AI ROI: how to calculate it (and improve it) without kidding yourself
Marketing Director
Hidden costs, real gains, the metrics to track: the complete method to measure the return on an AI project, with a simple calculation model.
"AI — what does it cost and what does it bring in?" It is the most legitimate question a leader can ask, and the worst-handled. Too many projects start on a vague promise ("save time") and end without anyone knowing whether they truly created value.
This article gives you a method to calculate an honest return on investment, avoid the classic traps, and track the right metrics over time. The goal: decide with numbers, not with enthusiasm.
The ROI of an AI project is calculated by comparing the value created (time saved, valued + additional revenue + errors avoided) to the total cost (setup + subscriptions and API + maintenance + human supervision). Basic formula: ROI = (value created − total cost) ÷ total cost. A positive ROI within the first 90 days is a good signal; maintenance and supervision are the most commonly forgotten costs.
Key takeaways
- The cost of an AI project is not just the subscription: it includes setup, API, maintenance and human supervision.
- Value is not just time saved: add additional revenue and errors avoided.
- Measure at 30, 90 and 180 days — ROI is judged over time, not in month one.
- According to MIT (2025, via Forbes), most generative-AI pilots fail: measurement is what separates a useful project from a sunk cost.
- One well-measured use case beats ten unmonitored experiments.
Step 1 — Count all the costs (including the forgotten ones)
Most ROI calculations are skewed from the start because they underestimate costs. Here are the items to include systematically:
| Cost item | Often forgotten? | Example |
|---|---|---|
| Setup | No | Scoping, configuration, integration |
| Subscriptions & API | Sometimes | Per-request cost of a language model |
| Maintenance | Often | Fixes, updates, adjustments |
| Human supervision | Almost always | Review, quality control, edge-case handling |
| Team training | Often | Upskilling, change management |
Human supervision is the most systematically overlooked item. An AI that produces a draft always implies someone to validate it. Include that time, or your ROI is an illusion.
Step 2 — Measure all the value, not just time
Time saved is the most visible gain, but rarely the most important. A company that responds to prospects twice as fast does not only save time: it wins sales. Three families of value to measure:
- Time saved, valued at the real loaded hourly cost of the people involved.
- Additional revenue: faster sales, better conversion, new billable offers.
- Errors avoided: re-entries, missed follow-ups, non-compliance — each has a cost.
As a public benchmark, Indeed reported (Forbes, May 2026) a 100% increase in developer productivity over about sixteen months and a 60% improvement in the time between idea and production. These are a large company’s figures, cited as an illustration — not a promise that transfers as-is, but an order of magnitude of what measurement reveals when a rollout is well managed.
Step 3 — Apply the formula
The formula fits on one line: ROI = (value created − total cost) ÷ total cost. A result of 1 means you doubled your investment; a result of 0.3 that you recovered 30% more than you put in. Always state your assumptions (hourly cost, volume, conversion rate) so they can be challenged.
What this changes concretely for your business
Calculating ROI forces you to choose. Rather than "doing AI" everywhere, you prioritise the use case whose value most clearly exceeds its cost. That discipline is also your best protection against gimmick projects: if a use case does not translate into numbers after 90 days, adjust it or stop it.
The metrics to track over time
| Milestone | What you measure | Decision |
|---|---|---|
| 30 days | Real team adoption, first time savings | Adjust the tool and usage |
| 90 days | Provisional ROI, effect on sales or quality | Continue, correct or stop |
| 180 days | Consolidated ROI, durable effects | Industrialise or reallocate |
Use cases by company size
- Solo: measure time saved on quote production and convert it into extra accepted assignments.
- Micro-business: track the effect of an automated follow-up on quote conversion.
- SME: compare the cost of a support agent to the cost of the hours it frees, plus the effect on satisfaction.
- Larger company: consolidate ROI by department before industrialising.
Limits and conditions for success
An ROI calculation is only as good as its assumptions. Two traps to avoid: overestimating gains (by forgetting supervision) and underestimating adoption time (a team takes weeks to absorb a new tool). Again: according to MIT researchers cited by Forbes in 2025, the vast majority of generative-AI pilots fall short of their goals. That is not an argument against AI — it is an argument for measurement and selection.
At what ROI is an AI project worthwhile?
As soon as ROI is positive, the project creates more value than it costs. In practice, aim for a clearly positive ROI at 90 days for a simple use case; heavier projects may need 6 months to materialise.
What is the most commonly forgotten cost?
Human supervision. An AI produces drafts or recommendations that must be validated. That review time is a real cost that must appear in the calculation, otherwise you greatly overestimate ROI.
How do I measure additional revenue from AI?
Isolate one variable (response time, conversion rate, number of proposals sent) and compare before/after over a representative period. Ideally, test on one segment while keeping a control group.
Do I need a specific tool to track ROI?
No. A spreadsheet is enough in most cases: cost column, value column, explicit assumptions, review at 30/90/180 days. What matters is rigour, not the tool.
An AI project is neither a bet nor an act of faith: it is an investment that can be measured. By counting all the costs, valuing all the value and tracking your metrics over time, you will know exactly where AI pays off — and where it is not worth the spend.
Want a clear ROI figure on your priority use case? During the free diagnostic, we quantify the real potential together, hidden costs included.
Get my free diagnosticRelated reading
Sources
- How Indeed Doubled Productivity With AI — Forbes — May 2026: +100% dev productivity, +60% time-to-value
- Are Small Businesses Adopting AI? — Forbes — Gene Marks, Sept. 2025 (MIT: 95% of pilots fail)
Go further with Optimal Prompt AI
1000 prompts to master AI — free
A ready-to-use library, sorted by use case. Instant access.