How to roll out AI in a company of 30 to 250 people

Phase 0: pick a problem, not a tool

The starting question is not “which AI should we buy?” but “what costs us more than it should?”. At this size the same candidates keep coming up: meetings that end without clear tasks, commitments nobody tracks, knowledge that lives in three people’s heads, and email and Slack eating the morning.

Pick one or two problems that meet three conditions: they recur every week, they affect several people, and you can tell whether they improve without building an elaborate measurement system.

Who decides and who signs off

A rollout without an owner turns into a never-ending trial. Assign roles before you start:

• Sponsor: someone in leadership (CEO, COO or head of operations) who sets the goal and unblocks decisions.
• Pilot lead: a person inside the pilot team with time set aside to collect questions and propose changes.
• IT: reviews access permissions, sign-in, which integrations get connected and with what scope.
• Data protection (DPO or adviser): reviews the processing agreement, data location and whether an impact assessment is needed.
• Employee representatives, if you have them: tell them early, especially if the tool reads email or meetings.

Phase 1: preparation (two weeks)

Before connecting anything, settle the things that would block you later:

• Data processing agreement signed and sub-processor list reviewed.
• A decision on where data is stored and whether you want a European AI model.
• A short AI usage policy: approved tools, data that never goes in, human review.
• Which actions need approval and which can run automatically.
• An internal announcement: what you are testing, why, what each person sees and what nobody sees.

Phase 2: a pilot with one team (six to eight weeks)

Choose a team of 8 to 15 people with a lot of coordination work: operations, account management, product or sales. Pick an enthusiastic team over the one with the most problems. The pilot needs to succeed for the rest of the company to want in.

During the pilot, the lead gathers the team for 20 minutes every two weeks: what helped, what got in the way, what to adjust. Those adjustments (what counts as urgent, which actions need approval, which channels to connect) are worth more than any generic training session.

Measure adoption without measuring people

This is where many companies go wrong. To check whether AI works, they look at who uses it and how much. That turns a helper into a monitoring tool, and people stop using it naturally. Measure the work and the group, not each person:

• Team-level usage, only for groups of at least 5 people.
• Meetings with minutes and assigned tasks versus meetings without.
• Overdue or blocked commitments at the end of each week.
• Time saved, which each person sees privately, plus the company total.
• A short survey at the end of the pilot: would you keep using it? What would you drop?

Phase 3: expand in waves

If the pilot works, expand team by team rather than to the whole company on the same day. Each new team starts with rules already tuned and someone from the pilot to ask. If it doesn’t work, that is a result too: you spent two months and one team, not a year’s budget.

Where Tessa fits in this plan

Tessa is designed to be introduced this way. A pilot takes minutes to set up: each person connects Google Workspace and Slack (Microsoft 365 is in beta), and Tessa starts turning emails, messages and meetings into tasks with owners. What she reads from each person stays private, data is stored in the EU and anything leaving the company waits for approval.

For companies this size, the Business plan includes single sign-on (SSO) and hands-on onboarding, and the usual way to start is a demo followed by a guided pilot with one team.