How to Estimate AI Project ROI Before You Start

How to estimate AI project ROI before you start comes down to four numbers you can pull together in an afternoon: what the task costs today in staff time, how much time AI would realistically save, what building it will cost, and what running it will cost every month. Divide the one-time build cost by the net monthly savings and you get a break-even timeline in months, not a guess. If that timeline runs longer than the project will realistically stay useful, don't build it yet. This is a pre-launch estimate, built on honest numbers, not a forecast dressed up as fact.

Why estimate AI project ROI before you build anything

Most write-ups about AI and return on investment happen backward: a team builds something, ships it, then tries to prove the spend was worth it. That's measurement, and it matters, but it's a different exercise from deciding whether to spend the money at all. If you already have a live project, how to measure AI ROI for a small business after launch is the right read. This is the step before that: pricing the return before anyone writes a prompt or signs a contract, the same discipline behind most sound decisions about AI for small business spending generally, whether the project costs $500 or $50,000.

There's a real gap between AI adoption and AI payoff worth building into your estimate as a dose of skepticism. In McKinsey's 2025 State of AI survey, 88 percent of organizations reported using AI in at least one business function, but only about 6 percent qualified as "AI high performers" attributing 5 percent or more of EBIT to AI, and that edge came largely from redesigning the workflow before deploying the tool, not bolting AI onto an unchanged process (McKinsey, The State of AI). An estimate done before you start forces that question early, while it's still cheap to answer.

The four inputs every AI project estimate needs

Every credible ROI estimate for an AI project, done before a line of code exists, rests on the same four inputs. Get sloppy on any one and the break-even number that comes out is fiction.

1. What the task costs today

Start with the fully loaded cost of the task as it's done now: minutes per instance, times how often it happens per month, times the loaded hourly cost of the person doing it. Loaded means salary plus payroll tax and benefits, not the number on a pay stub. A 15-minute task happening 200 times a month, done by someone at a $35 loaded hourly rate, costs roughly $1,750 a month today. That's your ceiling; AI can't save more than the task currently costs.

2. How much time AI will realistically save

This is where estimates go wrong, almost always toward optimism. Few tasks get fully automated; most keep a human reviewing exceptions before output ships, so a realistic estimate assumes AI cuts the time per instance rather than erasing it, and discounts the first month or two while staff learn the tool. Survey data supports building in that discount: in Thryv's 2026 AI and Small Business Adoption survey, 92 percent of small-business AI users said the technology saves them time, and 79 percent expected to get back between 11 and 60 hours a month, a wide but bounded range (Thryv, 2026 AI and Small Business Adoption Survey).

3. What building it will cost

This is the one-time number: developer or agency hours, any setup or integration fee, time spent preparing the data the AI needs, and testing before rollout. This is also the number that anchors how to write an AI project proposal that a partner or manager can actually approve, since a proposal without a build-cost line is just an idea.

4. What running it will cost every month

Recurring cost is the line item most estimates skip, and it erodes ROI every month the project runs: API usage that scales with volume, hosting, and staff time spent monitoring output. Check these assumptions against a broader view of how much a small business should spend on AI tools for comparable projects, rather than pulling a number from a vendor's pricing page.

The break-even formula

Once you have all four inputs, the math is one line:

Break-even (months) = One-time build cost ÷ (Monthly value of time saved − Monthly run cost)

Treat the result as a threshold, not a pass or fail switch. As a rough guide for small-business budgets:

• Under 6 months: an easy yes if the estimate is reasonably conservative.

• 6 to 12 months: worth building only if the task is stable and won't change shape before you recoup the cost.

• Over 12 months: narrow the scope, cut the build cost, or shelve the project until one of those changes.

Worked example: a document-review assistant for a 12-person firm

A 12-person accounting firm reviews client-submitted documents before intake, currently done manually by an office coordinator. They're weighing an AI tool that pre-reads documents, flags missing fields, and drafts a summary for the coordinator to check rather than write from scratch. Here's the estimate they built before approving the budget.

At roughly five months, this clears the "easy yes" threshold, assuming review time genuinely drops to 8 minutes and volume holds steady. Cut volume in half, to 80 documents a month, and the same build cost takes over ten months to recoup, a different decision even though nothing about the AI changed. The estimate, not the tool, is what tells you that.

Three places these estimates go wrong

• Assuming full automation. If a human still checks the output, that review time belongs in the "after" number, not zero.

• Ignoring the ramp-up month. The first four to six weeks after launch usually run slower than the steady state while staff adjust, so estimate that period lower than the rest.

• Forgetting run cost scales with success. A tool that works well gets used more, and API costs that looked trivial in a pilot can climb once the whole team adopts it, so re-check the run-cost line at expected full-volume usage.

Set a go/no-go threshold before you start

Pick your break-even ceiling before running the numbers, not after, so the estimate can't be reverse-engineered to justify a decision already made. A common approach ties the ceiling to cash runway: a nine-month break-even is a harder sell with six months of runway than with a year of it. Write the four inputs, the formula, and the threshold into whatever document covers how to write an AI project proposal, so whoever approves the budget sees the assumptions, not just the conclusion.

If the math doesn't clear your bar, that's useful information too. It usually means the task isn't a good early candidate, worth revisiting against which tasks to automate with AI first, or the business isn't quite positioned to capture the savings yet, worth checking against the signs your small business is ready for AI automation before trying again with a smaller scope.

Frequently asked questions

What is a good ROI for an AI project?

For a small business, a break-even timeline under six months on the formula above is a comfortable yes, six to twelve months works for stable, recurring tasks, and anything past twelve months needs a narrower scope or lower build cost first. There's no universal percentage return that applies across projects, since the right threshold depends on the business's cash position, not a fixed benchmark.

How do you calculate AI ROI in advance?

Multiply time saved per task by how often the task happens per month and by the loaded hourly cost of the person doing it, to get the monthly value of time saved. Subtract the estimated monthly run cost for net monthly benefit. Divide the one-time build cost by that net monthly benefit for a break-even timeline in months. Every input should stay conservative, especially the time-savings assumption.

Is an AI project worth it if the task is small?

Usually not on its own. A task costing $200 a month to do manually rarely justifies a $4,000 build, even with generous time savings, since the break-even stretches past a year. Small tasks tend to pencil out only when bundled into a tool handling several related steps, or when build cost itself is low, such as configuring an existing platform rather than commissioning custom development.

What's the difference between estimating and measuring AI ROI?

Estimating happens before you spend anything and relies on assumptions about time savings and cost that haven't been tested. Measuring happens after launch and uses logged data: actual time per task, actual API bills, actual adoption rates. The estimate is a go/no-go tool; the measurement is an accountability tool that shows whether the estimate held up and what to fix if it didn't.