AI for mid-size companies: start with one department, not a strategy deck.

The mid-size squeeze: too big for one agent, too small for a consultancy programme

A company with 25 to 250 people sits in an awkward place when it comes to AI. It is too big for one clever agent to make a difference: the work is spread across sales, service, finance and operations, each with its own tools and habits. And it is too small for the programmes large consultancies sell, which assume an internal AI team, a transformation office and a budget that runs to six figures before anything is built.

So mid-size companies tend to fall into one of two patterns. Either everyone gets a chat assistant licence and the results stay personal, or the leadership team commissions a strategy and waits. Neither changes how the work gets done.

Why the strategy-deck route fails at 25 to 250 staff

A strategy deck is built for organisations with people whose job is to execute it. A mid-size company rarely has them. Three things usually go wrong:

1. Nobody builds. The deck lists twenty use cases. The people who would build them are the same people running the departments. Nothing starts.
2. Everything is scoped at once. A company-wide plan treats every department as equally ready. They are not. The plan stalls on the slowest one.
3. The foundation is skipped. Without a shared place where the company’s knowledge is written down, each pilot builds its own. The second pilot cannot reuse the first.

The alternative is not “no plan”. It is a short plan that ends with something running. We explain the difference between buying advice and buying working agents in AI agent vs AI consultant.

How to choose the first department

Put three numbers against each candidate department:

• Where the hours go. Hours per week spent on recurring work: answering the same questions, moving data between systems, checking documents, preparing the same reports. A rough tally over one week is enough.
• How rule-based the work is. Could a new hire do it from a written procedure? If yes, an agent probably can too, with a person approving.
• What an error costs. A wrong draft reply that a person catches is cheap. A wrong payment is not. Start where errors are cheap to catch.

The department that scores highest on hours and rules, and lowest on error cost, goes first. In most mid-size companies that is sales operations, tier-1 customer service or finance operations.

What 60 days look like: five agents, one brain, one KPI

A first department typically needs a small team of agents, not one. In our Workforce Pro build that is five coordinated agents covering one department in 60 days, for a fixed €15,000 with a 14-day money-back guarantee. The shape matters more than the package:

• Five agents across the levels of the work. Some prepare information for the head of department, some run the daily flow, some take over high-volume tasks. Each has a written list of what it may do alone and what needs a person.
• One company brain. The procedures, prices, policies and exceptions the agents need are written once, as plain files the team can read and correct. This is stage 1 of the AI Company Brain, and every later department builds on it.
• One KPI. Response time, invoices processed per day, hours handed back. Measured before launch and every week after.

Before any agent goes live, it passes a set of test cases with known answers. After launch, a person reviews its work closely for the first weeks.

Signals that it is time for the second department

• The KPI has moved and stayed moved for four weeks.
• The team overrides the agents less often, but still checks them.
• The company brain is being kept up to date without being chased.
• Another department head starts asking when it is their turn.

When those are true, the second department joins on the same brain and the same rules. That is the point where a roadmap becomes useful, and our AI transformation roadmap shows how the phases run over twelve months.

Lessons from smaller builds that carry over

Most of our published field notes come from smaller firms, such as the accounting practices in 5 AI agents for accounting firms: what broke. They are not mid-size case studies, but three lessons apply directly to a first department in a larger company:

• “Our standard process” is rarely one process. Departments often run two or three variants. Ask before building.
• Test on the busiest period, not a quiet month. Agents tested only on calm weeks raise false alarms when the real volume arrives.
• Find out where the data actually arrives. Customers reply to whichever channel they were contacted on, not the one the process assumes.

Where to start

For a mid-size company, the first step is a map of where the hours go and which tasks agents should assist with or take over. A Transformation Assessment (€3,000, four weeks) produces that map with a phased plan and a fixed price per phase. If you want a shorter first conversation, the €297 AI Audit covers one session and a written report.

How the whole programme runs, for mid-size and large companies, is on our AI transformation services page.

---

Key terms in this post: AI agent · AI workforce · eval suite · human-in-the-loop · AI transformation