The 4-week AI activation playbook for a 100-person firm

Four weeks is the right shape for an AI activation sprint in a 100-person company. Less than that and you don’t have time to measure anything. More than that and you’ve drifted into a discovery project that produces slides instead of software. This is the week-by-week playbook we run — deliverables, exit criteria, and the three checkpoints that decide whether the workflow scales or gets killed.

We’ve run this shape on services firms, manufacturers and non-profits between 50 and 250 employees. It works for the boring reason that the constraints are honest: one workflow, one named owner, one production outcome, one ROI number you can carry to the next leadership meeting.

The opposite of an activation sprint is a six-month “AI transformation roadmap” that produces 80 slides and zero production software. We’ve never seen one of those pay back.

Pre-flight: the two-week setup nobody wants to do

Before week one, two things must be true. If they’re not, the sprint will fail and we’d rather tell you that in week zero than week three.

• You’ve named a single business owner.Not “IT.” Not “marketing leadership.” One person, ideally a department head, who owns the workflow and will decide whether the new version is better than the old one. If three people are accountable, nobody is.
• You’ve picked one workflow. Specific, repeatable, lives mostly within one team’s span of control. Examples that work: “client meeting prep pack,” “month-end invoice intake,” “inbound RFP triage.” Examples that don’t: “marketing automation,” “use AI in operations.”

If both are true, run two pre-flight sessions: a 60-minute baseline timing exercise (three users, three reps each), and a 30-minute system-of-record map. You now have a stopwatch number for the workflow and a list of the systems it touches. That’s the artefact you’ll grade everything against.

Week 1: Baseline + design

Outcomes

• Documented baseline: avg. minutes per workflow run, frequency, headcount affected.
• One-page solution design: what we build, on which tile, with which integrations.
• Risk register: data, IP, access, vendor dependencies.
• Named internal lead (often a power user, not IT).

What happens

Three working sessions. First session: the business owner walks the workflow end to end while we time it and screenshot every artefact. Second session: we propose two solution patterns (e.g. “Copilot Studio agent over SharePoint + Power Automate trigger” vs “M365 Copilot prompt-pack + shared SharePoint site”) and the owner picks. Third session: solution review with IT + security to clear the data path.

Week 2: Build (first cut)

Outcomes

• Working version of the workflow in a sandbox environment.
• Demoable to the business owner end-to-end at end of week.
• Three test cases authored by the owner, not by us.

What happens

Build days. For most engagements this is one practitioner pair- programming with the named internal lead inside the customer tenant — not behind a curtain. Most of the work is in Copilot Studio, Power Automate, Power Apps or some mix of the three. If we’re writing custom code, this is also where the Azure AI Foundry parts get scaffolded.

End-of-week demo against the three test cases. Two outcomes are acceptable: “it works on all three” or “it works on two and we know exactly what’s wrong with the third.” Anything worse and we surface it now.

Week 3: Pilot (real users, real load)

Outcomes

• 3–5 real users run the workflow daily on real work.
• Stopwatch timing on every run, captured in a shared sheet.
• Daily 15-minute stand-up with users to surface friction.
• Punch list of fixes, prioritised against impact.

What happens

The workflow goes live for the pilot group. Crucially, the old process stays available — users can fall back. We’re not measuring whether people will use the new flow under coercion; we’re measuring whether they do when given a choice.

Daily 15-minute stand-up: one question to each user — “did you use the new flow, and if not why?” Two outputs: a friction log and a fix backlog. Fix backlog gets worked in real-time by the internal lead with us in support.

Week 4: Hand-over + scale decision

Outcomes

• Production cutover with the old process retired or archived.
• Owner-authored runbook: how to operate, how to change, who to call.
• 90-minute training session with the broader affected team.
• Final ROI math: hours × frequency × headcount × realisation factor.
• Scale-or-kill decision documented and signed by the business owner.

What happens

Cutover. The old process is retired — not deprecated, retired. The runbook is written by the internal lead, not by us (this is a load-bearing detail; if we write it, nobody else can read it). Training session for the whole affected team, 90 minutes, hands-on.

Then the ROI math gets put on the wall. Hours saved per run, runs per week, headcount, realisation factor, contribution- margin conversion. We argue about every multiplier; the business owner signs the final number. That number is the defensible value of this one workflow and the basis for the scale decision.

Anti-patterns that derail this playbook

1. The committee. If approval requires four people in three time zones, the four weeks are over before you start.
2. The phantom owner. The named business owner who actually attends one of nine working sessions. Cancel and reschedule the sprint until the owner is available; do not proceed.
3. The scope creep. “While we’re at it, can we also handle invoices?” No. Park it. The next sprint will handle invoices if the first one earns the right.
4. The missing baseline. Without timed pre-rollout data, the final ROI math is unverifiable and the board conversation falls apart.
5. The hidden licence cost. Some workflows push you into premium connectors, AI Builder credits, or Copilot Studio message packs. Surface those in week one, not week four.

What it actually costs

Honest numbers for a 100-person services firm running one sprint:

• External consulting (us, fixed-fee): typically $35–$55K.
• Internal time: ~ 0.3 FTE for four weeks (the named lead).
• Licence delta: $0–$1,500/month depending on connectors and AI Builder.
• Total all-in year one: $50–$80K for one production workflow.

A workflow that saves a 100-person firm 1,000–2,000 hours per year pays back inside the first year at any reasonable contribution-margin conversion. If your candidate workflow can’t clear that bar even on the optimistic side of the math, pick a different workflow.

Bottom line

Four weeks. One workflow. Three checkpoints. One business owner. One ROI number. That’s the entire shape of an AI activation sprint that produces software instead of slides. Everything else is window dressing, and most six-month AI “transformation” engagements are window dressing.

If you want help running this shape on your own first workflow, the free AI Readiness Assessment is the lowest- friction starting point — it surfaces the top candidate workflows and rough-sizes the ROI before any commitment.