Relevance score

A signal feed with no ranking is a chronological list, and a chronological list is worked from the top until the person gets bored. Scoring changes the question from what arrived most recently to what deserves the next twenty minutes — which is the only question that matters when there are more signals than hours.

That is all a score should be asked to do. The moment it is presented as a likelihood to buy, people start treating a number derived from a handful of public attributes as a forecast, and the first few misses discredit the whole feed.

Three properties, and they are all about the reader rather than the maths:

• It is reconstructible. Someone can look at a signal and see why it scored what it did. A score nobody can explain is a score nobody acts on.
• It is comparable. Two signals of different types produce numbers on the same scale, or the feed cannot be sorted at all.
• It is filterable. A floor — show me nothing under this — is what turns a score into a working queue rather than a decoration on a card.

Lead scoring ranks people by fit and engagement, accumulates over time, and belongs to the CRM. Signal scoring ranks events by how much attention they deserve today, and it should decay fast. Conflating them produces a number that means one thing on Monday and another on Friday.

Kept separate, they compose well: signals decide which accounts to work this week, lead scoring decides who inside them to write to and how urgently to follow up.

Every signal in the SalesShift pool carries a relevance score. The feed sorts by it by default, and a minimum-score filter is part of the query — so the same pool can be worked as a wide sweep or as a short high-signal queue without maintaining two lists.