Business question

How do AI systems evaluate and recommend vendors?

In stages. An AI first forms an opening list for a broad category question, then narrows it as the buyer adds requirements: cutting companies that do not visibly meet each new requirement, moving the leader when a requirement exposes it, and settling on the companies that survive the full sequence. The final recommendation reflects all the requirements added along the way, not the opening question.


The problem

Treated as a black box, AI recommendation looks random. Seen as a staged process, it becomes something you can place yourself inside: where you enter, where you are removed, and where you survive.

Why it matters

Understanding the stages is what makes the other questions answerable. Each stage is a place where a company can be won or lost, and we study each one on its own.

What we can conclude

We have observed this staged narrowing across several categories: an opening list that is not the final answer, companies cut at specific requirements, the leader exposed at the first requirement it cannot meet, and fields that survive, collapse, or move to a different tier depending on the category. Every observation is held at its honest confidence level, and we have now seen the pattern on both ChatGPT and Google AI.

What remains unknown

How closely this resembles the way a human buying committee evaluates, and how much of it holds across different AI systems and over time, are open questions the research program is built to keep testing.


How Upstream Zero helps

From the evidence to your answer.

Upstream Zero studies each stage and turns that into a diagnosis for your company: where you enter, where you are cut, and what must become true to survive. The research is public; the diagnosis is the product.

See the research, or how the work is done in the methodology.