You’ve started using AI tools in your business. Maybe you’re using ChatGPT to draft emails, a customer service chatbot on your website, or an AI tool that helps you generate quotes. The monthly subscription seems reasonable. But there’s a catch that many Irish business owners don’t see coming: the hidden costs of running AI can add up fast.
New research from the enterprise AI world has identified a critical blind spot. While everyone focuses on how much it costs to actually generate AI responses (the “generation” part), the retrieval layer — the behind-the-scenes work of finding and fetching the right information — often goes entirely unaccounted for.
The Hidden Layer Nobody Meters
Think of it this way: when you ask an AI assistant a question about your business, it doesn’t just pull the answer from thin air. First, it has to search through your documents, knowledge base, or database to find relevant information. That search process — called retrieval — uses computing power, memory, and APIs. And in most AI systems, nobody is tracking those costs.
The research paper, published on arXiv under the title “Cost-Governed RAG,” studied this exact problem. The authors found that in multi-tenant systems (where multiple clients share the same AI infrastructure), retrieval costs can be silently cross-subsidised. One client’s heavy usage gets invisibly paid for by everyone else. The same principle applies within a single business: the marketing department’s AI usage might be eating into the operations budget without anyone noticing.
What This Means for Your Business
If you’re running any kind of AI system that searches through business documents — whether it’s a chatbot for customer support, an internal knowledge base tool, or an AI that helps your team find information — you could be paying for usage you never see on the bill.
The researchers found that retrieval infrastructure costs can range from 3 to 9 times higher than what you might expect, depending on the system design. That means an AI tool you thought cost €100 a month might actually be consuming €300 to €900 worth of backend resources. The subscription just masks the true cost.
How to Control AI Costs in Your Business
The good news is that the same research points to practical solutions. Here’s how Irish business owners can avoid the AI cost trap:
Ask your provider about retrieval costs. When evaluating an AI tool, don’t just ask about the subscription price. Ask whether there are additional charges for searching or retrieving information. A good provider should be able to explain their cost model clearly.
Match the tool to the task. Not every AI task needs a full document search. If you’re just drafting a quick email, a simple language model without retrieval will do the job for a fraction of the cost. Reserve the expensive retrieval systems for tasks that genuinely need them — like searching through years of customer records.
Audit your AI usage quarterly. Just as you’d review your phone bill or electricity costs, set a reminder to audit your AI spend every three months. Look for unexpected cost growth, especially if you’ve added new users or data sources.
Consider hybrid approaches. The most cost-effective AI setups often combine cheap, fast models for simple tasks with more expensive, retrieval-heavy systems for complex queries. Your business doesn’t need a Ferrari to go to the shop for milk.
The research behind cost-governed AI systems is advancing quickly, and the technology to track these hidden costs is getting better. For now, the most important thing is knowing the question to ask: what’s the real cost of that AI tool — not just the subscription, but everything underneath?