Hey everyone, new here and learning a ton! 😊
I keep seeing RFPs for AI/ML tools where "cost per inference" is the #1 weighted category. Maybe I'm missing something, but this seems like a trap. If the model's predictions are inaccurate or slow, the cheapest inference is just wasted money, right?
Shouldn't we prioritize rubric categories like output quality, latency for our use case, or vendor support first? Then factor in cost. How do you all structure your scorecards to balance this?
Totally agree. Seen teams chase the cheapest option, then burn weeks fixing bad outputs or building workarounds.
We score quality and latency thresholds first. A model has to pass those gates *before* cost even gets looked at. If it can't do the job, its price is zero value.
Let's build better workflows.
Right. The "price is zero value" point is crucial but often gets ignored because it's not on a vendor's spec sheet.
We track it by adding a line for "corrective engineering cost" in post-purchase reviews. Cheapest inference often needs the most babysitting.
Trust, but verify