Signal, Not Noise: An Account-Scoring Engine That Actually Works

Most account-scoring models are confidence theater — a spreadsheet of weights that feels rigorous and predicts nothing. The good ones are different in a specific, learnable way.

Why most scoring models fail

• Weights chosen by intuition. Someone decides "firmographic fit is 40%" because 40 felt right. It wasn't tested.
• No feedback loop. The model is never measured against outcomes, so it never improves.
• Too many signals. Twenty inputs, most noise, produce a score that means nothing.

The model that works

A useful account score is built backwards from an outcome:

1. Start from what you're predicting. Won deals. Churned accounts. Expansion. Pick one.
2. Find the signals that actually correlate. Not the signals you think matter — the ones your data says matter. Look at your won accounts and find what they shared before they bought.
3. Cut ruthlessly. A model with five strong signals beats one with twenty weak ones. Every signal you add dilutes the ones that count.
4. Score against the outcome, not in isolation. A 90 means "looks like accounts that won," not "passed a vibe check."
5. Close the loop. Track which scored accounts actually closed. Re-weight. Repeat quarterly.

The signals worth testing

• Engagement depth — not opens, but meaningful actions (a demo, a reply, a shared doc).
• Relationship proximity — is there a warm path in?
• Trigger events — hiring, funding, leadership change, product launch.
• Stack signals — what tools they use and the gaps that imply.

The principle

A scoring engine works when it's an empirical model that updates itself, not a one-time opinion. Signal, not noise — and the only way to know the difference is to measure against the outcome you care about.