Follow the money.
The savings live in the gap between
a payment and its check.
A $120M revenue company saved $3.6 million with OHM — 3% of outflow. That number sounds implausible until you follow one invoice from the mailbox to the payment run and watch where money escapes at each step. This is that walk-through: where the leaks open, why the controls you already run miss them, and what changes when every transaction is checked before the money moves.
The flow of money — and the four places it leaks
Money-out is a pipeline: an invoice arrives, gets entered, gets taxed, gets booked, gets paid. Each stage asks questions nobody has time to answer on every invoice. The leak is the sum of the unanswered ones.
The invoice arrives — and the first leak opens
Have we seen this invoice before? Is the price the price we agreed? Do the PO, the goods receipt and the invoice agree? Duplicates rarely look identical — a re-sent PDF with a new number, the same delivery billed by two group entities. Checking the price means opening three documents, and on invoice 400 of the week nobody opens three documents. The Duplicate, Pricing and 3-way match agents answer all three questions on every invoice, at entry, in under a second — an invoice that fails is held with the documents that prove why.
Tax and compliance — the leak that compounds
A credit claimed before the vendor files reverses with interest; one never claimed is cash left with the government. Withholding at the wrong rate is money out the door or a notice later. A small-vendor clock that breaches accrues statutory interest automatically. None of these is a judgment call — they are rule checks against registries and filings that fail in batch processes for one reason: the data changes daily and the check runs monthly. The GST, TDS, PAN and MSME agents verify each entry against live status before it posts.
The books — where errors go to hide
Payments go out daily; reviews happen at month-end. So there is, on average, a month between an error and the review that finds it — and by then the money has moved and recovery means emails, debit notes and leverage you no longer have. With OHM, entries post to the ERP already compliant — checked, matched, inside approval limits — so the book stays clean without a cleanup pass.
The payment run — the last gate, checked twice
The run is where every upstream mistake becomes irreversible. It is built from approved entries only, then checked again before it leaves: duplicates across the run, amounts against limits, expiring discounts captured rather than forfeited. Exceptions are not escalated into someone's inbox to age — they are held, with the evidence attached, until a person decides.
Why the controls you already run miss it
Not because your team is careless — because the volume makes conscious checking impossible. 85% of bills are approved without a conscious check; approval becomes a keystroke. And the checks that do run, run on a sample, after posting, because matching every line manually doesn't scale.
Recovery after payment is pennies on the dollar. Prevention before payment is the whole dollar.That is the entire economic argument for real-time checking. The 3% was never one big leak to find — it is many small ones, each individually invisible, escaping through gaps in timing. Close the timing gap and the money simply never leaves.
- Every transaction, not a sample. A sampled control catches the error rate of the sample. A leak spread thin across thousands of invoices needs every one checked to show up at all.
- Before the money moves, not after. The same finding is a prevented loss before the run and a collections project after it.
- Held with evidence, not escalated. An exception that arrives as an argument gets waved through. One that arrives with the contract, the PO and the earlier invoice attached gets decided.
- No negotiation required. Recovering the 3% needs no vendor renegotiation and no headcount change — it was always your money, leaving through timing.
The full breakdown — six ordinary failures, what each is worth on $120 million of outflow, and which agent catches it — is on What it found. The short version: duplicates and double payments, contract and PO price variances, credits missed and rates misapplied, discounts forfeited. Individually, rounding errors. Together, $3.6 million.
"Why hasn't software fixed this already?"
Fair question — ERPs have had 3-way match for thirty years. Three independent studies point at the same answer, and none of the blockers they name is a model problem.
| Finding | What it means | Source |
|---|---|---|
| Adoption is flat — 59% of finance departments using AI, against 58% a year earlierA year of intense attention moved the number by one point. The pilots are not converting into production. | A year of intense attention moved the number by one point. The pilots are not converting into production. | Gartner, Nov 2025 · n=183 CFOs and senior finance leaders |
| Only about a third of organisations reach maturity on agentic governance and controlsSecurity and risk is the number one barrier to scaling agents, cited by nearly two-thirds. | Security and risk is the number one barrier to scaling agents, cited by nearly two-thirds of respondents. | McKinsey, State of AI Trust 2026 |
| Three infrastructure obstacles hold agentic AI back in financeLegacy system integration, data architecture constraints, and governance and control frameworks. | Legacy system integration, data architecture constraints, and governance and control frameworks. | Deloitte, Tech Trends 2026 · CFO guide |
The savings above only materialise if an agent is allowed to touch every transaction — and that is a governance question, not a model one. It is why OHM is built as an ecosystem for deploying finance agents rather than an AP tool: permissions resolve before the model runs, the ERP stays the system of record, ambiguity has a defined destination, and the audit trail lives once, underneath every agent. The checks run where the money already lives — SAP, Oracle, NetSuite, Tally, Zoho Books, QuickBooks, Sage, Odoo — starting from an extract your team already produces. How that governance works in practice is on Security.
Sources
- Gartner, survey of 183 CFOs and senior finance leaders, November 2025 — AI adoption 59% vs 58%; use cases: knowledge management 49%, AP automation 37%, anomaly detection 34%; leading obstacles data literacy and data quality. Reported by CFO Dive.
- McKinsey, State of AI trust in 2026: shifting to the agentic era — AI Trust Maturity Survey, ~500 organisations, Dec 2025–Jan 2026.
- Deloitte, Tech Trends 2026 and the accompanying CFO guide — infrastructure obstacles to agentic AI in the finance function.
The Gartner figures are quoted from CFO Dive's reporting of the survey, with the sample size and date as published. The McKinsey and Deloitte findings are summarised from their published research. The 85%, one-month and 3% figures describe what we observe across OHM deployments; the savings-check exists so you can test them against your own data rather than take them from us.
The argument is only worth as much as your own data.
Ninety days of invoices, one export, findings back within a working day.
$Check your savings→