
AI ROI: measuring whether AI is worth the money
In short: AI ROI compares the measurable value of an AI system, such as time saved or revenue gained, with its full cost, including review, rework and maintenance.
AI ROI (return on investment) is the measurable value an AI system produces, in money saved, revenue gained, risk reduced or capacity freed, compared with its full cost: models, infrastructure, integration, change management, review time and maintenance. It is hard to measure honestly. In a randomised trial published in July 2025, 16 developers estimated AI had cut their task time by 20% when it had increased it by 19%.
AI ROI in plain words
Was the AI worth the money? Did it pay for itself, counting everything it took to make it work, and can you show that with numbers someone else would accept?
How it works
- Pick a specific process and baseline it before AI: volume, time per unit, error rate, cost.
- Deploy to a defined group, ideally with a comparison group.
- Measure the same metrics after, including new costs such as review and rework.
- Convert time saved into value only if it is redeployed to something measurable.
- Repeat over months, because early gains and costs both shift.
A measured result looks like the Generative AI at Work study, first submitted in April 2023. Its abstract reports data from 5,172 customer support agents and a 15% average rise in issues resolved per hour, an output the business already counted.
Example
As an illustration, an accounts-payable team processes 10,000 invoices a month. AI extraction cuts handling time per invoice, but adds review of low-confidence cases and a platform fee. ROI is positive only after counting the reviewer hours honestly and confirming the freed time was used to clear a backlog that previously caused late-payment penalties.
Where you'll meet it
- Board and budget discussions about AI programmes
- Vendor business cases and analyst surveys
- Debates about which pilots deserve a production budget
- Pricing of AI-powered services
What it changes for you
For leaders, ROI needs baselines before rollout, not surveys after. For engineers, cost per task and quality metrics feed the business case directly. For services firms, outcome-based pricing makes ROI a contract term, not a slide.
AI ROI vs productivity claims
Productivity claims often report self-estimated time saved. ROI requires that saved time to turn into measurable value or avoided cost.
Common mistake
My view: counting "hours saved" from user surveys as money. Unless the hours are redeployed to measurable work, they are a feeling, not a return. Both sides of the AI ROI debate often run on these soft numbers.
Related terms
Cost per task · Build vs buy (AI) · Demoability · AI governance
From the archive: Both sides of the AI ROI debate run on soft numbers · AI ROI needs an acceptance-to-EBIT bridge
Hours saved become a return only when they are redeployed to work you can measure.


