The question we get asked most often after the pricing one is how long an AI rollout actually takes. The honest answer for a 25-person SME is six working weeks, give or take a fortnight depending on the stack and how clean the documents are. Not six months; the long-rollout stories are usually about an enterprise with a procurement gate and a regulator. For an SME with three founders and a clear stack, six weeks is the shape, and the shape repeats.
We have run this cadence enough times that the weeks have names. Here is what each one looks like and what the team should be doing inside it.
Week 1: discovery
The week the rollout is fastest to get wrong if it is skipped. The aim is not to talk about AI; it is to understand the work the team actually does, so the rollout has a target rather than a slogan.
What we run: four 45-minute conversations across the business, one per main work area. A founder, a sales person, an operations person, a finance person. The conversation is not "what AI workflows do you want"; it is "walk me through your Monday morning, your Wednesday afternoon, the thing that ate three hours last week". The AI use-cases drop out of those conversations more honestly than they do out of a brainstorm.
What the client owes us: access to the people, a calendar, and a list of the tools they already pay for, which is the input we need to be useful. The output of week one is a shortlist of three to five workflows where AI looks like it would land, ranked by how clearly the risk band sits in band one or two.
Week 2: policy and stack
The week most rollouts skip and then regret. The aim is to have the acceptable-use policy written, the approved tools chosen, and the data-handling commitments in writing before any seats are bought.
What we run: a 90-minute session with the SLT to agree which AI product fits the stack (M365 shop, Workspace shop, or other), what the approved-tools list will say, and what the data-classification table looks like. The session produces a draft policy by end of day. The draft is then circulated for a 48-hour comment window, refined, and signed by Friday.
What the client owes us: someone with authority in the room and a willingness to make calls. The policy that gets stuck in committee delays everything downstream.
By the end of week two, the business knows which AI product it is buying, has a signed policy that covers it, and has a configured tenant or workspace ready to provision seats.
Week 3: pilot start
The week the work starts being visible. The aim is to get five seats live, the pilot users briefed, and the first workflows running before Friday.
What we run: the seat provisioning (usually a few hours of admin work, sometimes a day if the tenant needs cleanup first), a 60-minute briefing with the pilot five, and one-to-one 30-minute sessions through the week with each user to get their first AI-supported workflow up and running. The briefing covers what the product does, what the policy says, where the data-classification line is, and what the four pilot measurements will be.
What the client owes us: the five users available for an hour each in week three, not later in the month. The pilot that loses its first week is a pilot that takes nine weeks instead of six.
By end of week three, the five are using Copilot or Gemini or Claude or ChatGPT in real workflows. The dashboard has its first numbers.
Weeks 4 and 5: pilot run with light training
The weeks where the team finds out what works. The aim is to let the use settle into a real pattern, with light support and the first round of workflow tuning.
What we run: a 15-minute check-in at the end of week four with each pilot user (five conversations, an hour and a quarter total), plus a 60-minute group session in week five where the pilot users compare notes. The group session is where the patterns become visible; one person discovers a use-case that another had not tried, and the pilot gets richer.
We also begin building the prompt library in these weeks. Two or three prompts per pilot user, the ones that produced the best output, captured with the source workflow they fit. The prompt library is a deliverable; we do not leave the rollout without one.
What the client owes us: the time. The check-ins are short but they have to happen. A pilot user who is too busy to do a 15-minute check-in is a pilot user who is too busy to use Copilot, and that is itself a data point.
Week 6: decision and roll-wider plan
The week the rollout becomes a permanent capability or a refund conversation. The aim is to run the structured 60-minute review, make the keep-or-drop call, and write the plan for the next 20 users if the answer is keep.
What we run: the review meeting with the pilot five plus the SLT, the dashboard summary across the three months, and the roll-wider plan if appropriate. The plan covers who gets seats in months two and three, in what order, with what onboarding, and at what budget.
If the decision is to roll wider, the next three to four months are mostly business-as-usual: more users onboarded a few at a time, the prompt library grows, the policy gets refined as edge cases come up. The intensive work is done.
What runs in parallel through the six weeks
Three workstreams that do not get a week of their own but matter to the rollout landing cleanly.
Information security review. The vendor's data-handling commitments, retention configuration, audit logging, SSO and conditional access. This is set up in week two alongside the policy and verified in week three before the pilot users go live. By week four it is invisible, which is the right state for it.
Cost tracking. The licences, the consultancy time, the internal time spent on the rollout. We build a simple sheet in week one and update it weekly. By week six, the business has a real per-user-per-month cost figure including the rollout overhead, which is the figure that matters for the roll-wider conversation.
Documentation. The policy, the prompt library, the use-case map, the dashboard. All live in one place the team can find. We hand it over at week six and it becomes the SME's own asset.
The cost shape
Indicative for a 25-person SME, six-week engagement:
- Assessment and policy work, weeks 1 to 2: from £7,500.
- Pilot deployment, configuration and training, weeks 3 to 6: from £15,000 depending on stack complexity.
- Five seats of the chosen AI product, three months: about £350 to £450 at SME pricing.
- Review and roll-wider plan: included in the deployment fee.
Total cash outlay for the six weeks: roughly £22,850 to £22,950. If the decision at week six is "do not roll wider", the seats stop costing and the business has paid for the discovery, the policy, and the honest answer. If the decision is to roll wider, the per-user cost going forward is the seat licence plus a light support layer that folds into the Managed Services contract.
Where this lands with us
This cadence is the AI Enablement engagement. We sit alongside the team for the six weeks, run the discovery, draft the policy, configure the deployment, run the pilot, and write the review. The team owns the outcome; we own the cadence.
The mistake we see most often is treating week one as administrative and rushing to week three. The discovery is the difference between a pilot that lands and a pilot that drifts, and it is also the cheapest part of the engagement to do well.
Thinking about an AI rollout and want to see whether the six-week shape fits your business? Drop us a note at info@jmopartners.co.uk.
JMO|Partners · Enterprise IT, sized for SMEs.