You justify an AI investment the same way you justify any operational spend: benefit minus cost, over cost, with the benefit counted rather than promised. The formula is simple – what changes is where the numbers come from: hours and errors you can count today, a range instead of a point estimate, and the cost of doing nothing entered as a real line item. Below: the calculation, what to do when the benefit resists estimating, and the payback period that actually gets approved.
The formula: ROI = (annual benefit – all-in annual cost) ÷ all-in annual cost, and its companion, payback = all-in cost ÷ monthly benefit. That is the entire calculator – two divisions – and anyone selling you a more elaborate one is selling precision the inputs cannot support. The work is in the inputs. Benefit is hours or errors the project removes, priced at loaded rates. Cost is everything: subscriptions, implementation time, training hours. And the do-nothing scenario belongs in the comparison as its own line, because the status quo has a price – the same wasted hours, compounding every month while the decision waits.
Worked numbers: a sales team produces 40 quotes a month at three hours each. An AI-assisted process cuts each to one hour, removing 960 hours a year; at a $60 loaded rate that is $58,000 at full adoption, stated honestly as $35,000 to $58,000. All-in cost is $12,000 a year. Low end: ROI of roughly 190 percent, payback inside four months. High end: payback in ten weeks. Either end clears any reasonable bar – which is the point of choosing a countable problem.
Then say so with a range, because the range is the honest answer – and executives trust it more than precision. A point estimate (“this will save $47,200”) claims knowledge nobody has and hands a skeptic one number to break; a range with visible assumptions (“$35,000 to $58,000 depending on adoption, error count from the Q2 log”) shows you know exactly where your uncertainty lives. Executives run on ranges – revenue forecasts, guidance, scenarios – and a proposal shaped like their own tools reads as competence.
If even a range will not hold, two moves remain. First, count backwards instead of forwards: the cost of the current process is not a forecast, it is history – timesheets, error logs, overtime. You can always put a hard number on doing nothing, and “here is what the problem costs us; the pilot is a fifth of that” is a justification with no forecast in it. Second, if nothing about the problem can be counted at all, that is the signal to pick a different first project – justification does not get easier after the money is spent.
Under six months, payback is barely a debate – the spend feels reversible, and reversible decisions get made quickly. Six to twelve months is the normal approval zone, where the outcome rides on how defensible your inputs are. Past twelve months, expect real scrutiny and the need for a sponsor; past two years you are pitching a strategic bet, and it will be judged as one no matter what the spreadsheet says.
Payback also beats ROI percentage as your lead number for small projects. A 190 percent ROI sounds like marketing; “it pays for itself by June” sounds like operations – and it answers the question an approver is actually asking, which is when the risk comes off their books. Pair it with the do-nothing line: if the process you are fixing wastes $3,000 a month, every month of deliberation is a purchase too. Someone is already spending that money; the only question is whether anything is bought with it.
Five questions. One clear picture across all five dimensions of AI readiness. No credit card, no sales call, no fluff.
Take the QuickScan →Free · 5 questions · 3 minutes · Instant results
