Between 2020 and 2024, the share of firms using AI across OECD countries more than doubled, rising from 5.6% to 14%. The number looks good until you break it down by company size: only 11.9% of firms with 10 to 49 employees used AI in 2024, versus 40% of firms with 250 employees or more. A gap of more than three times.
The obvious question is "why". The data's own answer isn't lack of interest. It's lack of execution capacity: according to the same OECD research, half of small and mid-sized companies that haven't adopted generative AI cite the same cause — staff lack the skills to use the tool properly, and fewer than 30% of those that have adopted offer any formal training for it.
A JPMorgan Chase Institute study, published in 2026 using real transaction data from 4.6 million U.S. small businesses, helps pinpoint where the real obstacle sits. Cost stopped being the problem: the average monthly price of an AI tool fell from roughly $50 in 2019 to $20-30 in 2025. What still separates adopters from laggards is something else — companies with at least one employee on payroll adopt AI at nearly twice the rate of solo operators. It isn't the subscription price. It's having someone, even just one person, able to set the thing up and keep it running.
The project a mid-sized company can't afford
When a large company decides to adopt AI seriously, the path usually involves an evaluation committee, a controlled pilot, months of integration with existing systems, and often one person dedicated solely to it. It's an expensive path, but it exists because the company has the budget and the headcount to walk it.
A forty-person company doesn't have that luxury. There's no budget for a six-month pilot, and the person closest to "responsible for technology" is also handling the office network, the internet provider, and the sales system. Salesforce's survey of more than three thousand small and mid-sized business leaders shows exactly this gap: 75% of companies already experiment with or use AI in some form, but only 34% have actually implemented it in any routine. The distance between "trying" and "implementing" is precisely the project that lacks the budget and the team to carry it through.
The practical result is that the mid-sized company ends up with the worst of both worlds: people using personal AI tools with no standard at all, because the official path never made it off the drawing board. The problem isn't willingness to adopt. It's that the adoption model was designed for companies with a full technology team available to run the project.
Copying the large-company playbook doesn't work
The most common reaction, once leadership at a mid-sized company decides to take AI seriously, is to try shrinking the large-company project down to a smaller size: hiring a consultancy for a few months, buying the "enterprise" tier of a generic tool, or waiting until the company grows enough to justify a dedicated technical hire.
None of the three solves the real problem. The consultancy delivers a report and leaves, without anyone internally left able to maintain what was set up. The enterprise tier of a generic tool still requires someone to configure permissions, connect systems, and decide what's allowed — the underlying work is still nobody's job. And waiting to grow just postpones the problem: whoever grows without that foundation inherits the same mess, just at a larger scale, with more people already used to solving it the wrong way.
What changes the outcome isn't shrinking the large-company project. It's starting from a different point — one where the identity, the process, and the budget the company already uses day to day get reused, instead of rebuilt from scratch for AI.
What actually needs to exist
For a mid-sized company, four things need to exist before any conversation about results.
Log in with what the company already uses. If the team already signs into the network and email with the corporate directory, AI should come along with it — no separate signup, no extra password to remember, no identity project just for this. Whoever's already staff simply logs in; whoever leaves loses access at the same instant, without depending on someone remembering to shut down a standalone account.
A ready-made process template, not a project from scratch. A mid-sized company doesn't have time to design a contract-approval flow or a customer-response policy from a blank page. It needs a configurable starting point, adapted in hours, not months.
One routine first, not the whole company. Trust builds by solving one concrete thing well — answering one type of finance question, organizing one HR process — before trying to cover every department at once. A routine that works becomes the template for the next one.
One visible shared budget, not a seat license for every idle person. Instead of every department buying its own subscription, a shared budget makes clear how much is being spent and by whom, without paying for people who barely use the tool.
This is how Skyller was designed: identity coming from the directory the company already has, ready-made process templates to adapt instead of building from scratch, and credits shared across the team instead of a per-seat license.
What changes once the first routine works
The most immediate gain shows up in the month-end bill: fewer people paying for uncontrolled personal subscriptions, less rework fixing what a generic tool produced with no context about the company. But the gain that really matters shows up later, once the first routine actually works.
At a forty-person company, whoever figures out the right way to organize a finance response is usually the only one who knows how to do it. If that knowledge stays locked to them — in a personal folder, in a chat history only they can see — the company keeps depending on one specific person being available. When that same process can be reused by someone else on the team, with the same permissions and the same standard, the gain stops being individual and becomes the company's.
That's the effect that makes the difference between a mid-sized company that buys an AI tool and one that gradually builds its own way of working with AI — without ever needing a technology team the size of a multinational's to get there.
A checklist to start with
Before comparing vendor proposals, it's worth deciding these points internally:
- Pick a routine, not a project. Favor something that already repeats every week — a standard response, a recurring report — over trying to cover the whole company at once.
- Confirm the login comes from what already exists. If the vendor's proposal asks for a signup separate from the company directory, someone will have to manage two access lists forever.
- Ask who maintains it after the vendor leaves. If the answer depends on an outside consultancy coming back every time something needs to change, the process will never really belong to the company.
- Measure the budget by team, not by person. A fixed, visible amount is easier to justify internally than adding up individual subscriptions scattered across the corporate card.
- Ask for a configurable starting point. A ready template the team adapts in hours is worth more than a blank screen and the promise that "you can build anything".
