The short answer
To measure AI ROI, pick one business number the AI should move, record a baseline before launch, count the full cost (build, licenses or usage, integration, training and upkeep), then compare the change against that cost over 12 months. Count saved time as value only when the hours go into work that earns or saves money.
To measure AI ROI, you need three things before launch: one business number the AI should move, a baseline for that number, and the full cost of the project. Most companies skip at least one. Only 13% of chief information officers (CIOs) say AI has delivered significant value so far, according to Gartner's 2027 CIO survey, and only about 29% of executives in an IBM discussion said they can measure AI ROI with confidence.
What is the ROI of AI?
AI ROI is the value the system adds, minus what it costs, divided by what it costs. The formula is ordinary. The hard part is being honest about both halves.
IBM splits the value into hard and soft returns. Hard returns show up in the accounts: lower labor or operating costs, more leads, more revenue. Soft returns are real but slower, such as better morale or customer experience. For a budget decision, count hard returns and treat soft ones as a bonus you note but do not bank.
IBM also reports that only about 25% of AI initiatives deliver the ROI that was expected. That gap usually comes from the measurement plan, or the lack of one, more than from the model.
How do you measure AI ROI, step by step?
Decide what success looks like in money before anyone writes code, then measure the same thing the same way after launch. This is the method we recommend:
- 01Pick one business number the AI should move: cost per invoice, tickets closed per agent, days to close the books, qualified leads per month.
- 02Record a baseline for 4 to 8 weeks before launch, so normal ups and downs are visible.
- 03Add up the full cost: build and integration, licenses or usage fees, staff time for training and change, and yearly upkeep.
- 04Run the AI with one team first and keep a similar team working the old way, so you can see what changed because of the AI.
- 05Turn time saved into dollars only when the hours go into work that earns or saves money.
- 06Review at 3, 6 and 12 months against kill criteria you wrote down before launch.
The comparison team matters more than it sounds. People are poor judges of their own speed. In a 2025 trial by the research group METR, experienced developers took 19% longer on tasks with AI while believing it had made them faster. A side by side comparison stops that kind of error from reaching your business case.
If you are still pricing the build itself, our guide to what AI integration costs gives published ranges by company size.
A worked example: AI for a 10 person support team
Here is the calculation with illustrative numbers. Plug in your own. Say a support team of 10 costs $40 an hour each, fully loaded, and a pilot shows an AI assistant saves each person 3 hours a week.
That is 30 hours a week, or about 1,500 hours over 50 working weeks, worth $60,000. The first year costs $30,000 to build and integrate, $12,000 in licenses and usage, $8,000 of staff time for training and process changes, and $6,000 of upkeep: $56,000 in total.
| Scenario | Value in the year | Cost in the year | ROI |
|---|---|---|---|
| Year 1, all saved hours go into useful work | $60,000 | $56,000 | 7% |
| Year 1, half the saved hours drift away | $30,000 | $56,000 | -46% |
| Year 2, all saved hours used, build already paid | $60,000 | $18,000 | 233% |
Two lessons come out of this. Most AI projects look weak in year one, because the build cost lands up front, so judge them over a sensible period. And the same tool swings from a loss to a gain depending on what happens to the time it frees up.
Why is time saved not the same as money saved?
Saved hours only become money when someone decides what those hours are for. If nobody does, the time fills up with other tasks and the saving vanishes from the accounts.
BCG has seen this directly. In its work with clients, efficiency gains of up to 50% tended to fade within a few months, as freed time was absorbed by backlog and new priorities. Its 2026 survey of nearly 12,000 workers found that 42% of regular AI users save about eight hours a week, yet 66% get little or no guidance on what to do with that time.
So write the redeployment plan into the business case. "We will handle 20% more tickets without hiring" is a plan. "Staff will be more productive" is a hope, and it will not show up in your results.
What is the 10/20/70 rule for AI?
BCG's 10/20/70 rule says 10% of the effort in an AI program should go on algorithms, 20% on technology and data, and 70% on people, organization and processes. It matters for ROI because it tells you where the real cost sits.
Most business cases price the model and the integration carefully, then treat training and process changes as free. By BCG's rule, those are the largest part of the work. BCG also estimates that companies investing in training and change management reach about 60% adoption, against roughly 30% for those that do not. Adoption is what turns a working tool into a return, so that budget line pays for itself.
What counts as a good AI ROI?
There is no universal benchmark, so be wary of anyone who quotes one. The useful comparison is against your own baseline and your next best use of the same money.
Published surveys show how wide the spread is. Among technology companies, EY found that about 16% report zero ROI from generative AI copilot projects, while 43% see returns above 50%. In BCG's 2025 study of more than 1,250 firms, 60% reported no material value at all.
Our rule of thumb: set a payback period your finance team would accept for any other investment, often 12 to 24 months, and a kill threshold before launch. If the pilot has not moved the business number by the first review, stop or change course. Gartner found that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025. Dropping a weak pilot early is cheaper than funding it for another year.
Before scaling a pilot, our AI production readiness checklist covers the operational questions that sink returns later. If you want a second pair of eyes on the numbers, our AI and automation team can review a business case before you commit. The next step costs nothing: pick the one number your AI project should move, and start recording its baseline this week.
Thinking about this for your own team?
Tell us what you are trying to do. We reply within one business day with how we would approach it, including when the answer is not to build anything.
Questions people ask
What is the ROI of AI?
It is the value an AI system adds minus what it costs, divided by what it costs. Value can be hard, such as lower costs or more revenue, or soft, such as staff morale. IBM reports that only about 25% of AI initiatives deliver the ROI that was expected.
What is the 10/20/70 rule for AI?
It is BCG's rule of thumb for where AI effort should go: 10% on algorithms, 20% on technology and data, and 70% on people, organization and processes. For ROI it means the change work is the largest cost, so leave it out and your numbers will look better than they are.
How long does it take to see ROI from AI?
Hours saved show up as soon as people use the tool, but AI that supports decisions can take years to show up in results, as IBM notes. Plan to judge most projects over 12 months, and set the review dates before launch.
Why do so many AI projects fail to show ROI?
Mostly because the value was never defined. Gartner found that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025, citing poor data, weak risk controls, rising costs or unclear value. Picking the business number first fixes the last of those.
Sources
- 1.IBM: How to maximize AI ROI in 2026
- 2.BCG: The Widening AI Value Gap (September 2025)
- 3.BCG: The Stairway to GenAI Impact
- 4.BCG: AI at Work, Why Strategy Matters More Than Tools (June 2026)
- 5.TechTarget: Chasing AI ROI? Here's why it keeps slipping away
- 6.EY: How tech companies can break out of the AI ROI trap