Skip to main content

Scoring rules

How the signals turn into a decision about who to talk to, and when.

The trigger rule​

One High signal fires outreach. Two Medium signals stacked at the same account, with a decision-maker present, also fire. A Weak signal never does, no matter how many.

Who fired it​

Technical influence and budget authority are separate things. A "SigNoz vs Datadog" search from a Staff SRE outranks a VP browsing the pricing page. The SRE is the trigger. The VP is the closer.

When two High-tier accounts compete for attention, role convergence breaks the tie. Engineers plus a budget-holder title active at the same account goes before engineers alone. Count tells you adoption. Convergence tells you a decision is forming.

When it fired​

Signals don't decay at one rate. Two families, two directions:

  • Frustration signals (a paid ceiling hit, an error on the migration doc) go stale in about a week. Yesterday's is already cooler than today's.
  • Deadline signals (a renewal date, a migration in progress) get more urgent as the date gets closer.

So the queue isn't first-in-first-out. See Sequence spec for how that plays out in Slack.

No weights on day one​

I wouldn't ship a weighted score at the start. Giving an economic buyer 3× the points without any conversion data is a guess wearing a model's clothes.

Phase one is instrumentation. Capture the signals. Tag every closed-won and closed-lost with which signals fired and when. Phase two lets that data set the weights.

Sequencing​

Start with the signals that need zero engineering: docs search logs, GitHub activity, deanonymised traffic, job changes through Clay. Prove the motion inside a month. Asking engineering for new product events comes after the first win, not before.

The score is a sort order​

Never show a bare number. Show the number with the signals that produced it:

82   VP Eng hit the Datadog migration doc twice this week
     three engineers from the same domain active
     retention at 80% of limit

A senior rep can audit that in five seconds and overrule it. A newer rep can follow it. Same system serves both.

Why scoring systems die​

Not because the maths is wrong. Because experienced reps don't trust a black box and quietly stop opening it. Someone with ten years in the job trusts their own read over a tool's first pass. Someone in their first year treats the tool as truth. The model has to serve both without pretending to replace the first.

Overrides are the best training data​

When a rep ignores a low score and wins the deal anyway, that disagreement is worth more than a hundred correct predictions. It's the place the model is wrong in a way the data alone won't show you.

Source: signoz_assignment_q3.md, Slide B and appendix. Tiebreak rule added 6 Sep 2026.