You will always have more candidates than engineering weeks. The skill is killing most of them before anyone writes code.
The two axes
Customer pull: how many existing customers and live deals asked for this connection by name. Distribution upside: whether the partner has a marketplace, a co-sell motion, or a field that will actually mention you.
High pull, high distribution: build first. High pull, low distribution: build for retention, do not expect leads. Low pull, high distribution: a bet. Validate with prospects before you commit a quarter. Low pull, low distribution: skip. A nice logo is not a reason.
Build cost is the third filter
A partner with a sandbox, clear docs, and a published review checklist is cheaper than a partner who answers API questions in a private Slack three weeks later. Cheap is not the same as right, but two equal scores should lose to the cheaper build.
The “one loud customer” trap
One prospect saying they will not sign without Salesforce is a deal problem. It is not automatically your first integration. Check whether five others said the same thing, and whether Salesforce will actually list you. Sometimes the right move is a CSV export and a honest timeline, not a surprise AppExchange project.
Write the kill list out loud
Share the scored list with product and the CEO. The political version of this job is saying no to the CEO’s conference intro. Do it with the matrix, not with vibes.
If you cannot point at a killed partner this month, you are not prioritizing. You are collecting.