decides the outcome

The second constraint is time. Engagements typically review invoices 18 to 36 months after payment. At that distance, a correction has become a claw-back. The buyer who negotiated the clause has changed roles. The supplier’s account manager has too. Contractual claim windows have closed or are closing. The commercial relationship has moved on, and reopening a three-year-old line item puts a strain on it.

Documentation decays on the same calendar. The delivery notes are archived, the email that carried the applicable grid left with its inbox owner, and the person who could confirm which amendment was in force in March of that year no longer works on the account - on either side.

A deviation raised weeks after invoicing, against a live contract, is a correction between partners. The same deviation raised three years later is a claim between adversaries.

And while the calendar runs, so does the error. A misconfigured grid found in an audit of year N-3 has, by definition, been billing wrong for three more years, through this year’s budget and into next year’s. Nothing in the engagement touches the flow.

Reconciliation now covers past and present

Stock and flow used to be different problems, and the division of labour followed: an audit firm to claw back the closed years, nothing at all on the running ones.

But controlling the flow means loading the contracts and recomputing every line against what was signed - and that is the same computation pointed backwards. The first run of a continuous system over three closed years of invoices is a historical audit, with one difference: it has no claim threshold. Nothing is deselected. The $572 pattern is found in year N-3 for the same reason it is caught in the current quarter: every line is recomputed, not sampled. And the finding arrives with its evidence already attached - the clause, the grid, the line, the delta.

What sits outside the engine is the prosecution of the hardest claims: cross-border VAT, duty recoveries, disputes that end in front of a mediator. That is work for lawyers, and the deviation ledger is what you hand them. Engaging a recovery auditor was the right decision under the old constraint. The constraint moved; the decision can too, without the first one having been a mistake.

The difference that remains is what happens after the first pass. An audit ends, and comes back in three years to find the same errors, three years older. A standing control returns a number every cycle: how much of this quarter’s spend matched the contract, and which lines did not, while they can still be corrected. The $572 line only matters under that question. It occurs 2,000 times a year either way.