Glossary
What is ROI? Return on investment of an AI project
In short
ROI (return on investment) is a measure that compares the benefit of an investment with its cost. It shows how much money came back for each unit of money invested.
Business and adoption · return on investment
MMOOZZGG.com editorsChecked: 1 min read
For a child
You buy lemons and sugar, make lemonade and sell it to your neighbors. In the evening you count the coins. If you took in more than you spent on lemons, the idea paid off. If less, think about what to change: the price, the recipe or the place. Adults do the same math when they spend money on new software.
In plain words
The measure answers a simple question: was the money worth spending? You take what the investment brought, subtract its cost and divide by that cost. The result is a share or a percentage, so different investments are easy to compare. For example, an office bought a coffee machine, and staff stopped going out to a coffee shop. In a year the saved time was worth twice the price of the machine. So the ROI is 100%: every unit of money spent came back and brought one more.
For business
In AI projects, the benefit comes from three places: saved staff time, extra revenue from faster answers to customers, and fewer errors and fines. Costs include not only development but also model subscriptions, integrations, support and staff training. A common mistake is to invest where the effect is visible, not where it is larger. The authors of the 2025 MIT NANDA report write that budgets go to sales and marketing, while high returns are often in internal processes. In 2024, BCG noted that 74% of companies had not yet seen real value from AI. So measure the baseline before the launch and check the effect in money during a pilot.
For a specialist
ROI (return on investment) is calculated as benefit minus cost, divided by cost. For projects with a long horizon, add measures that account for time: payback period, net present value (NPV) and internal rate of return (IRR). In AI projects the main difficulty is attribution: the effect is mixed with seasonality and other changes, so use a control group or an A/B test. Count costs as the total cost of ownership (TCO): inference paid by tokens or run on your own GPUs, integrations, quality monitoring, retraining and the time employees spend checking answers. Count intangible effects, such as the speed of decisions, separately, and add them only after you turn them into money.
Example
A company calculated that an AI agent that handles requests costs less per year than the pay of one manager, yet takes over most of that person’s routine work. A pilot confirmed the payback: the company compared the number of requests processed before and after the launch.
Related terms
Questions and answers
Sources
Where companies use it
Not sure whether your company needs this?
AI solutions map