Position checked 4 November 2026

If someone spends Friday afternoon retyping orders, we’d start there. Write down how long it takes, who checks it and what usually goes wrong. That gives you something to compare with after a trial.

Pick one job with a clear owner and a result you can measure. Drafting a service report, sorting incoming enquiries or finding an answer in approved documents are easier places to start than automating a decision about a person.

The CSO found AI use in 20.2% of surveyed Irish enterprises in 2025, up from 15.2% in 2024. That survey excludes firms with fewer than ten people and covers specified sectors. It is a benchmark for uptake, not evidence that an AI project will save your business a particular amount. CSO AI results · CSO methods

Run a small pilot you can check

Measure the current task over a normal period: volume, staff time, rework and delay. Write down what the proposed tool may receive and what it must never receive. Give a named person responsibility for checking the output.

Test ordinary cases and awkward ones, including missing or misleading information. Compare the new process with the baseline using the same scope and period. Include the time people still spend checking results, dealing with exceptions and keeping the tool running.

For example, saving 20 staff hours a week at an assumed €28 per hour would represent €29,120 in illustrative annual value of staff time released over 52 weeks. Cash savings require a reduction in spending. Improvements in service or quality are separate benefits you can measure.

If personal data is involved, the Irish Data Protection Commission advises organisations to understand what data an AI product uses, where it goes, whether the supplier retains or reuses it and how GDPR duties will be met. A data protection impact assessment is mandatory where processing is likely to pose a high risk to people’s rights and freedoms. DPC AI and LLM guidance · DPC DPIA guidance

At the end of the pilot, decide explicitly: scale, revise or stop. Keep the result, assumptions and owner with the decision. If funding is relevant, check your agency route first; the Grow Digital Voucher requires a prior Digital for Business project and funds only eligible costs.

Choose a first use with a clear review point

  • Back office: Extract fields from incoming orders into a draft record. Flag missing product codes for staff and compare the draft with the original order before it enters the system.

  • Customer service: Draft answers using approved policies. Give staff an escalation route for refunds, complaints and anything the policy does not cover.

  • Reporting: Summarise figures from a known dataset. Check every figure against the source and measure whether the review takes less time than writing the report manually.

A 90-day pilot you can stop

  1. First month: Ask staff which tools they already use, including accounts opened outside the approved toolset. Agree the permitted data, named reviewer, baseline and stop criteria before testing.

  2. Second month: Run a small trial alongside the existing process. Keep source records and log errors, rejected outputs and exceptions. Prevent the tool from sending a message or updating a system before a person checks it.

  3. Third month: Compare turnaround, rework, error rate and total cost with the baseline. Include checking, licences and maintenance. Give the workflow owner the evidence to decide whether to scale, change the design or stop.

For example, a draft-order tool that saves four hours of rekeying but adds three hours of review frees one hour before ongoing costs. If it also makes mistakes that could release the wrong order, fix or stop the pilot before expanding it. Data Protection Commission guidance for AI users.

Could AI save time on this job?

Use the scorecard to compare the time you spend today with what you’d spend using AI. Include time to check the results. We can help you decide what to test first.