The gap between large and small companies in AI is not shrinking. It is growing. And not because small companies don't want to.
I say this as someone who sells AI mostly to companies without an IT department.
Eurostat counts that in the EU, AI is used by 55 in 100 large companies, 30 in 100 medium ones and 17 in 100 small ones. In Poland, PARP data shows the same divergence: in large companies AI use grew from 17 percent in 2021 to 33 percent in 2024, in small ones from under 2 to under 4 percent.
McKinsey sees the same thing from the speed side. In companies with revenue above a billion dollars, 54 in 100 are scaling AI across the organization; in smaller ones, a third. AI agents in at least one function are already being scaled by 40 in 100 large companies, up from 27 a year earlier. In smaller ones that share stands still at 22.
Why is that? Not because of the price of models. Token prices fell by about 80 percent within a year. The reason is what happens around the model: data, integrations, people who know how to put it together. A large company has a team that spends six months building a pilot. A small company has an owner trying ChatGPT after hours.
What follows from this if you don't have an IT department.
First: buy, don't build. MIT calculated that a tool from a specialized vendor succeeds twice as often as one built in-house. Large companies can afford failed pilots. You can't.
Second: pick a process that doesn't require integrations to start. Meeting transcription, document reading, a website chat that answers from your documentation. Each of them works without plugging into systems you have or don't have.
Third: demand from the vendor what a large company would do itself. A quality test before launch, the cost per case, a human in the loop, and a report that tells you what to fix. If the vendor doesn't provide that, the deployment will end like most pilots: quietly.
That is how we built Nexly Agents. One line of code on the website, an exam of 100 questions from your documentation before launch, a daily report from conversations, and a handover of the customer to a human when they want to buy. Everything a large company gets from its IT department, just without the IT department.
There is one limitation: a ready-made tool won't tidy up your documentation for you. If the price list exists in three versions and the terms are in an e-mail from two years ago, the agent will answer from what it has. The first weeks are always about cleaning up knowledge.
Today's task: find the one document in your company that customers ask about most often, and check whether it is up to date.
Want to see how the agent handles your customers' questions? Enter your website address and e-mail in the demo form. In 20 minutes you get an agent trained on your website's offer, ready to be quizzed.
Sources
- Eurostat, Use of artificial intelligence in enterprises (2025 data): https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
- McKinsey, The State of AI in 2026: On the road to ROI: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- PARP and Jagiellonian University, Human Capital Balance: https://www.parp.gov.pl
- MIT NANDA, The GenAI Divide: State of AI in Business 2025 (preliminary report)