A Signal Scoring Framework That Ranks Accounts

Every signal should be evaluated on three axes. Fit asks whether the account matches your ICP at all, because a strong signal from a bad-fit account is still a bad account. Strength asks how predictive the signal type is, where a demo request outweighs a topic surge by an order of magnitude. Recency asks how fresh it is, because signals decay, some in days.

Score fit at the account level as a gate or multiplier, and score strength and recency per signal. The final account score is fit times the decayed sum of its signals. Keeping the axes separate makes the model debuggable: you can always answer why an account ranks where it does.

Start with a simple tiering rather than false precision. High tier: hand-raises, pricing page visits, product usage triggers, and multi-person engagement. Mid tier: content engagement on high-intent topics, relevant hiring, and funding plus a follow-on signal. Low tier: third-party topic surges, single blog visits, and social follows.

Then let your own data adjust the tiers. Look back at closed-won accounts and ask which signals appeared in the ninety days before the opportunity was created, and at what rate those signals appear in accounts that never convert. A signal common in both populations deserves a lower weight than intuition suggests.

Apply time decay so scores fall automatically without fresh activity. Fast-decay signals include site visits and usage spikes, which lose most of their meaning within a couple of weeks, while slower-decay signals include hiring and stack changes, which stay relevant for a quarter. Without decay, your ranked list becomes a museum of accounts that were interesting once.

Score combinations above sums. Two different signal types from one account inside a short window, like a hiring signal plus a pricing visit, is stronger evidence than double the score of either alone, because independent sources corroborating each other is exactly what an active buying process looks like. A simple multi-signal bonus captures this without complicated math.

A score is useless until it maps to behavior. Define action tiers: above the top threshold, route to a rep within a day with the signal history attached; in the middle band, add to targeted marketing and light-touch sequences; below that, keep accumulating silently. Publish the thresholds so sales and marketing argue about the model, not about individual accounts.

Review the model monthly with one question: are top-tier accounts converting to meetings and pipeline at meaningfully higher rates than mid-tier ones? If not, the weights are wrong or the signals are weak, and the honest move is to simplify back to the few signals that demonstrably predict revenue for your business.