24 August 2026 · 6 min read
TL;DR
One live AI workflow for an Irish SME is a fixed-price build from €8,000, or day work at €600–€1,200, plus a monthly running cost to budget at €50–€300 and half an hour a week of somebody's review time. LEO and Enterprise Ireland supports typically offset 40–80% of the build. Whether it pays back is arithmetic you can do yourself in two minutes, and it turns entirely on how often the task recurs.
The number nobody publishes
There is no shortage of research on AI adoption. There is almost none on price. The reports count who is using it and how much time they say they save, then go quiet exactly where an owner's question starts: what will this cost me, and when does it pay for itself.
I'd rather publish the cost side than let it be guessed at, so below are the figures that are genuinely public, the figures that are mine, and a clear line between them. Where a number is self-reported, I say so.
Where Irish businesses actually are
The CSO puts AI use by Irish enterprises at more than 20% in 2025, up from over 15% in 2024 — measured, official, and a lot lower than the impression you'd get from LinkedIn. Trinity's AI Economy in Ireland 2026 report with Microsoft, based on a survey of 250 organisations here, finds Irish adoption running above the EU average.
Against that, research from OpenAI with Dogpatch Labs and Skillnet reports close to nine in ten Irish SME leaders using AI and gaining around five hours a week. That gap is not a contradiction. The CSO is counting organisations that have put AI into the business; the SME research is asking leaders whether they personally use it, and the five hours is self-reported. A founder using ChatGPT to write emails is real value and it is not a workflow.
The QuickBooks 2026 AI Impact Report, across more than 34,000 responses, names the same split from the other side: usage is broad, paid commitment is much narrower. Plenty of businesses are using AI. Far fewer have paid to have something built that runs whether or not the owner remembers to open a chat window.
The gap between using AI and having a workflow is the gap between a good habit and an asset.
The cost side, in euro
These are my published rates, not a market average. Anyone quoting you a market average for AI work in Ireland is making it up.
| What | Figure |
|---|---|
Discovery and costing Pick the workflow, count the hours it eats, decide what a good result looks like. Ends in a ranked shortlist and a price, including an honest don't-build where the numbers don't work. | 1 to 3 days |
Day rate Depends on the work: workflow design and enablement at the lower end, build and integration at the upper. | €600 – €1,200 |
Fixed-price build One workflow, live, with a named human approving the output and a record of every run. Fixed price so the risk of the estimate sits with me, not you. | from €8,000 |
Running it, monthly Model usage plus whatever hosting you already pay for. A text-drafting workflow at SME volumes sits at the bottom of that; voice, documents at volume or a hosted vector store push it up. Check it against your own build rather than treating it as a quote. | budget €50 – €300 |
Review time, weekly The part nobody budgets. Someone reads and approves the output. If nobody has that time, the workflow does not pay back — no matter what it cost to build. | 15 – 45 minutes |
What you probably don't pay
The headline price is rarely the cash figure. Local Enterprise Office and Enterprise Ireland supports commonly offset 40–80% of a project like this, depending on the scheme, your size and where you are. Eligibility and rates change, so confirm the current terms with your LEO before counting on a number — but do check, because a €8,000 build that lands at €2,400 net is a different conversation entirely.
Now work out your own number
You don't need my figures for this part. You need yours. Take the one task you'd point at first — the invoice run, the quote follow-ups, the enquiries someone re-types into a spreadsheet:
- Hours a week the task takes, honestly, across everyone who touches it.
- × your loaded hourly cost for the person doing it — salary, employer PRSI and overhead, not the gross wage.
- × 46 working weeks = what the task costs you a year.
- − the review time it keeps (say half an hour a week, costed the same way) and the annual running cost.
- Compare what's left against the build price, net of any grant.
If the annual cost of the task is smaller than the build, stop. That's a real answer and it saves you money today. If it's several times the build, you have a business case you can take to a board or a bank without a vendor's help.
What the figures won't tell you
Price is the easy part. What actually decides whether a workflow pays back is whether the work genuinely recurs, whether it produces a record you can audit, and whether a named person has the time and authority to approve the output. Those three conditions are the difference between a workflow and an expensive demonstration — and no adoption statistic will tell you which one you're about to buy.
Two of the numbers above will move under you. Model pricing keeps falling, which pushes running costs down. Review time does not fall, because it is the point. Any business case that assumes the human gate goes away is the one to be suspicious of.
Sources
- CSO, Information Society Statistics – Enterprises 2025 (released 6 February 2026)
- CSO, Business in Ireland 2025 – Sustainability Through Innovation and Technology (24 March 2026)
- Trinity Centre for Digital Business and Analytics with Microsoft Ireland, AI Economy in Ireland 2026 (n=250 organisations)
- OpenAI with Dogpatch Labs and Skillnet Innovation Exchange, research on Irish SME leaders (self-reported)
- Intuit QuickBooks 2026 AI Impact Report (34,000+ SMB responses)
Rates, running-cost ranges and grant offsets are Artellis's own published figures and general guidance, not survey findings.
Want the arithmetic done on your actual numbers?
Bring one recurring task and a rough sense of the hours. A discovery conversation costs nothing and ends with a costed shortlist — including a don't-build where the maths says so.