Insights
What should AI cost my business in year one?
Far less than a transformation, and far more than the licence fee. The gap between those two numbers is where most AI budgets go wrong. The per-seat price is what everyone quotes, because it is the number a vendor puts in front of you. It is also the smallest line in the true cost of getting value from AI in the first year. Budget only for the licences and you will either underspend and get no adoption, or overspend on seats nobody uses.
The cost structure below covers the whole thing, so you can budget for it rather than for the visible tip.
The licence is the cheap part
A business AI plan runs roughly the price of another productivity subscription per person. That is real money at scale, but it is not the expensive part, and treating it as the budget is the classic mistake. The tool makes value possible. Capturing that value takes the work around it: enabling your people, wiring it into your systems, and governing how it is used. Fund the licence alone and you have bought the possibility and none of the delivery.
The four cost lines
Budget across four lines, not one.
| Cost line | What it covers | Rough weight in year one |
|---|---|---|
| Tools & licences | Per-seat plans, API usage, any platform fees | The smallest line, and the only one most budgets include |
| Enablement & change | Training, adoption support, redesigning how work is done | Usually the largest line, and the most forgotten |
| Integration & data | Connecting AI to your systems, readying the data it draws on | Varies most: trivial for off-the-shelf use, significant for grounding AI in your own data |
| Governance & risk | Policy, EU AI Act work, security review, oversight | Modest but non-optional; skipping it is a false economy |
The exact split depends on what you are doing, but the shape holds: the line you can see (licences) is rarely the line that costs the most.
The hidden line: change and adoption
The single most under-budgeted cost is getting people to actually use the thing. A tool that is bought and not adopted returns nothing, and adoption does not happen for free. It takes training, support, and the redesign of how the work is done around what the tool now makes possible. Firms that budget for licences and skip enablement reliably reach the worst outcome: full cost and no return, plus a verdict that "AI didn't work for us" when what didn't work was the budget.
If you cut one line, do not let it be this one.
Budget for outcomes, not seats
The temptation is to buy licences for everyone and hope value follows. It rarely does, and you end up paying for hundreds of seats with a handful of active users. The better shape is to fund a narrow, high-value use properly, all four cost lines included, prove it pays back, then let the returns fund the next expansion. That turns AI from a fixed cost you are hoping to justify into an investment that earns its own scale-up, which is a far easier number to defend.
A sensible year-one shape
For most mid-market companies, a defensible first year looks less like a platform rollout and more like a focused programme: a modest licence spend, a larger enablement and change spend, whatever integration your chosen use genuinely needs, and a small but real governance line, all aimed at one or two uses that pay back rather than a wall-to-wall deployment. Start there, measure it, and let the evidence set the second-year budget.
The headline for the board: the useful question is what it costs to make one valuable use actually pay back, and whether the budget covers all four lines of that or only the one on the invoice.
Get the year-one number right, across all four cost lines and pointed at a use that pays back. That costing sits inside the board-ready roadmap Firestarter produces in six weeks.