Most rebate programs don't fail loudly. There's no crash, no alert, no bad day. The quarter closes, the numbers look fine, and everyone moves on. Then — months later, often at year-end, sometimes during an audit — something surfaces. An accrual that was never released. A claim that was rejected for a reason no one can reconstruct. A promotion that kept paying out long after it should have stopped. The money was leaking the whole time. Nobody could see it, because nobody could see how the numbers were made.
That invisibility is the actual problem. And it's structural, not careless.
Spend any time inside rebate operations and the same failure modes repeat, regardless of how careful the team is.
A distributor submits a bill-back claim. Your ERP has the product as ABC123; their system recorded it as ABC-123. Your automated reconciliation treats those as two different products and rejects the claim. No one is notified — the distributor assumes you're slow-paying, you assume the claim was invalid, and weeks later it's a dispute. In one case documented by the rebate-software firm AIMDek, a single manufacturer found that 18% of valid claims were being rejected purely because of product-ID mismatches — pure leakage, hidden inside a process that looked like it was working.
Or: a promotional rebate is set to end September 30. Nobody turns it off in the system. It keeps accruing into the next quarter. There's no workflow to flag the expiration and no rule to kill the program automatically, so — again per AIMDek's account — finance discovers it four months later, during a historical payout review. By then the over-accrual is baked into closed periods.
Underneath both is the same root cause: the contract terms live in PDFs and email threads, not in executable logic. A PDF can't enforce a tier threshold, an expiration date, or an exclusion rule. The moment reality drifts from what someone remembers the contract saying, execution drifts with it — and reconciliation becomes manual detective work every single cycle.
This is why the consequences tend to land at audit. As the rebate platform Enable has noted, most ERP systems can't cope with the complexity of real rebate deals, and reconciliation failures of this kind "raise audit concerns" — with errors going unnoticed until year-end, at which point a rebate balance can become uncollectible and spill into the next financial year. The leak doesn't just cost money. It costs the ability to stand behind your own numbers.
The instinct, once the leaks are visible, is to buy your way out with a big implementation — an enterprise settlement or revenue-management platform, a systems integrator, a multi-year program. Sometimes that works. Frequently it doesn't.
Roughly 70% of ERP projects fail to achieve their goals, according to accounting and advisory firm CLA, and cost overruns on these programs routinely run well beyond the original estimate. The failure case currently working its way through the courts is instructive: in a 2024 lawsuit, the medical-device maker Zimmer Biomet alleged that its ERP go-live slipped five times — from February 2023 to mid-2024 — and that it incurred tens of millions of dollars in remediation while the system still suffered significant defects. The detail most analyses miss is structural: the suit alleges that roughly $50 million of the integrator's $63 million in fees were tied to time-based milestones rather than measurable business outcomes. An integrator paid to hit dates has every incentive to hit dates — whether or not the system actually works when it gets there.
You can spend a hundred million dollars and still not get control. The UK's National Health Service abandoned a large finance and HR ERP program after spending over £100 million, with limited benefit, because the software never aligned with how the organization actually worked.
The expensive fix and the quiet leak are the same disease wearing different clothes.
Look closely at every failure above — the rejected claim, the runaway promotion, the audit surprise, the implementation that never landed — and they share a single property: you cannot reconstruct how a number was reached.
The truth is scattered. It lives in spreadsheets, PDFs, email threads, a batch job nobody fully remembers configuring, and someone's memory. When a figure is questioned — by a partner, by finance, by an auditor — answering means archaeology. And when numbers can't be traced back to their inputs, leaks have somewhere to hide. That's the whole game. The leak isn't enabled by carelessness; it's enabled by opacity.
The alternative isn't a better spreadsheet or a bigger consulting project. It's a different starting principle: the record of how a number was produced should be created as the number is produced — never reconstructed afterward.
In practice that means a few specific commitments. An append-only ledger, where nothing is silently overwritten. Inputs frozen at the moment of calculation, so you can always see exactly what a figure was computed from, even a year later. And the ability to start from any number — an accrual, a settlement, a payment — and drill straight through to the source transactions and the contract terms behind it. When that's true by design, a leak has nowhere to hide, and an audit stops being an excavation and becomes a query.
That principle — audit-defensible by construction — is the foundation Aurgus is built on. Two things we're building from there follow directly from it: configuration handled by an AI that already understands rebate structures, rather than a months-long consulting engagement, and pricing designed for companies that have outgrown spreadsheets but shouldn't have to buy an enterprise suite to get control. Those are the direction we're heading. Traceable-by-construction numbers are where we start, because everything else is decoration if you still can't defend the figure.
If your rebate numbers currently live across spreadsheets, PDFs, and a batch process no one fully remembers setting up, the question isn't whether there's a leak. It's where, and how long until it surfaces.
If you'd find it useful to talk through where yours might be hiding — not a demo, a conversation between people who've seen how these break — we're happy to spend twenty minutes on it.