Is AI worth the investment? A practical guide for Irish SMEs.

Start with an expensive piece of work

Choose a task you can describe in an ordinary sentence. Perhaps a team member spends Friday gathering figures for a report, sales staff search several systems before every call, or enquiries wait while someone finds the right information. Write down who does it, how often and what a useful improvement would look like.

The CSO reported that 20.2% of enterprises in its 2025 survey used AI. That survey covers businesses with ten or more people in manufacturing, construction and selected service sectors. It is useful context, but it does not tell you whether a particular project will pay for itself in your business. CSO, CSO methodology.

Find the real starting point

Observe a normal week, including the awkward cases. Count the time spent finding information, entering it, checking it and correcting mistakes. Ask the people doing the work to help: the process described in a meeting may differ from the one they actually follow.

For the test, agree a small set of measures. Time per completed job, the proportion that needs correction and the time customers wait are often more useful than the number of answers an AI tool produces. Keep the quality standard the same before and after.

• How many times does the task happen?
• How long does it take, including checks and rework?
• What would the team do with the time recovered?
• What happens if a result is wrong?

Put numbers on the value, carefully

Here is an illustrative calculation, not a client result or a forecast. Suppose a task takes ten staff hours a week. A trial reduces that to four hours, including human review. At an assumed staff cost of €30 an hour and 46 working weeks, six hours recovered each week represents €8,280 of annual capacity.

If the assumed extra software and support cost is €2,400 a year, the net annual capacity value is €5,880. Against an illustrative €15,000 setup cost, simple payback is about 31 months. Over three years, the value after those costs is €2,640. This assumes the improvement lasts and the team uses the capacity productively.

Include more than the price of the tool

Allow for connecting systems, preparing information, training staff and keeping the solution working. Include the time someone will spend checking results and handling exceptions. Ask which costs are one-off, which recur and which could increase with use.

Run a cautious version of the calculation too. What if the time saving is smaller, fewer staff use the tool, or running costs rise? If the case only works when every assumption is optimistic, make the test smaller before making the commitment larger.

Test the uncertain part before committing

A demonstration shows what can happen. A useful trial checks what happens with representative work from your business, including incomplete information and unusual requests. Agree what would justify proceeding, changing the approach or stopping.

NIST’s voluntary AI Risk Management Framework treats evaluation and risk management as part of designing and using AI. Our practical application is to assess quality, consequences and value together, rather than treating a fast answer as a successful result. NIST.

Keep the decision commercial

The outcome might be an AI assistant. It might be a straightforward connection between two systems, a better process or a decision to leave things alone. The useful question is which option improves the work enough to justify its cost and disruption.

Before agreeing a project, ask for a clear scope, a realistic measure of success, the total expected cost and a named person who will own the result. That is a stronger basis for investment than buying because everyone else appears to be doing it.