The 7 most common ROI calculation mistakes, and how to fix the performance model

The expert paradox

There is an irony in finance departments and management control teams. The professionals who rigorously validate marketing business cases, scrutinise industrial investments and arbitrate other departments’ transformation budgets are, in practice, the most exposed to modelling errors when evaluating the return on investment of software meant for their own processes: the Procure-to-Pay cycle or supplier invoice matching.

The cause is simple. We evaluate first what is visible and immediate: the licence cost, integration fees, the budget line. We quantify far less naturally what the organisation loses every day without the tool. The cost of inefficiency, latent risk and scattered human time are diluted into operational routine. They are real, but nobody adds them up.

Here are the seven most common mistakes in building the ROI of procurement-finance software, and for each one the correction to apply to the model.

Mistake 1: comparing licences with each other, not with the cost of manual processing

This is the most widespread mistake. The cost of the old solution is compared with the cost of the new subscription. If the licence is more expensive, the project is dropped. Yet the expense line of an accounts payable team is not IT infrastructure: it is human time.

The real cost of manual matching includes time spent on data entry, verification and routing for approval, plus the inherent error rate (duplicate payments, keying errors) and the time spent correcting them afterwards. A correct model starts from the full cost of processing an invoice in your organisation, measured on your volumes, and compares it with the cost once the tool is in place. The licence differential is just one line in that calculation.

Mistake 2: ignoring the cost of undetected supplier disputes

A discrepancy between purchase order, goods receipt and invoice is not an accounting line on hold. It is a dispute in the making, a potential credit note forgotten, and often the starting point of a supplier relationship that deteriorates. A supplier paid late because of internal friction may delay deliveries, tighten terms at the next renegotiation or withdraw a preferential rate.

The model therefore has to include two things: the time spent resolving discrepancies, valued at fully loaded hourly cost, and the amounts paid wrongly for lack of detection: overpayments, discounts not applied, credit notes never recovered. On the perimeters we have analysed, recovered margin sits between 1 and 7% of the analysed perimeter. An honest ROI includes that shortfall, and documents it line by line rather than estimating it.

Mistake 3: demanding ROI within 12 months and ignoring the ramp-up

The habit inherited from on-premise investments pushes organisations to demand profitability in year one. If the curve does not cross break-even within 12 months, the project is deemed risky. That reading ignores how an agentic AI solution creates value.

Unlike a fixed-rule engine or traditional OCR, which quickly reach their limits, an agent platform reads your document formats without prior configuration and relies on a repository that grows richer as contracts, orders and invoices are linked to it. The controlled perimeter expands, from 5% to 100% of your purchases checked, and each extension adds detected discrepancies with no marginal data-entry cost. Value is cumulative; the model has to represent it over three years, not one financial year.

Mistake 4: forgetting the opportunity cost of the teams involved

While your buyers and management controllers manually tick invoice lines to verify discrepancies of a few cents, what tasks are they not doing? Opportunity cost is one of the most powerful profitability levers and the most systematically ignored.

The point is not only to price a task that disappears, but to value the task that replaces it. What does an hour of a buyer’s time yield when spent on sourcing or renegotiating a framework contract? What is a faster monthly close worth in cash visibility? Time given back to teams (from 15 to 45% per FTE, depending on perimeter and data maturity) only has value if the model states what it is reallocated to.

Mistake 5: undervaluing compliance risk as a financial lever

Compliance is often perceived as a legal obligation rather than a financial lever. As long as there has been no painful tax audit, the risk is valued at zero. That is a major valuation error in a regulatory context that tightens every quarter.

Mandatory electronic invoicing, algorithmic tax audits, the CSRD and its ESG reporting, the French LME Act on payment terms: a badly matched line, a fake-supplier fraud or the absence of a reliable audit trail cost far more than a keying error. Avoided crisis costs belong in the ROI. They are modelled like insurance: probability, exposed amount, cost of cover.

Mistake 6: comparing two tools rather than the tool with the status quo

During a tender, thinking narrows to a feature battle between vendors. That is a framing bias: the real competitor to your project is neither solution B nor an ERP module. It is your current manual process. It is inertia.

Ask the question head-on: what does doing nothing cost this year? Wage inflation on positions assigned to repetitive tasks, the departure of people who refuse to work with outdated tools, the loss of responsiveness against competitors who are already automated. The investment in the tool is subtracted from the cost of maintaining the status quo, and that cost rarely appears in the model.

Mistake 7: leaving procurement out of the calculation

This is the historic blind spot of finance departments. Matching software budgeted by the CFO is evaluated solely on accounting indicators. As a result, half of the real ROI is ignored.

The Procure-to-Pay cycle is an unbroken chain. Friction in accounts payable almost always originates upstream, in the act of purchasing. A solid business case is built with procurement and includes their indicators: working capital, visibility on commitments, and early-payment discounts. Invoices matched and approved quickly open the door to the early-settlement discounts written into the contract. That is money left on the table for lack of velocity, and it is measurable on your own negotiated terms.

A three-pillar model

Calculating the ROI of a procurement-finance automation solution is not a matter of subtracting a licence cost from an hourly wage. To convince an executive committee, structure the model around three pillars:

• Avoided costs: less time on manual entry and verification, no regulatory late-payment penalties, errors corrected at the source.
• Direct gains: credit notes recovered, duplicate payments detected, early-payment discounts captured.
• Strategic value: time reallocated to analysis, healthier supplier relationships, a secured audit trail.

Each pillar must rest on your data, not on a market average. That is the only way to produce a business case a management controller cannot take apart.

This is the logic of cumulative value in which AI agents operate: they read the documents, confront each invoiced line with the clause that governs it, show the discrepancy with its evidence and leave the decision to a person. ROI stops being a theoretical promise once it is measured on a real perimeter. The 3-week diagnostic on your data exists for exactly that: establishing the three pillars with your figures, before any purchasing decision.