Services
Explore
Free AI assessmentAnswer 15 questions about one process and see where AI could help.
Start an assessmentIs AI worth the investment for a business your size? How to work it out
A method for estimating the time an AI tool could recover on a task, the costs to set against it, and when the numbers suggest keeping your current process.
The short answer
AI pays back when it improves a task by more than it costs
The return on AI for a small business usually comes from time recovered on a specific task. Measure it as net hours: the time the task takes now, minus the time it takes with AI, including checking, corrections and maintenance. Then compare the value of those hours with what the tool costs to set up, run and support.
AI does not guarantee a return, any more than a new hire or a new product line does. Whether a particular use is worth it depends on how often the task happens, how much time it takes now and how much checking the AI output needs.
A common reason for a poor return is starting with a tool and looking for uses, with nothing measured before or after. A clearer test starts with one task, records how long it takes now, and compares that with a trial on real work.
Where value comes from
Four sources of return from AI
Usually the easiest return to measure. For example, a task that takes 2 to 3 hours a day adds up to several hundred hours a year. If AI shortens it, the difference is time the team can use elsewhere, after allowing for checking.
Manual data entry, document processing and repetitive checks all produce some errors. In structured work, an AI step can apply the same rules to each item and flag exceptions, which can reduce avoidable mistakes. Measure the error rate before and after to know.
AI can produce first drafts, summaries and analyses quickly. The person responsible still decides what the work says and checks it, so measure the time including review.
Time recovered can let the same team take on more clients or more output. How the business uses that time is its own decision, and it only becomes a cash saving if costs fall or the time goes into paid work.
Worked examples
Illustrative examples of net time recovered
| Task | Time now | Time with AI, including checking | Net time recovered |
|---|---|---|---|
| Example: drafting client documents | 4 hours a week | 1 hour a week reviewing and correcting drafts | 3 hours a week, or 138 hours over 46 working weeks |
| Example: answering routine customer enquiries | 10 hours a week | 4 hours a week checking drafted replies, plus 1 hour a week updating approved answers | 5 hours a week |
| Example: summarising meeting notes | 45 minutes per meeting, 4 meetings a week (3 hours) | 15 minutes per meeting checking the summary (1 hour) | 2 hours a week |
| Example: data entry from documents | 6 hours a week | 2 hours a week checking extracted data, plus 30 minutes fixing failed extractions | 3.5 hours a week |
These illustrative examples use assumed times. None of them is a measured client result. Net time recovered is the time now minus the time with AI, including checking, corrections and maintenance. Time recovered is capacity, which is different from cash saved. Your own figures depend on task volume, the tool and how much checking the output needs.
The cost side
What to include on the cost side
Costs fall into several groups: subscriptions and licences, usage charges, setup work such as scoping, building and connecting systems, training, and ongoing support and maintenance. Check current prices with each provider, and note whether they include VAT.
Custom tools need looking after once they are live: prompts and reference material need updating, connected systems change and someone has to handle problems. Include that time and cost in the estimate. If the net hours recovered do not cover the costs, keep the current process or try a simpler approach.
Common questions
Questions about AI ROI
How do I calculate the ROI of AI for my business?
Pick one task your team repeats and record how long it takes each week. Trial the AI approach on real work and record the time it takes, including checking, corrections and upkeep. The difference is your net time recovered. For example, 3 net hours a week at an assumed internal cost of £25 an hour is £75 of staff time a week, or £3,450 over 46 working weeks. Compare that with the subscription, usage, setup and support costs for the same period.
What's a realistic payback period?
It depends on the task, the volume and the costs. An off-the-shelf tool on a frequent task can pay back faster than a custom build, which has higher setup costs. Work it out from your own trial: divide the setup cost by the net weekly value once running costs are deducted.
What if the AI doesn't perform as expected?
Agree before you start what result would make the tool worth keeping. If the trial falls short, look at whether the task, the tool or the instructions were the problem. Changing approach or stopping are both reasonable outcomes.
Are there hidden costs I should plan for?
Yes. Licences, usage charges, integration work, training time and ongoing maintenance all cost money or time. Include them in the business case from the start. We set out the costs we know of before any work begins.
Does the ROI hold up as AI tools get more expensive?
Prices and plans change. Use current prices in your calculation, review the numbers at each renewal and keep the option to switch tools or stop if the return no longer covers the cost.
Talk to us
Want help applying this to your business?
Bring the task that takes your team the most time. We reply within one working day.