Key takeaways
- Budget transformation in stages, because each stage has a different cost shape.
- Stage 1 is mostly seats and a one-off setup. It is predictable.
- Stage 2 adds workflow redesign and usage-based costs. This is where caps matter.
- Internal time is the line nobody budgets and often the largest real cost.
Ask a large consultancy what AI transformation costs and you will get a number with a lot of zeros and a long timeline. For a business of 20 to 200 people, the honest answer is much smaller and much more predictable, provided you budget it in stages and cap the parts that can move.
These are planning ranges, not quotes. Seat prices assume mainstream business-tier assistants at around 20 to 30 US dollars per user per month on annual billing, checked October 2026.
The cost lines, by stage
| Line | Stage 1: access | Stage 2: workflows | Stage 3: organisation |
|---|---|---|---|
| Seats | Main cost | Some premium seats added | Seats shift with roles |
| Setup and configuration | One-off fee or internal time | Small additions | Small |
| Workflow redesign | None | Main cost | Ongoing |
| Usage (API, agents, automations) | Low | Grows, needs caps | Can become significant |
| Internal time | Moderate | High | Ongoing |
Worked examples, first year
| 30 staff | 80 staff | 180 staff | |
|---|---|---|---|
| Seats (weekly users) | 25 seats, about 6,000 to 9,000 USD | 60 seats, about 14,000 to 22,000 USD | 130 seats, about 31,000 to 47,000 USD |
| Setup | Fixed fee in the low thousands | Fixed fee in the low to mid thousands | Mid thousands plus identity work |
| Three to five workflows redesigned | A few thousand, or internal | 5,000 to 20,000 USD | 15,000 to 50,000 USD |
| Usage allowance | Under 100 USD a month | A few hundred a month | Several hundred to a few thousand a month, capped |
| Internal time | An owner at a few hours a week | An owner at a day a week | A part-time role |
Three things stand out. Seat costs scale with users and are easy to forecast. Workflow redesign is where the range is widest, because it depends on how many workflows you take on and whether anything custom is built. Usage starts small and is the line that can surprise you, which is why it should be capped from day one.
Where the money goes wrong
Paying for seats nobody uses
Buying a seat for every employee on day one, when only two thirds will use it weekly, wastes a third of the largest line. Start with likely users and add on request. Review every quarter.
Building before configuring
Commissioning a custom tool before staff have even used a well-configured off-the-shelf assistant is the most expensive mistake in the category. See build or buy AI.
Uncapped agents
An automation or agent with no ceiling can generate a month's budget in a weekend. See AI agent spending limits.
A long, slow programme
Internal time compounds. A rollout done in 30 days costs a fraction of the staff hours of one that meanders for two quarters.
How to justify the spend
- Pick five named tasks across the business and time them now.
- Do the stage 1 rollout and time them again after 30 days.
- Multiply the hours saved by a loaded hourly cost.
- Compare with the seat cost. At normal salary levels, an hour a week per user usually covers it many times over.
- Use the result to choose which workflows to fund in stage 2.
We cover the measurement in more depth in how to measure AI ROI. To model your own numbers, use the AI cost calculator.
Frequently asked questions
How much should a small business budget for AI?
For the first year, a business of 30 people can expect roughly 7,000 to 12,000 US dollars in seats, a one-off setup fee in the low thousands, and a modest usage allowance. Workflow redesign and automation add more, depending on how many workflows you take on.
What is the biggest cost in AI transformation?
For most smaller businesses, seats are the largest cash line but internal time is the largest real cost. A rollout that drags for six months consumes far more staff time than the licences cost.
Is AI transformation worth the cost?
If the first stage gets most staff saving even an hour a week, the seat cost is covered many times over at normal salary levels. The return becomes clearer and larger once a few high-volume workflows are redesigned. Measure time on named tasks before and after.
How do we stop AI costs growing out of control?
Set hard spending limits at the provider for anything usage-based, review seats every quarter and reclaim unused ones, and map routine tasks to cheaper models. Most uncontrolled AI spend comes from usage, not seats.
Get a straight answer for your business
The Clarity Package gives you a written business case and a detailed implementation plan for 500 USD. The Implementation Package does the whole build, live in 30 days, for a fixed 5,000 USD.