How to Present ROI to Your CFO: What Finance Actually Checks

The headline ROI figure is the least-examined number in your business case. Finance leaders assume it is optimistic before you open your mouth — that is priced in. What they actually check is whether the machinery underneath it survives contact: where each assumption came from, which input the result is hostage to, what the floor looks like, and whether the timing of money has been handled honestly. Prepare for those four checks and the meeting stops being an interrogation.

Check one: provenance of every assumption

The fastest way to lose a finance audience is one untraceable number. It only takes one. The moment an input turns out to be "from a vendor deck" or "industry standard, I think", every other input inherits the doubt.

The trap is that widely repeated figures are often mis-quoted versions of real research. A famous example: the claim that "81% of customers prefer self-service" circulates constantly, but the underlying Harvard Business Review research says 81% of customers attempt self-service before contacting a live representative — an attempt rate, not a success or preference rate. Gartner's consumer research found only 14% of issues are fully resolved in self-service. A CFO who has seen the first number waved around will respect you for volunteering the second.

The standard to hold yourself to: every input is either your own operational data (say which system it came from and when), or a published source you can link, quoted with its caveats. If you can't trace it, cut it or replace it with a conservative placeholder labelled as such.

Check two: sensitivity — which input owns the result

Any experienced CFO reflexively asks some version of: "if this assumption is 20% off, what happens to the number?" A model is really only as strong as its most influential input, and you should know which one that is before the meeting does.

This is what a tornado chart is for: flex each input across a plausible range, hold the others constant, and rank inputs by how far each one swings the result. The output is usually humbling — two or three inputs dominate, and the rest are noise. In a support-cost model, for example, the result typically lives or dies on the deflection-rate assumption, because the unit economics are so asymmetric: Gartner put live channels at an average $8.01 per contact against roughly $0.10 for self-service, and ContactBabel puts the average US inbound call at $7.20. When one input has an 80× cost gap running through it, small errors in that input dwarf large errors everywhere else.

Presenting the sensitivity analysis yourself changes the dynamic entirely. Instead of finance hunting for the weak point, you are showing them you already found it — and that your evidence is strongest exactly where the model is most exposed.

Check three: the downside scenario is read first

Finance reads scenarios bottom-up. The optimistic case is decoration; the conservative case is the decision. So the conservative case must be genuinely conservative — anchored to measured evidence, not just your moderate case minus 10%.

The published record gives you honest floors. For automation, Deloitte's UK study measured a 16% average cost reduction among implementers — a citable conservative anchor, against the 32% that Deloitte's 2022 survey reports from the minority who actually measured. For self-service, Gartner's 14% full-resolution figure is the realism check against every optimistic deflection claim. A conservative case built on numbers like these does something subtle: it makes approval a decision about the floor. If the project clears the hurdle in the downside case, everything above it is upside, and the argument is over.

Check four: NPV over simple ROI, and honest unit costs

Simple ROI ("312% return") ignores when money moves — and timing is most of what a CFO thinks about. Costs land in months one to six; benefits ramp over years. Net present value discounts each year's cash flow back to today at the organisation's discount rate, so a pound of benefit in year three is correctly worth less than a pound of cost in month two. Presenting multi-year NPV, alongside IRR and payback, signals you are speaking finance's native language rather than marketing's.

The same honesty applies to unit costs inside the model. Price labour at fully-loaded cost — BLS data shows US private-sector benefits add roughly 43% on top of wages, so salary × 1.4 is a citable floor — and use hours actually worked, not contracted: OECD figures for 2025 put the US at 1,800 hours a year and the UK at 1,533, a long way from the 2,080 in most spreadsheets. Small choices like these are exactly what a finance reviewer spot-checks, and getting them right on the first pass buys credibility for everything else.

The audit trail changes the meeting

Put all four together and the artefact you bring is less a pitch than a set of accounts: every assumption listed with its source and caveats, a sensitivity ranking of the inputs, a defensible downside case, and discounted multi-year cash flows. The meeting changes shape. Instead of defending a headline number, you are walking through a ledger, and the conversation moves to the only question that matters — do we believe the two or three assumptions this decision actually turns on?

This is the structure LeadersToolset produces by default: conservative, moderate and optimistic scenarios, a receipt-style assumptions audit trail, sensitivity analysis, and multi-year NPV and IRR — with every benchmark's methodology public on the methodology page. You can see the full output format in the sample report, or build your own from the calculator library — your first calculator is free, with unlimited re-runs, so the version you take into the meeting is the one you have already stress-tested.