Signals in the CRM, with Decay

The play in six steps

1. 01

   Step 1: Capture signals, dated

   Signals are only worth something when they reach the place the rep already works, with a date attached.

   • Job posts, funding, tech stack changes, exec moves, website visits, post engagement.
   • Detect them in Clay or in Claude Code. Either works; where they land is what matters.
   • Every signal carries its date, a one-line summary and a source URL.
     • An undated signal cannot be decayed or checked for freshness, so it is not stored.
2. 02

   Step 2: Store each signal as its own CRM record

   A text field gets overwritten every week. A record keeps the history and the why.

   • Create a Signals custom object in HubSpot or Salesforce, associated with the company.
   • Properties: signal type, signal date, summary, source URL, weight, age in days.
   • The company keeps the rollups reps filter on: signal count, last signal date, priority score.
   • Dedupe on account, signal type and source, so the same event is not scored twice.
3. 03

   Step 3: Score it, and let it decay

   An account with three signals a month ago and nothing since should not still sit at the top.

   • Each signal adds its type's weight: an exec hire or a funding round counts more than one page visit.
   • Each signal's contribution decays exponentially with age, on its own curve per signal type.
     • A storm or an intent spike fades in weeks. A new leader still matters months later.
   • Keep weights and decay rates in a config table, not in code, so they can be tuned without a deploy.
   • The score drifts down every day and jumps back up when something new lands.
4. 04

   Step 4: Set two thresholds per signal

   Ranking an account on a signal and naming that signal in an email are different decisions.

   • Decay window: how long the signal still counts toward ranking the account.
   • Copy window: how long it may be named in an email or on a call. Always tighter than the decay window.
   • Hard hide at around six months: past that the signal is not shown at all.
   • Intent topics may steer which subject gets chosen, but are never stated as a claim. Naming them tells the buyer they were watched.
   • Generators check a yes or no flag per signal before naming it, rather than a prompt rule.
5. 05

   Step 5: Let thresholds move accounts both ways

   Accounts climb and fall on their own as signals arrive and age.

   • Score thresholds set HOT, WARM and COLD on the company record.
   • Reps sort by score: the list re-ranks itself every morning before they log in.
   • An account that cools drops out of the hot view without anyone touching it.
6. 06

   Step 6: Hot account: map the committee and alert the rep

   The rep is told why the account is hot and who to talk to, not just that it is hot.

   • Source the buying committee with AI Ark.
   • Find verified emails with Prospeo and Findymail, verified with BounceBan.
   • Claude writes a short account brief from the live, copy-safe signals.
   • Push contacts and brief to the CRM next to the score, create a task for the owner and alert them in Slack.