A Twelve-Month AI Roadmap for a Business Without a Data Team
Last updated:
Sequence matters more than ambition
Small businesses do not fail at AI because they picked the wrong model. They fail by starting with the most visible, highest-risk application — usually something customer-facing — with no evaluation, no baseline and nobody who can tell whether it is working.
The roadmap below is deliberately slow at the start and fast later. It is designed so that being wrong in month three costs a few thousand pounds rather than a customer relationship.
Quarter one: tools, policy, and one choice
- Give people an approved tool. Staff are already using AI; the only question is whether it is a tool you chose. This costs little and removes most shadow usage.
- Write a one-page policy. What may be pasted in, what may not, and who to ask. One page that gets read beats twelve that do not.
- Pick one internal use case using volume, error tolerance, a clear definition of correct and existing documentation.
- Take a baseline. Time and error rate for the current process, measured for two weeks.
Total spend this quarter is modest — licences and a few days of attention. What it buys is the ability to judge everything that follows.
Quarter two: build one thing properly
One use case, built with evaluation and a review interface, in eight to twelve weeks. Internal, not customer-facing. Typical budget £15,000–£40,000 depending on scope.
Insist on three deliverables beyond the feature itself: an evaluation set with a measured accuracy figure, a way for a person to correct output, and documentation good enough for someone else to maintain it.
Quarter three: measure, then decide honestly
Re-measure the baseline metrics. Compare properly. Then make one of three decisions, and make it explicitly rather than by drift.
- Working. Widen it to adjacent processes.
- Partly working. One more iteration with a defined threshold and a date.
- Not working. Stop. A stopped project that cost one quarter is a good outcome compared with one that limps for two years.
The willingness to stop is what separates businesses that get value from AI from those that accumulate half-finished pilots.
Quarter four: outward, if you have earned it
Only now consider something customers touch — support, onboarding, self-service. By this point you have an evaluation habit, a correction interface, a policy, and staff who know what the technology is and is not good at.
Launch it to a slice of traffic, watch re-contact and satisfaction rather than deflection, and widen by evidence.
What this costs across a year
| Quarter | Spend | What you get |
|---|---|---|
| Q1 | £1,000–£4,000 | Tools, policy, a chosen use case, a baseline |
| Q2 | £15,000–£40,000 | One production internal system with evaluation |
| Q3 | £0–£10,000 | Measurement, iteration or a clean stop |
| Q4 | £20,000–£50,000 | A customer-facing feature, if justified |
A business that follows this ends the year with one or two systems it trusts and a clear-eyed view of where AI helps. That beats five pilots and a vague sense of disappointment.
Frequently asked questions
Do we need to hire an AI specialist?
Is it too late to start?
What if our competitors are moving faster?
Can we skip the internal phase and go straight to customer-facing?
Want the roadmap fitted to your business?
Tell us what your team does day to day. We will suggest the one internal use case worth starting with, and what a realistic first phase would cost.