Say you've decided AI genuinely matters for your business. You're past the "is this hype" question. You're not interested in another pilot that quietly dies in a year. You want to know: what do I actually do first?
Here's the plan I'd run — the same one I run with clients — for the first 90 days. No tool recommendations in week one. That's deliberate, and I'll explain why.
Days 1–30: Find the constraint, not the opportunity
Most companies start by asking "where could we use AI?" That question has too many answers and produces a scattershot list of ideas that never gets prioritized. Instead, start with: "Where is our business currently constrained by something slow, expensive, or error-prone that a leadership team member could name in one sentence?"
In the first month, I'm not looking at AI at all. I'm sitting with department heads and asking questions like:
- What takes your team the longest to do, relative to how much it matters?
- Where do errors or inconsistency actually cost you money or customers?
- What do you currently pay people to do that feels beneath their actual skill?
You'll get five to ten answers. Most won't be a fit for AI. That's fine — the goal of month one is to end up with one, maybe two, specific, well-understood bottlenecks, each with a rough sense of what it currently costs you in time or money.
This is the step almost everyone skips, and it's the single biggest predictor of whether an AI initiative works. If you can't point to the constraint before you pick the tool, you're going to pick the wrong tool, or the right tool for the wrong problem.
Days 31–60: Size it, name an owner, and set a number
Once you have your one or two candidate problems, month two is about turning them into something you could actually be held accountable for.
For each candidate:
- Put a number on the current state. Cost per unit of work, hours per week, error rate, turnaround time — whatever's honestly measurable with what you already track.
- Name one owner who is accountable for the outcome, not the technology. This should usually be the department head who feels the pain, not IT.
- Set a target and a date. Not "improve efficiency" — something like "cut average response time from 4 hours to 1 hour within 90 days of launch."
- Pick the smallest viable approach, which is very often not a big platform purchase. Sometimes it's a well-configured off-the-shelf tool. Sometimes it's a narrow, custom workflow. Resist the urge to buy something impressive before you've proven the narrow thing works.
This is also the point where I'd bring in specific tool or build recommendations — not before. By now you know exactly what problem you're solving and what "solved" looks like, which makes the tooling conversation five minutes instead of five months.
Days 61–90: Run a real pilot, with a real decision at the end
A pilot without a decision date is just a permanent low-grade experiment. In the last month, you're running the thing you scoped in month two, against the baseline you set, with a hard date to look at the number and make a call: scale it, adjust it, or kill it.
A few rules I hold clients to here:
- The pilot runs with real work, not a sandboxed demo. If it doesn't touch actual customer tickets, actual contracts, actual whatever-the-work-is, you won't learn anything real.
- Someone checks the metric weekly, not just at the end. You want to catch a false start early, not discover it on day 89.
- "Kill it" is a genuinely acceptable outcome. If the 90-day pilot doesn't move the number, that's a successful use of 90 days and a modest budget — not a failure. The failure is spending eighteen months not knowing either way.
By day 90, you should have one of three things: a working process you're ready to expand, a clear reason why it didn't work and what you'd try instead, or a validated "not now" for this particular bottleneck. All three are strategy. None of them are what most companies have after a year of scattered AI activity.
Why this order matters
Notice what's not in the first 60 days: tool selection, vendor demos, an "AI strategy" slide deck for the board. Those come later, and they come easily once the problem is actually understood. Executives who skip straight to the tool are the ones who end up, a year later, writing the "why isn't our AI working" search query.
Where I'd fit in
I don't run this process alone in a room and hand you a document — I run it with your leadership team, in real working sessions, and I stay through the pilot so month three has an honest, external voice checking the number instead of someone who's incentivized to call it a win regardless.
If you're past "should we" and ready for "what's the first concrete step," that's exactly what the first call is for. Book 30 minutes with me and we'll leave that call with your one or two candidate constraints already named.