Mid-market AI rollouts split cleanly into two flavors: planned (written brief, named owner, 90-day calendar, deliver gate, kill criteria) and gambled (license + demo + hope). Most teams never make the decision and end up gambling by default. This piece lays out the five-line plan checklist and the Microsoft tooling that maps cleanly to each row.

The decision most mid-market AI teams skip

The decision is whether the project has a plan or a gamble shape. It is not whether to use AI. That question is settled. The interesting question is the rollout shape.

A plan is boring on purpose. It looks like a one-page brief on a wall, a name next to each row, and a 90-day calendar with a check-in every week. A gamble looks like a Copilot license purchase, a vendor demo, and a Loom sent to the team.

Both shapes are common. The plan shape ships value inside 90 days. The gamble shape shows up next year in a slide deck about why AI 'is not ready yet.' The middle does not exist.

The five-line plan checklist

1. Written brief

One page. Who the agent serves. What the agent does. When it ships. If the brief lives in someone's head, it does not exist. Mid-market teams underrate this row because they are used to small projects that did not need a brief. AI projects do.

2. Named owner

One person. Not 'IT' and not 'the team.' One person owns the outcome and meets weekly with the executive sponsor. The owner does not have to be technical; the owner has to be accountable.

3. 90-day calendar

Pilot in weeks 1 to 4. Deliver in weeks 5 to 8. Expand in weeks 9 to 12. Anything longer is a gamble in a sweater. Mid-market teams that try for six-month rollouts usually end up shipping nothing in six months.

4. Deliver gate

A human reviews the AI output before it ships. Every time. Not 'eventually a review process.' Now. The gate is what separates AI-as-tool from AI-as-decision-maker and keeps liability where it belongs.

5. Kill criteria

What makes you turn the agent off and try a different one? Write that down before the pilot starts. Mid-market teams skip this row because killing a pilot feels like failure. It is not. Killing a pilot at week 4 is a 90-day saved.

What the gamble actually looks like in the wild

The gamble shape has a vocabulary you can spot from across the room.

'We are exploring AI.' 'We are doing a Copilot pilot.' 'We bought licenses and we are seeing what people do with them.' 'We are slow-rolling it.' 'We are following Microsoft's roadmap.'

None of those sentences contain a written brief, a named owner, a calendar, a gate, or a kill criterion. The team has not made the rollout decision. The team has bought time before making the rollout decision. That time is expensive.

Why mid-market needs a 90-day rollout, not a Center of Excellence

Large enterprises can afford to start with a Center of Excellence: a chartered group, a budget, an 18-month plan, and a research-grade approach to evaluation. Mid-market teams cannot. The math does not work.

What works at mid-market scale is a 90-day rollout. One agent. One owner. One pilot. One deliver gate. One kill criterion. Ship value in 90 days or kill the project and try a different agent.

This is closer to how mid-market teams already ship CRM rollouts, ERP migrations, and security upgrades. The mistake is treating AI as a different category of project that needs a different process. It does not. AI is software with a probabilistic interface. The rollout discipline is the same.

Microsoft tools that map cleanly to the plan

The Microsoft stack now has clean mid-market entry points for each row of the plan.

Written brief: Microsoft Foundry Skills and Toolboxes

Foundry is now the standard place to define what an agent does. Skills describe behaviors; Toolboxes describe APIs the agent can call. Both are written artifacts that map cleanly to the brief.

Named owner: Agent 365

Agent 365 sits at the tenant level and gives every agent an identity that Conditional Access can hold accountable. The 'named owner' for the agent's behavior in production is the security plane, not just the project sponsor.

90-day calendar: Copilot Studio multi-agent

Copilot Studio multi-agent orchestration is now generally available. The platform supports the handoff patterns mid-market teams actually run: one agent triages, one agent acts, one agent reports. A 90-day rollout maps cleanly to the platform's deployment model.

Deliver gate: Purview + human review

Purview for data lineage and a human review step at the deliver gate. Together they answer the auditor's questions about what data the agent saw and what output a human signed off on.

Kill criteria: telemetry and termination

Agent 365 + Defender for Cloud Apps make the kill criterion enforceable. If the agent does not hit the metric, the agent does not get the workload next quarter.

What to write on the one-page brief before you buy the first license

If you do nothing else this week, write the one-page brief. The brief has five sections.

Section 1: the job. What does the agent do, in plain English. Two sentences.

Section 2: the trigger. What event starts the agent's work. One sentence.

Section 3: the output. What does the agent produce. One sentence.

Section 4: the gate. Who reviews the output before it ships. One name.

Section 5: the kill condition. What makes you turn the agent off. Two sentences.

That is the brief. Fits on one page. Anyone in the company can read it in 60 seconds and tell you whether the rollout has a plan shape or a gamble shape.

The webinar

ArchitectNow is walking through this checklist live on a 45-minute webinar Tuesday June 24, 1:00 PM CT / 2:00 PM ET. Free. Register at architectnow.net/webinars/ai-rollout-plan.

Frequently asked questions

Is a 90-day calendar realistic for mid-market AI rollouts?

Yes, when the scope is one agent and one workflow. A 90-day calendar for 'all of AI' is not realistic; a 90-day calendar for 'route inbound support tickets to the right queue' is.

Do we need Agent 365 to run a plan-shaped rollout?

No, but the identity-governance row of the plan gets harder without it. Agent 365 is licensed per user and went generally available May 1, 2026.

What if we have a Center of Excellence already?

Keep it for strategy. Run the 90-day rollouts inside the operating units. Mid-market teams do not need the CoE to ship; they need the CoE to set the standards the units ship against.

How do we know if our team is gambling?

Ask the project lead for the one-page brief. If there is no brief, you are gambling.

What this means for your next 30 days

Write the brief for one AI rollout your team is considering. Find a named owner. Put a 90-day calendar on the wall. Define the deliver gate and the kill criterion. If any of those rows are missing 30 days from now, you have not made the rollout decision. You are gambling. That is fine, as long as everyone is honest about which shape the project has.

About ArchitectNow

ArchitectNow is a Microsoft Solutions Partner with four designations (Digital and App Innovation, Data and AI, Infrastructure Azure, Support Services) and the AI Apps on Azure Specialization. We design and deliver mid-market AI rollouts on the Microsoft stack. Talk to us at architectnow.net.