The real ROI of an AI project is often lower than vendor claims and higher than sceptics expect. To calculate it honestly, you need to count all the costs (not just licences), value time saved realistically, include quality and revenue effects where you can evidence them, and discount for the fact that not everyone will use AI fully. Here is a simple method and a worked example.
Step 1: Count all the costs
- Licences and tools.
- Training, including people's time in sessions.
- Setup and integration, if any.
- Governance: policy, legal and data protection time.
- Ongoing management: the owner's and champions' time.
- Learning curve: a temporary productivity dip in the early weeks.
Most ROI calculations that look too good to be true have left out the people costs.
Step 2: Value time saved realistically
Estimate hours saved per person per week on specific tasks, using before-and-after measurement where possible. Then:
- Multiply by the number of people actually using AI, not the number licensed.
- Apply a realistic hourly cost.
- Apply a conversion factor: not every saved hour becomes productive value. Be conservative.
Step 3: Add other benefits you can evidence
- Quality improvements, such as fewer errors or less rework.
- Revenue effects, such as more proposals sent or faster response winning more business.
- Avoided costs, such as reduced outsourcing or hiring you no longer need.
Only include what you can reasonably support. Unsupported benefits damage credibility.
Step 4: Calculate
ROI = (Total benefits minus total costs) divided by total costs.
Also calculate the payback period: how many months until cumulative benefits exceed cumulative costs.
A worked example (illustrative figures)
A 20-person bid team pilots AI for proposal drafting for a year.
| Item | Annual figure |
|---|---|
| Licences for 20 users | £6,000 |
| Training and learning time | £5,000 |
| Owner and champion time | £4,000 |
| Total cost | £15,000 |
| Hours saved: 15 active users x 3 hrs/week x 45 weeks | 2,025 hours |
| Valued at £35/hour | £70,875 |
| Conversion factor 50% | £35,438 |
| Net benefit | £20,438 |
| ROI | About 136% |
The figures are illustrative; use your own costs and measured savings. Notice how much the result depends on the number of active users and the conversion factor. Those are the numbers leadership will question, so base them on evidence.
Common ROI mistakes
- Counting licensed users instead of active users.
- Assuming 100% of saved time becomes value.
- Ignoring training and management time.
- Including speculative revenue benefits.
- Measuring only once, rather than tracking over time.
Making ROI more likely
The biggest levers on AI ROI are adoption and conversion: how many people use AI properly, and what happens with the time saved. Both are leadership responsibilities. For timing, see the real ROI timeline for AI investments, and for broader return questions, the ROI on an expensive AI course.
Our Leading AI Innovation & Organisational Transformation course covers measuring results and scaling up, including tracking time saved, quality and customer impact in your own 90-day plan.
Frequently asked questions
What is a good ROI for an AI project?
It depends on your organisation's hurdle rates, but focused AI projects on frequent tasks can often show strong returns when adoption is high. Be wary of claims that seem extraordinary.
How do we measure time saved accurately?
Short before-and-after time logs on specific tasks, collected from a sample of users, are usually accurate enough for decision-making.
Should ROI include employee wellbeing?
You can report it as a supporting benefit, such as less time on tedious work, but keep it separate from the financial calculation unless you can quantify it credibly.