How to calculate AI adoption ROI with numbers that hold up

The short answer

The point of an ROI calculation is not accuracy but transparency of assumption. Showing a conservative number together with its basis gets approved far more often than throwing out one optimistic figure.

You need only four inputs: number of people, average salary, hours per occurrence, and times per week. Add an automation-rate assumption plus setup and running costs, and you have annual savings and a break-even point.

1. Settle the hourly labour cost first

Divide annual salary by annual working hours. The standard is 2,080 (40 hours × 52 weeks). A salary of ₩50M works out to roughly ₩24,000 an hour.

A common argument starts here: shouldn’t you include employer contributions and overhead? Including them makes the number bigger and the ROI look better, but it also means a long explanation when finance asks for the basis. Calculating on salary alone and presenting it conservatively — then footnoting "roughly 1.2–1.3× with overhead included" if needed — starts fewer arguments.

2. Calculate annual hours saved

The formula is simple: hours per occurrence × times per week × 52 weeks × number of people × automation rate.

The automation rate is the most sensitive variable in the whole calculation. Do not assume 100%. There will always be a stretch where a person handles exceptions, reviews output and corrects errors. Presenting three scenarios — 50% conservative, 70% standard, 90% aggressive — heads off the "isn’t that optimistic?" question before it is asked.

3. Establish the three-year total cost

4. Calculate the break-even point

Monthly net saving is (annual saving ÷ 12) − monthly running cost. Break-even in months is setup cost ÷ monthly net saving.

Once that figure passes 24 months the case weakens, because the technology and the pricing structure are both likely to change within that window. If break-even comes out long, the better move is to narrow the scope and bring the setup cost down rather than argue the number.

Three common traps

What the documented range actually looks like

Based on Divii Consulting engagements: marketing department automation cut report-writing time by 80%, and automating meeting notes and weekly reporting cut document work by 70%. Both were document-centred repetitive work where the automation target was unambiguous.

We would not recommend dropping those figures straight into your own calculation — different work and different data conditions produce different results. But they are usable as a reference point: reductions in the 70–80% range have genuinely been achieved on document-centred repetitive work.

Frequently asked questions