The €720m AI Mirage: Why Irish Companies Are Throwing Money at Dead Projects
Most Irish organisations buying AI right now are not buying a capability. They are buying a story to tell the board.
The number is €720 million. That is what Irish organisations collectively lost on failed AI implementations over a recent 12-month period, according to research from global consulting firm Kearney. Picture it this way: that figure would fund the entire annual capital budget of a mid-sized Irish county council, twice over, and still leave change. It vanished into pilot programmes that never scaled, vendor contracts that promised transformation and delivered dashboards, and internal projects that collapsed the moment the enthusiastic consultant left the building.
The Theatre Versus the Engine
There are two kinds of AI adoption happening in Irish business right now. The first is AI as theatre: a proof-of-concept that looks impressive in a presentation, generates a press release, and quietly dies when nobody can explain what problem it is solving or how you would know if it solved it. The second is AI as operational engine: a specific task, automated or accelerated, with a measurable before and after. The difference between them is not the technology. It is the question asked before the purchase order is signed.
The theatre version starts with the technology and works backward to a use case. A vendor demonstrates something impressive, a decision-maker gets excited, a budget gets approved, and six months later a team of highly paid people are trying to retrofit a genuine business problem onto a tool already bought. The engine version starts with a bottleneck, a cost, or a lost revenue opportunity, and then asks whether AI can address it faster or cheaper than the alternatives. One approach is producer thinking. The other is consumer thinking wearing a producer's hat.
The 4-Step Framework Separating Winners from Wasters
The Irish companies actually generating return from AI are not smarter. They are more disciplined. Their process follows a consistent four-step logic, and the order matters because each step filters out a different category of expensive mistake.
Step 1: Name the exact problem, in money. Not "we want to improve customer experience." That is a wish. The problem is "we are losing 18% of inbound leads because our sales team takes 72 hours to respond, and each lost lead costs us an average of €4,200." That is a problem AI can address, and you will know within 90 days whether it has.
Step 2: Set a floor, not a ceiling. Decide in advance what minimum return makes the investment worthwhile. If a €60,000 AI implementation cannot demonstrably save or generate at least €120,000 in year one, the project does not start. This sounds obvious. Almost nobody does it. Instead, organisations approve budgets based on vendor projections and optimism, two things that have never once appeared on an audited set of accounts.
Step 3: Run one real pilot, not five exciting ones. The most common waste pattern in Irish AI adoption is the distributed pilot: five departments each running a small proof-of-concept, none of them resourced properly, none of them generating enough data to draw a conclusion, all of them consuming management attention. One focused pilot, in the highest-value process, run to completion with proper measurement, teaches you more than five scattered experiments. Scarcity of focus is a feature, not a constraint.
Step 4: Scale what works, kill what does not, within 90 days. The failure mode here is the zombie project: technically still alive, occasionally mentioned in meetings, consuming small amounts of resource indefinitely because nobody wants to be the person who admits it did not work. Set a review date before the pilot starts. Honour it.
What the Profitable Minority Actually Have in Common
Dublin's logistics sector is already running this playbook, where companies that defined the exact cost of a misrouted pallet before touching any technology are now compounding efficiency gains quarter on quarter. The companies generating real AI returns in Ireland share three characteristics that have nothing to do with the sophistication of the model they are running.
First, they have clean data before they start. AI does not fix a data mess. It amplifies it. An organisation feeding three years of inconsistent, manually entered CRM data into a machine learning model is not building an engine. It is building a very expensive echo chamber for its own errors.
Second, they have a named owner for the outcome, not the project. A project owner manages a budget and a timeline. An outcome owner is accountable for the revenue saved or generated. Those are different jobs held by different kinds of people. Organisations that confuse them consistently produce projects that are delivered on time and on budget and produce nothing.
Third, they treat the first implementation as infrastructure for the second. The return on AI compounds when the skills, data pipelines, and measurement frameworks built in round one reduce the cost and time of round two. Organisations chasing one-off wins reset to zero every time. Organisations building capability accumulate advantage.
The Fear Worth Naming
The risk that Irish SMEs are losing the AI adoption race to better-resourced competitors is real. But the answer is not to spend faster. It is to spend on the right thing first. A smaller business that automates one genuinely painful process and measures the result has more genuine AI capability than a larger organisation running twelve pilots and producing a glossy internal report about its digital transformation journey.
The €720 million figure is not evidence that AI does not work. It is evidence that buying technology without a defined problem, a measurable outcome, and a person accountable for that outcome does not work. That has been true of every technology wave before this one, and the lesson has never been popular because it is boring and it requires saying no to things that look exciting.
The companies that will own their categories in five years are not the ones that started AI projects first. They are the ones that started solving problems first, and used AI as one of the tools to do it.
Build the outcome, then find the technology to serve it. Not the other way around.
