Forward deployment for a 20-person business

Forward deployment rests on a simple idea: an engineer sits with the customer until the system works on their data, in their workflow. AI and data companies use it to land large enterprise accounts, with teams embedded for months.

I think it fits a 20-person business even better. In a small company, the person with the problem, the person who signs and the person who will use the tool are often in the same room. There is no procurement cycle to survive. There is also no data team, no IT department and no patience for a six-month programme.

Start where the work happens

The first days are for watching, not building. How does a booking actually arrive? Who answers the same question twenty times a week? Which spreadsheet does the owner open first on Monday? You are looking for one workflow where time or money leaks every week, and one number that will say whether you fixed it.

Write that number down before you build anything. In a small business it is usually plain: bookings through your own website instead of a third-party platform, hours lost to admin, quotes sent the same day. If you cannot name the number, you are not ready to build.

Fit the day they already have

A small team will not adopt a new system because it is clever. It adopts what fits the day it already has, so the work happens inside the tools the team already uses, and the new thing should feel like a shortcut, not a project.

Data rules apply at every size

Twenty people still handle customer data, and privacy law does not shrink with headcount. Decide early where the data may go. If it cannot leave the building, the model has to come to the data. In Licita, my research project on public procurement, the same pipeline runs on a hosted API or on open-weight models on the customer’s own servers, by changing one setting. A small business deserves the same choice, sized to its budget.

Handover is the product

The engagement is finished when the owner can run it without me. That means a short runbook where every error says what happened, why, and what to do next. It means a named person on their side, a monthly check that takes minutes, and running costs that fit inside a small margin.

I run it in three blocks of thirty days:

• Days 1–30: one workflow live on real data, reviewed with the people who use it.
• Days 31–60: part of the weekly routine, owned by someone on their side.
• Days 61–90: measured against the number we wrote down, then a decision on what comes next.

A small business does not need an AI strategy. It needs one workflow that works on Monday, and someone who stays until it does.