Process Automation ROI: Turning Hours Saved into Money You Can Defend

Most automation business cases follow the same recipe: count the hours a task takes, multiply by an hourly rate, present the product as annual savings. It is quick, it produces a big number, and it falls apart in the first ten minutes of a finance review. The gap between hours saved and value realised runs through four checkpoints — real working hours, fully-loaded cost, utilisation, and ramp time — and each one has published evidence you can anchor to.

Start with real working hours, not 2,080

The default divisor in most spreadsheets is 2,080 hours a year (40 hours × 52 weeks). Nobody works that. The US federal government's own OPM divisor is 2,087 hours — but that is gross paid hours, before any leave, holiday or sickness comes out.

Hours actually worked are much lower, and they vary sharply by country. OECD data for 2025 puts average annual hours actually worked at 1,800 in the US, 1,533 in the UK, 1,498 in France and 1,332 in Germany, against an OECD average of 1,736. If your model prices a UK employee's hour using a 2,080-hour year, you have understated the value of each hour by roughly a quarter — or, seen the other way, you may be claiming more recoverable hours than exist. Use the figure for the country you operate in, and say so in the assumptions.

Price the hour at loaded cost — but call it a floor

Salary divided by hours is not what an hour of labour costs. The BLS Employer Costs for Employee Compensation series shows US private-industry benefits running at about 30% of total compensation — wages of $32.36 per hour worked against total compensation of $46.15, a multiplier of roughly 1.43× on wages. A salary × 1.4 loading is therefore citable to a government statistical series. It is also a floor: it excludes recruiting, training, equipment and office overhead. Present 1.4× as the defensible minimum, not the true figure.

Utilisation honesty: saved hours are not redeployed hours

Here is where most cases quietly overstate. Automating a task that took 30 minutes a day does not put 30 minutes of productive labour on the balance sheet. Freed time fragments across the day; some of it is reabsorbed by other low-value work; none of it reduces payroll unless headcount, hiring plans or overtime actually change.

An illustrative example. Suppose automation removes 2,000 hours a year of manual processing across a UK team. At a £22 hourly wage and a 1.4× loading, the naive claim is 2,000 × £30.80 = £61,600 a year. An honest model applies a utilisation factor — the share of freed time that converts into avoided hiring, reduced overtime or genuinely redeployed capacity. At 50% utilisation the claim is £30,800; at 30%, £18,480. The gap between the naive and honest versions is not pessimism. It is the difference between a number finance will sign and one they will strike out.

The measured evidence supports restraint here. Deloitte's UK robotics study found implementers achieved an average 16% overall cost reduction — every organisation surveyed was confident of financial benefit, but the measured average was 16%, not 40%. Deloitte's 2022 intelligent automation survey of 479 executives reports a 32% average cost reduction — but more than half of respondents had never calculated their actual reductions, so 32% comes from the self-selected minority who measured. A 16%–32% band, conservative to optimistic, is what the published record will bear.

Ramp time: value does not start on go-live day

The same Deloitte 2022 survey found average pilot payback lengthened from 16 months in 2020 to 22 months by 2021/22. Automations need exception handling, process stabilisation and adoption before they deliver at full rate. A model that books 100% of annual savings from month one will show a payback period the evidence says is unrealistic. Phase the benefit in — and let the payback land where it lands.

Where automation value is best documented

Document-heavy back-office processes have the strongest published benchmarks. Ardent Partners puts the all-in average cost of processing an invoice at $9.40, against $2.78 for best-in-class organisations — with average exception rates of 14%. Error reduction is a real, separate line of value: Panko's research on knowledge-work error rates found an average cell error rate of 3.9% across 14 studies, with 94% of inspected real-world spreadsheets containing errors. If your process involves manual keying between systems, a 1–5% error rate is the evidence-based assumption — and the rework it drives belongs in the baseline.

The model your CFO will actually sign

Put the four checkpoints in the open: country-correct working hours, a 1.4× loaded-cost floor, an explicit utilisation factor, and a benefits ramp. Run a conservative case near the 16% measured evidence and let the optimistic case earn its way up towards 32%. A smaller number with visible provenance beats a large one with none.

The process automation calculator builds exactly this structure — conservative, moderate and optimistic scenarios, multi-year NPV, and an assumptions audit trail showing every input and where it came from. Your first calculator is free, with unlimited re-runs, so you can stress-test the utilisation factor before the meeting rather than during it.