A 90-Day AI Implementation Roadmap for UK Businesses

Ninety days is a planning window, not a delivery guarantee. A useful AI implementation roadmap tells you what must be true before you move forward. It should also make stopping a weak project an acceptable outcome.

Start with one workflow. Trying to launch a chatbot, a calling system, a reporting platform and a marketing campaign together makes it harder to establish which change helped and which introduced a problem.

Before the clock starts: name the problem

Write a short brief: the recurring task, who does it, where information comes from, what goes wrong and who owns the result. Include examples of exceptions. A process that works only for the easy cases is not ready to take over live work.

Choose a baseline you can actually collect. For enquiries, record first response time and whether the enquiry reached an accountable person. For reporting, record reconciliation effort and disputed figures. Avoid treating software activity as a business result.

Phase one: scope and prepare

Use the first part of the planning window to establish the current process and agree the smallest useful change. The exact duration depends on access, data quality and the people involved.

Get exclusions into the proposal. If the pilot drafts a response but does not send it, say so. If a booking needs staff approval, keep that boundary visible in the interface and the acceptance criteria.

Phase two: test a controlled pilot

Run representative cases before wider access. Test missing inputs, duplicate events, ambiguous requests and disconnected services. Agree which actions require human approval and how a user can reach a person.

Keep a simple test record: input, expected result, actual result, destination checked, reviewer and any defect. An attractive demonstration is not enough. The underlying record must match what the user was told.

For a call-handling pilot, a message taken and a confirmed booking are different outcomes. For a reporting pilot, a missing figure and a true zero must remain distinguishable. These details matter more than whether the output sounds fluent.

Review errors as well as successes. If staff spend more time correcting output than the workflow releases, investigate before expanding. Do not hide review work outside the cost calculation.

Phase three: decide, document and hand over

Compare the pilot with its baseline using the same definitions. Record external changes that could affect the comparison: staffing, demand, seasonality or a simultaneous campaign.

There are three legitimate decisions: expand, revise or stop. Expansion needs evidence that the process works reliably enough for the proposed use, plus the capacity to handle exceptions. A calendar milestone alone is not an acceptance test.

Before handover, the owner should be able to pause the workflow, find failed tasks, change approved content and reach support. Record the software costs, access arrangements and responsibilities for future changes.

What belongs in the budget?

Separate discovery and implementation from software, usage, maintenance and staff review. Ask which costs vary with activity and what is excluded. Do not use a generic project range as a substitute for a scoped quote.

If the benefit is released time, describe the work that time will support. If the benefit is more sales, measure collected revenue and profit after delivery costs. Neither benefit is guaranteed by installing the system.

What if the pilot is not ready by day 90?

Identify the cause. Missing access needs an owner. Unreliable source data needs correction. A poor business case may warrant stopping. Extend the plan only when the remaining work and its value are clear.

For help choosing the workflow, read AI automation for UK SMEs. For the current service scope, see automation.

Bring the process you want to improve and the evidence you already have. Get unstuck.