You’ve decided to pursue AI. The board is supportive. Budget is allocated. Expectations are high. The pressure to “do something” is real, and the temptation to move fast is strong.
This is exactly the moment to slow down.
The first 90 days of your AI initiative set the trajectory for everything that follows. They determine whether you’ll build something valuable or something that consumes resources without delivering value. They establish patterns of work — good or bad — that will either accelerate you or anchor you for years.
Here’s a tested roadmap for those first 90 days.
Days 1-30: Foundation
Stakeholder alignment and conflict surfacing. Spend this time getting everyone in the room. Not just the executives who decided to pursue this, but the people who’ll actually have to use it, support it, integrate it. Where are the disagreements? What’s the unstated resistance? It’s better to surface this now than to discover it halfway through implementation.
Current state assessment. Understand how work actually happens right now. Not how org charts suggest it happens, but what people actually do. Where is the pain? Where is there friction? This is often different from what the original problem statement suggested.
Success definition with explicit sign-off. What does success actually look like? Not “improved efficiency” — that’s too vague. Not “better outcomes” — that doesn’t mean anything until you define it. What are the specific, measurable outcomes that would indicate success? And critically, who has signed off on this definition? If leadership can’t agree on success metrics, your initiative is already in trouble.
Solution direction recommendation. Based on everything above, what’s the right direction? Build vs. buy? Custom vs. off-the-shelf? This should be a recommendation, not a final decision, but it should be directional.
Spend the full 30 days on this. Resist the pressure to move faster. Foundation matters.
Days 31-60: Pilot
Define the narrowest useful pilot. The temptation is to pilot everything. That’s the wrong instinct. Define the smallest useful scope — one department, one workflow, one problem — where you can prove the concept works and learn what you need to learn.
Build minimum viable AI solution. Not a complete solution. The most basic version that lets you test your hypothesis. Build it fast. It’s going to have limitations. It’s going to be rough around the edges. That’s fine. You’re learning, not delivering production.
Deploy, gather feedback, iterate quickly. Get it in front of users. See what works. See what breaks. Iterate fast — days or weeks, not months. This is where you learn whether your assumptions were correct.
Days 61-90: Scale Planning
Honest pilot review. What worked? What didn’t? Where did you learn something unexpected? This isn’t a celebration or a failure analysis — it’s a genuine assessment of what the pilot showed you.
Production architecture design. Now that you know what actually works, design the system for production. This is different from pilot architecture. Production needs to be robust, secure, integrated with existing systems, and maintainable.
Change management planning. How will you bring the broader organization along? If your pilot succeeded with enthusiasts, how will you get the skeptics to adopt it? What training do people need? How will you address resistance?
Business case and roadmap for leadership. What’s the business impact? What’s the investment required? What’s the timeline for scaling? Leadership needs clarity on all of this before you ask them to fund the next phase.
Why This Structure Works
Foundation work feels slow when you’re doing it, but it prevents you from building the wrong thing.
Piloting with the right scope lets you learn without breaking the organization.
Scale planning based on pilot results means you’re scaling something that actually works, with a clear plan for how to bring people along.
Most organizations try to skip the foundation work and rush to the pilot. Then they’re shocked when the pilot struggles because no one aligned on what success looks like or what the current state actually is.
Don’t do that. Spend 30 days on foundation. Spend 30 days on pilot. Spend 30 days on planning. In 90 days, you’ll have learned more than organizations that rushed and failed to learn.
And then the real work of building organizational AI capability begins.
