This is a point more product teams should quantify early. We often optimize prompts for quality but forget that response length is a product decision with a direct impact on unit economics.
One pattern that's worked well is designing for progressive disclosure: give the concise answer first, then let users expand for more detail only when they need it. It improves UX while avoiding unnecessary output tokens on every request.
The same thinking also makes model routing more effective. If a feature consistently needs long-form generation, reducing output length by even 15–20% can have a bigger cost impact than shaving a few input tokens, especially as usage scales.
This is a point more product teams should quantify early. We often optimize prompts for quality but forget that response length is a product decision with a direct impact on unit economics.
One pattern that's worked well is designing for progressive disclosure: give the concise answer first, then let users expand for more detail only when they need it. It improves UX while avoiding unnecessary output tokens on every request.
The same thinking also makes model routing more effective. If a feature consistently needs long-form generation, reducing output length by even 15–20% can have a bigger cost impact than shaving a few input tokens, especially as usage scales.