Organisations are starting to learn that there’s no such thing as an artificial intelligence (AI) free lunch. Businesses attempting to embed AI in operations have turned to token-maxing strategies where they require employees to use AI for a certain proportion of their time or a percentage of tasks.
However, as the cost of AI tokens mounts, the value of the outputs is increasingly called into question, forcing those businesses to calculate whether it would be cheaper to use human labour instead.
We look at the pushback against token maxing and ask if it is likely to be a significant constraint on AI adoption.
AI adoption is accelerating across Dublin businesses, but the next phase will increasingly be about demonstrating value rather than adoption for adoption’s sake, says Mia Finnegan, public affairs manager, Dublin Chamber. “Dublin Chamber research shows that businesses are primarily deploying AI to improve efficiency and automate processes, but businesses also report a range of barriers to greater adoption, including skills, internal resources and funding.
READ MORE
“As AI becomes more deeply embedded in day-to-day operations, businesses will become increasingly focused on the return they are getting from that investment. The question should not be how much AI an organisation is using, but whether it is improving productivity, reducing costs or enabling employees to focus on higher-value work. For many firms, particularly SMEs, successful adoption will depend on having the right skills and internal capability to identify where AI can deliver genuine business value.”
AI can be cheap to try and expensive to run at scale, says Mark Kelly, founder, AI Ireland. “A single interaction costs almost nothing. But then usage spreads across hundreds or thousands of employees, agents start performing multiple steps, and you add integration, monitoring, security and human review. That is when the economics change.”
Across more than 1,000 artificial intelligence projects that Kelly has reviewed through the AI Awards, the technology itself is rarely the only issue. The bigger challenge is often the total cost of operating the solution reliably at scale, he says. “The mistake is looking only at the price of the model rather than the cost of the complete process.”
We are increasingly seeing artificial intelligence being priced and consumed through tokens, the units of text and data processed by AI models, says David Lee, chief technology officer, PwC. “Every prompt, response, document summarised or workflow automated consumes tokens, meaning that as AI usage grows, so does the associated cost. However, the real challenge for businesses is not controlling token consumption, but ensuring that token spend delivers measurable business value from the AI.
“The most successful organisations focus on the outcomes and value gained from the AI consumption. Rather than only measuring how many tokens are used, they also track the outcomes generated, such as productivity gains, time savings, revenue growth, improved customer service and operational efficiencies. This allows businesses to understand the true cost per outcome and determine whether AI investment is delivering a return.”
There are dangers in focusing solely on AI costs and usage, Lee says. “Uncontrolled usage can lead to rising costs without corresponding business benefits, while blanket deployment of premium models for every task can create unnecessary expense. Organisations also risk generating productivity gains that never translate into financial value if workflows, governance, accountability, and the right guardrails are not redesigned around AI.
“As AI becomes more widespread, competitive advantage comes from using AI for positive disruption, re-imaging how the business can work differently, efficiently and for specific uses and outcomes. Businesses that combine strong governance, visibility of token consumption and a clear link between AI spending and business outcomes will be best positioned to maximise value while keeping costs under control.”
Tokens become important when volume increases, the information being given to the artificial intelligence becomes larger, or the AI starts performing multiple steps on its own, says Kelly. “As you move up those levels from basic chats to agents, the amount of computing and AI usage generally increases dramatically.
“A simple question to a chatbot might involve one interaction. An AI agent might search information, call another system, analyse the result, retry something and then produce an answer. One business task can therefore create many AI interactions behind the scenes.”
That is why something that looked extremely cheap during a pilot can become material when it is running thousands of times every day, says Kelly. “The key measure should be cost per successful task, not cost per token.”
The value challenge is clear in Ireland, says Lee. “PwC’s 2026 CEO Survey found that only 17 per cent of Irish CEOs reported higher revenue from AI in the previous 12 months, while 23 per cent reported lower costs. PwC’s AI Agent Survey shows a similar gap: 53 per cent of Irish respondents saw measurable productivity gains from AI agents, but only 38 per cent said these gains had turned into real cost savings.”
Artificial intelligence is a positive disrupter and does not come cheap, Lee says. “The findings show you need to use AI to disrupt and reinvent your business to really get value from it – turning higher productivity into financial returns.
“Productivity improvements only create value when businesses reinvent their businesses, redesign workflows, strengthen governance, and link AI spending directly to measurable business outcomes.”
At the same time, artificial intelligence introduces new costs, particularly through token consumption, which can reduce or offset savings if usage is not well managed, explains Lee. “As a result, companies need a disciplined approach that focuses on cost per outcome and return on AI, rather than simply increasing AI adoption or usage.”
The most important number is not what artificial intelligence costs, says Kelly; it is what a good outcome costs. “An expensive AI model can be the cheaper option if it completes a task correctly without intervention. A cheap model can be extremely expensive if its work has to be checked, corrected and completed again by a person.
“After training more than 10,000 leaders, the pattern I see is that the organisations making progress are moving beyond experimentation. They can increasingly answer three simple questions: What problem are we solving? What does it cost us? What are we getting back? That is where the AI conversation is heading: from adoption to economics, and ultimately to business value.”
Rising artificial intelligence consumption costs could become a constraint if organisations fail to manage them effectively, says Lee. “As AI usage grows, token consumption and associated costs can increase rapidly, making it essential to link AI spending to measurable business outcomes. The real challenge is not the cost of AI itself, but whether organisations can demonstrate value from that spending.
“Leading organisations are responding by focusing on cost per outcome, using governance and monitoring to control consumption, and matching workloads to the lowest-cost model that can deliver the required result. In the long term, rising AI costs will drive greater discipline around how AI is deployed, measured, and governed, rather than preventing adoption.”
“If I showed you a shortcut that saved five minutes on your drive home, you could immediately measure the benefit because you know how long the journey normally takes,” says Kelly.
“AI should be evaluated in exactly the same way,” he continues. “Measure the current process first: how long it takes, what it costs, how many mistakes happen and how much rework is required. Then run the AI process beside it and compare the same measures.”
Importantly, Kelly says to include the full cost of the AI process: software, model usage, integration, supervision, mistakes, security and human review. “The number I like is cost per correctly completed task. One of the most commonly missed costs is human supervision. If someone has to check every AI output, their time is part of the business case.”
