"Where do we even start with AI?" is the most common question we get from UK business owners. The honest answer is that the first 90 days matters more than the next 900. Most companies that fail with AI failed in the first three months, by trying to do everything or by trying to do nothing.
Here is a clean 90-day roadmap that we've seen work across very different businesses. Boring. Sequential. Effective.
Days 1 to 30: see what's actually there
The first month is not about buying anything. It is about looking honestly at where your team's hours go.
Audit the work, not the tools
Sit down with each function (sales, ops, support, finance) and list the work that gets done in a typical week. For each item, capture three things: how long it takes, how often it happens, and how much judgement it needs.
You're looking for items that are high-frequency, time-consuming, and low-judgement. Those are the obvious AI candidates.
Get the team using ChatGPT or Claude properly
Buy your team a paid ChatGPT or Claude subscription. Run a one-hour session on how to use it well: writing, summarising, drafting, research. Most teams get a 5-10% time uplift from this alone. Cost: trivial. ROI: almost immediate.
Pick the candidate with the highest ROI
By the end of the month you should have a shortlist of three to five jobs that look automatable. Pick the one with the biggest, most measurable payoff. The first agent you ship needs to obviously work, so the second one is easy to justify.
Days 31 to 60: ship one thing
The second month is build. One job. One agent. Done well.
Define success up front
Before anyone writes a line of code, agree what success looks like. Hours saved per week. Response time targets. Error rates. Pick metrics you can measure honestly.
Build narrow before you build wide
Resist the urge to scope it large. The first agent should do one job for one team. If it works, you'll add more later. If it doesn't, you've spent very little.
Connect to real systems
The agent must touch the real tools your team uses. Not a sandbox. Not a demo. The actual CRM, the actual inbox, the actual file storage. Anything else is theatre.
Get a human in the loop
For at least the first month of operation, the agent's outputs go through a human approval step. This builds trust. Once you can see it's making good decisions, you turn off the approval for the simple cases.
Days 61 to 90: prove it, then expand
The third month is about evidence and the next bet.
Measure honestly
Compare the metrics you set in week 5 against reality. Did it actually save the hours you predicted? Did response times improve? How many cases needed a human? Be brutal with the numbers. If it underperformed, dig into why before doing anything else.
Tell the story internally
If it worked, share the result with the rest of the business. Not in vague "we adopted AI" terms. In concrete "the operations team got six hours a week back" terms. This is how you get other teams asking to be next.
Pick the next one
Go back to the shortlist from month one and pick the next highest-impact item. By now you'll know what kind of work fits agent-shaped solutions in your business specifically.
The first 90 days isn't about doing AI at scale. It's about earning the right to do AI at scale, with one clear win behind you.
Mistakes to avoid in each phase
The most common ways this goes wrong:
- Month 1: trying to do AI strategy without doing the audit. Big-picture frameworks without specific work to point at are decorative. Skip them.
- Month 2: scope creep. "While we're at it, can it also..." kills first projects. Say no. Add it later.
- Month 3: declaring victory too early. One demo isn't a result. Wait for the metrics.
- Across all months: hiring the wrong help. A consultant who can't show you working agents from past clients should not be your first hire. A vendor selling generic SaaS dressed as AI is not a partner.
Doing this with help
Roughly half our clients hire us for the audit (month one) and then build something themselves or with another partner. The other half ask us to do the build too. Both are fine. The audit alone has value, because most owners come out of it with a much clearer picture of what's worth doing.
If you'd like to start the 90 days properly, that's exactly what our AI strategy and audit is built for. Two-week, fixed-fee, comes with a written report you can take to your team or your board. Or, if you already know the work and want a build partner, just tell us about it.