AI bills have a way of creeping up before anyone notices. More users, more requests, more tokens, and suddenly a tool that was supposed to save money is eating into margin instead.
Martech contributor Steve Petersen argues the fix is not a bigger budget. It is tighter process: better prompts, reusable frameworks, prompt logs, and spending guardrails.
The FOCUS Framework
The core prompt structure Petersen highlights is called FOCUS, which stands for function, outcome, context, usage, and specific. The idea is straightforward: prompts built around these five components produce outputs that are actually usable in a work context, not just technically correct responses that need heavy editing before anyone can act on them.
Prompt Libraries as Institutional Memory
One of the more practical angles in the piece is the prompt library concept. Teams that document what works create a shared resource with a second benefit most operators overlook: new team members can use the library to get up to speed on what actually produces results, rather than burning tokens on trial and error.
This is the kind of compounding infrastructure that solo operators and small teams rarely build, and then regret not having when they scale or hand off work to a contractor.
The Operator Takeaway
If your AI spend is rising faster than the value you’re getting out, the problem is usually upstream of the tool. A model running on a vague prompt is an expensive way to get mediocre output. A model running on a well scoped FOCUS prompt, inside a team that logs and reuses what works, is a different equation.
Guardrails on usage (who can run what, how often, at what cost per task) complete the picture. The teams spending the least per useful output are not the ones using cheaper models. They are the ones who treat prompts as reusable assets rather than throwaway inputs.
