Don't measure how much AI you use. Measure what it brought you.
Number of licenses, number of prompts, credits used. Companies measure how much AI they use and can't say what it brought them. We explain why this is the same mistake as measuring a salesperson by the number of calls.
Activity is not a result
A salesperson can hit their call target every week. But that number won't tell you whether they will hit the quarterly plan. The number of calls shows activity, not results.
The same happens to companies with AI. In the first year, companies mostly measured adoption: how many people use AI and how many tokens get used. A token is the basic unit in which AI processes text and by which it is billed. Usage became the measure of both engagement and cost.
Some teams even pushed for the highest possible usage, believing more use means more value. This is called “tokenmaxxing”.
With a person
When you reward the number of calls, you get lots of calls. Not necessarily more contracts.
With AI
When you track usage, you get high usage. People learn to hit the number, not to deliver value.
The opposite extreme is the same mistake
When the first higher AI bills arrive, some companies react the other way: they limit use, ban newer models, cut instructions. Saving on usage is the same trap as chasing it. Both approaches measure tokens, neither measures results.
A typical example: a team saves 500 tokens by cutting important context from an instruction. AI then produces something that works on its own but fails in the bigger picture. Fixing it costs 5,000 tokens. The cost doesn't disappear, it just moves.
“A marketing agency shortens its AI instructions to save credits. The client's brand rules disappear from them. The copy is cheaper to produce, but a copywriter has to rewrite every piece. The savings on credits turned into hours of work.”
What to measure instead of usage
Instead of asking “how much AI do we use”, ask the questions you would ask about any other investment:
- How many hours of work did we save, and in which area?
- How many fixes and errors went away?
- How many inquiries, invoices or quotes did we handle faster?
- How much shorter is our response time to clients?
Only once you know the result can you judge whether usage is reasonable. High usage with a big benefit is a good investment. Low usage with no benefit is money thrown away, just less of it.
Case study: Lead generation“For our sales system, we don't track how much AI it uses. We track how many companies it checks and how many outreach emails it sends. In two weeks that was about 509 companies and 173 emails. That is a result, not usage.”
Value comes from the system around the model, not the model
Access to good models is no longer an advantage. Everyone has it for the price of an ordinary monthly subscription. The difference is what surrounds the model: what material AI gets, how it is built into the work, who checks the result and how it is tracked.
The analyst firm IDC estimates that by 2028, 70% of large AI deployments will work with several models at once. For a smaller company, one thing matters. Don't tie yourself to one model, tie yourself to the result.
Who is responsible for results
People who use AI should learn to use it efficiently. Not less, but better: a clear brief, a plan before execution, only the material needed.
Management needs to see costs and results together. Reward benefit, not volume.
Tools should help: budgets and limits, an overview of usage and linking usage to a specific area of work.
Ask yourself
Can you say how much one completed task costs with your AI?
