What evidence should an AI show before recommending who to contact?

One design problem we keep running into while building Netlio: a useful people recommendation needs to explain itself.

Suppose a founder is deciding between seat-based and usage-based pricing. Returning a list of product leaders is easy.

A more useful result would identify one person and show:

- they led the same pricing transition recently

- they were close enough to the decision to know what actually broke

- there is a real relationship path

- the timing makes the conversation relevant now

Then the product should suggest a narrow question, not a generic networking message:

“What created more customer pushback: the pricing metric or the migration?”

The part I’m still testing is how much evidence to expose. Too little feels like a black box. Too much makes the result harder to scan.

If you were using an AI agent for this, what would you need to see before trusting the recommendation?