Most rollout advice is written for either 5-person startups or 5,000-person enterprises. Neither applies if you are 50 people. This guide is the actual playbook we use for B2B companies in the $5M-$30M, 30-80 employee range, where the rollout is too big for ad-hoc and too small for committees.
At 50 people, you have just enough complexity for AI to matter, multiple functions, real workflows, paying customers, but not enough headcount for a dedicated AI team. You usually have one ops person, one IT person (often the same person), and a CEO who is paying attention but cannot run the rollout day to day.
Your rollout has three constraints that 500-person rollouts do not face: (1) no AI center of excellence (and you should not build one), (2) limited engineering capacity, and (3) no slack, the people you need to involve are already maxed.
The single most common failure mode at this size: trying to roll out AI "to the whole company at once." It never works. The right shape is sequential, one function, then the next, then the next, with one shared platform underneath.
You will roll out to one revenue-side function first. Almost always: sales, marketing, or customer success, whichever one is most visibly bottlenecked AND has a willing senior leader. "Willing" is more important than "most bottlenecked."
Do not start with a function whose leader is skeptical, even if leverage looks high. You will spend all your time on internal politics. Start where the wind is at your back.
For your chosen function, set up: (1) the LLM platform (Claude Team is our default at this size), (2) 3-5 Projects that capture the most-leveraged workflows, (3) a shared prompt library, (4) a weekly office hour with the AI lead.
Keep it boring. No agentic frameworks. No custom code. No vector databases. Just Projects with good knowledge loaded and well-written system prompts.
Once the first function is in steady-state production (typically 6-10 weeks in), pick the next function. Same platform underneath, Claude Team seats expand, but you do not switch tools. Same shape: 3-5 Projects, shared prompt library, office hour.
Resist the temptation to do all functions at once. Sequencing is your friend. People in function #2 will see what function #1 did and want it, that pull is much more valuable than push.
Around month 4-6, with two or three functions humming, you build the company-wide layer: a one-page AI policy, a quarterly cross-function review ("what's working in your function that the rest of us should steal?"), and a single point of contact for new questions.
This is also when you start saying no to scope creep. "Should we build a custom AI agent that does X?" Almost always: no, not yet, here is the buy-or-defer alternative. The discipline of saying no is what keeps a 50-person rollout from collapsing into a 500-person committee.
| Role | Time commitment | What they do |
|---|---|---|
| CEO (or COO if delegated) | 2 hrs/week | Names the AI lead, approves budget, attends one in-person training, signs the policy. |
| AI lead | 8-12 hrs/week minimum | Owns the rollout end-to-end. Runs office hours. Picks workflows with function owners. Maintains prompt library. |
| Function owners | 3-5 hrs/week initially | Own the workflows in their function. Use the tools daily. Refine prompts. Train their team. |
| Users (everyone in the affected function) | First 2 weeks: 2 hrs of training. Ongoing: it is just their job, with new tools. | Use the tools as part of normal work. Surface friction. |