How to build a 90-day AI roadmap for your business
By Keith Hamilton · · 4 min read
A 90-day AI roadmap is the right unit of planning for a small or medium business: short enough to stay accurate, long enough to deliver and measure something real. A long AI strategy document rarely changes much in an SME, because by the time it is finished the tools have moved on and the team has lost interest. This guide sets out how to build an AI roadmap for a small business and what it should contain.
Why plan AI in 90 days?
Ninety days is one quarter, which fits the way boards already review the business. It gives you time to find the right opportunity, put something into use and measure it properly. It is also short enough that a wrong choice costs you a quarter, not a year.
The aim is not to transform the business in a quarter. It is to make one or two well-chosen improvements, prove whether they work, and use what you learn to plan the next quarter.
How do you find AI opportunities in weeks 1 to 3?
Start with the work, not the tools. Ask each team which tasks follow the same steps every time and take up hours each week. For each one, note:
- how many hours a week it takes, and whose time it is
- how often it happens, such as calls, orders or invoices
- what data it touches, especially personal or payment data
- how you would know if it got better
Turn the hours into money using the real cost of the people doing the work. This gives you a baseline, and it lets you compare tasks fairly rather than by who argues loudest.
Then score each opportunity on three things: value, ease of putting it in place, and risk. A score of one to five on each is enough.
How do you prioritise AI projects?
Prioritise one or two pieces of work that are valuable, easy to put in place and low risk. The most common roadmap mistake is trying to do too much with limited time and attention.
A worked example
A distributor has three candidates:
- Keying in orders from emails. High value, fairly easy, low risk.
- Answering routine delivery questions. Medium value, fairly easy, some customer data involved.
- Forecasting stock with AI. High potential value, but hard, and dependent on data that is not yet clean.
The right 90-day roadmap takes on the first, prepares the ground for the second, and parks the third with a note on what needs to be true before it is worth revisiting.
Writing down what you are not doing, and why, is part of the roadmap. It stops the same ideas being reopened every month.
How do you deliver and measure in weeks 4 to 10?
Deliver each piece of work with one named owner and a measure of success agreed in advance. The owner runs the process and makes decisions about it, and the measure comes from your baseline.
Before anything goes live, check the basics:
- you have an AI policy that covers the tools being used
- you know what personal data is involved and your lawful basis under UK GDPR
- someone checks outputs, especially in the early weeks
- staff affected by the change know what is happening and why
Keep the first version small. One process, one team, one clear measure is far easier to judge than a broad rollout.
What happens in the review in weeks 11 to 13?
At the end of the quarter, measure the same things you recorded at the start and decide what happens next. Compare the result with what you expected, using the low, central and high ranges from your AI business case.
For each piece of work, make one of three decisions:
- Continue, and extend it to more volume or another team
- Adjust, because it works in part and needs changes
- Stop, because the benefit does not justify the cost or risk
A stopped project is not a failure if it was small, cheap and measured. It is information.
What should a 90-day AI roadmap include?
A 90-day AI roadmap fits on a page or two and shows:
- the one or two pieces of work chosen, with an owner for each
- the baseline and the measure of success for each
- the main milestones by week
- the budget, including internal staff time
- the key risks and how they are handled
- the ideas parked for later, with the reason
If your board cannot read it in five minutes, it is too long.
What are the common AI planning mistakes?
The common AI planning mistakes are all avoidable:
- Starting with a tool a supplier has demonstrated, then looking for a use for it
- Setting no baseline, so nobody can say later whether it worked
- Giving ownership to a committee rather than a named person
- Ignoring the time staff will spend testing and checking outputs
- Planning the next quarter before reviewing this one
Where Forwardcycle fits
Our AI Opportunity Diagnostic produces a process inventory, scored opportunities, a value model and a 90-day roadmap. Through our commercial leadership service, Keith Hamilton can then own that roadmap and report progress to your board. To see where your business stands first, take the free AI Readiness Assessment.