Research
Irish SMEs should ask what their data can support before buying AI analysis
A small, inconsistent customer history needs a different plan from a large clean dataset.
Zocri · 2 min read · Sources checked
Popularity does not establish data readiness
Ireland's CSO reports data mining as the most common AI technology among surveyed enterprises in 2025. This is an adoption observation, not evidence that every small firm's customer history can support a useful prediction. The condition of your own records should come before the sophistication of the proposed analysis.
Start with a decision you would change if the answer were reliable. A distributor might want to know which repeat orders need attention. A vague request to find insights creates room for impressive charts that do not alter any useful action or explain how a conclusion could be wrong.
Inspect the history behind the question
Check whether customer names, dates and order states are consistent. Look for duplicate customers, missing cancellations and changes in how staff recorded work. If the definition of a completed order changed, the software may interpret an administrative change as a shift in customer behavior.
Identify the amount of relevant history rather than the total number of rows. Years of unrelated records do not necessarily help answer a new product question. Keep exceptional periods and one-off promotions visible so the reviewer can consider whether they should influence the next decision.
Compare with a simple rule
Before purchasing predictive software, test a transparent baseline. For repeat orders, that might be a list of customers whose usual reorder interval has passed. Ask whether the AI result improves on that list after accounting for the effort required to review its suggestions.
Keep a portion of recent cases out of the initial setup and use them to judge the result. Review mistaken suggestions and overlooked cases. Avoid declaring success because the system explains old outcomes elegantly; useful analysis should also support decisions made before their outcomes are already known.
Choose the next investment accordingly
If the records are weak, improving intake and definitions may be the better first purchase. If a simple rule handles the job well, use it. If the AI approach adds a material improvement, document the evidence and assign responsibility for reviewing changes over time.
These are evaluation choices, not a claim that CSO recommends a particular analytical method. A modest system built around trustworthy records can be more useful than an advanced feature operating on data the business cannot explain, reconcile or maintain consistently from month to month.