How much should an SME budget for AI adoption?
Budget for training, change, integration and management time, not just licences, which are usually the smallest cost. Phase spend as returns are proven.
Why licences mislead the budget
Licence fees are easy to see, compare and approve, so they tend to dominate AI budget discussions. They are usually the smallest part of the real cost. The larger costs sit in people's time and in the changes needed to make AI part of normal work. These are the costs most often left out.
A budget built only on licences looks affordable at first. It then overruns quietly through unplanned training, consultancy and management effort.
The cost components to include
Licences and subscriptions: Per-user fees for AI tools or for add-ons to software you already use. Check whether pricing rises with usage and whether every user needs the same tier.
Training: Structured learning for staff and managers, plus the temporary productivity dip while people adjust. In the first year this is often larger than the licence cost.
Change management: Time spent redesigning processes, updating documentation, communicating with staff and handling concerns about jobs or workload.
Integration and setup: Connecting AI tools to your systems and data, preparing documents and templates, and any external technical support.
Governance: Writing and maintaining policies, reviewing data handling and checking supplier terms. Regulated businesses should expect this to take more time.
Management time: Leaders and champions overseeing pilots, reviewing outputs and making decisions. It rarely appears as a budget line, but it is a real cost.
Ongoing review: Human checking of AI outputs continues after rollout, so include it in running costs and do not treat it as a one-off.
Phase your spend
Committing a full year's budget on day one concentrates risk. Phasing spreads it and lets evidence shape later decisions:
Phase 1, discovery and pilot: A small spend on a limited number of licences, focused training for a pilot group, and time to set a baseline and basic guardrails.
Phase 2, targeted rollout: Extend proven use cases to the teams that benefit, with proper training and process updates.
Phase 3, wider adoption: Broaden use, consider integration or automation, and invest in deeper capability where results justify it.
Set a review point between each phase. Move forward only when the previous phase has shown results against agreed measures. For guidance on choosing between off-the-shelf tools and custom work at each stage, see whether to build, buy or wait on AI.
Fund later phases from proven returns
The most defensible approach is to let early returns pay for what comes next. Suppose a pilot frees capacity that goes to billable work, or it reduces rework. That value can make the case for the next phase. Spending stays tied to evidence, and finance gets a clear story behind each request.
This only works if you measure credibly from the start, with a baseline and a full count of costs. How to measure the ROI of AI explains how to build that picture.
What the right figure depends on
No standard percentage or per-head figure fits every SME. Your budget depends on:
Headcount.
How many roles involve information-heavy work.
Your current systems.
Your regulatory exposure.
How much internal capacity you have to lead change.
A small team with simple systems will spend mostly on training and time. A business with complex systems or sensitive data will spend more on integration and governance.
This is general guidance. Take advice from your accountant on how AI spending should be budgeted, capitalised or claimed in your own circumstances.
More questions