Ireland's AI Adoption Gap Is a Size Problem, Not a Technology Problem

The Central Statistics Office reported that just over 20% of enterprises in Ireland used artificial intelligence in 2025, up from over 15% in 2024. That headline hides the number that actually matters for anyone running a mid-sized business here: 57.7% of large enterprises (250+ employees) reported AI use, against 17.2% of small enterprises (10–49 employees).

That is a three-fold gap in a market where every company has identical access to the same models. Nobody in Ireland is locked out of frontier AI by price or availability — Stanford's AI Index 2025 recorded inference costs for GPT-3.5-equivalent performance falling from around $20.00 per million tokens to roughly $0.07 in about eighteen months. Capability is not scarce. The scarce resource is the operating layer: clean data, connected systems, and a process worth automating.

Look at what Irish enterprises actually use AI for and the pattern confirms it. The most common applications reported were data mining (10.8%) and automating administrative processes (7.8%). Both depend entirely on having structured data in one place. A large enterprise already has an ERP, a CRM and a warehouse feeding them. A fifty-person firm running on spreadsheets, an inbox and a legacy quoting tool has nothing for a model to sit on.

The same CSO release shows why: 34.7% of enterprises used ERP software and 28.4% used CRM software, but among large enterprises those figures were 78.7% and 66.0%. Among small enterprises, 29.2% and 23.3%. The AI adoption gap is downstream of a systems gap that opened a decade earlier.

This reframes the investment decision. If you are a fifty-person business asking whether to buy an AI tool, the honest answer is usually that the tool is not the constraint. Connecting your quoting, delivery and finance data so that one record describes one customer is the constraint. Do that and AI becomes a cheap layer on top. Skip it and you buy a pilot that never leaves the pilot stage.

We would rather clients spend the first quarter on data plumbing than on model selection. It is less exciting, it does not demo well, and it is the only sequence that has produced measurable returns in the work we have done. The 17.2% figure will rise. It will rise fastest for the firms who fixed the foundation first.