Building a Lead Score From Real Engagement, Not a Guess

Keep these as two separate numbers rather than merging them into one. Fit score answers whether this prospect matches your ideal customer profile: company size, industry, role, and it barely changes over the life of the lead. Engagement score answers whether they're currently paying attention: opens, clicks, reply sentiment, meeting requests, and it should move up and down within days, not stay flat for months. A high-fit, cold-engagement lead and a lower-fit, hot-engagement lead need different next actions, and combining them into one score hides that difference.

Not every signal deserves equal weight. A reply, even a short one, should weigh far more than an open, since opens are noisy and increasingly unreliable with privacy features that pre-fetch images. A click on a specific link, like a pricing page or case study, is a stronger signal than a generic open because it shows intent about a particular topic. Multiple opens of the same email without a click or reply over several days is a weaker signal than people often assume, since it can just as easily mean someone forwarded the email internally as that the original recipient is warming up.

Engagement from three weeks ago shouldn't count the same as engagement from yesterday. Build a decay curve into the score so recent activity carries more weight and older activity fades, rather than accumulating forever. Without decay, a prospect who engaged heavily once and then went quiet keeps showing up near the top of a rep's list long after they've moved on, which trains reps to distrust the score and go back to working leads by gut feel.

A score is only useful if crossing a threshold changes what happens next. Define what a rep, or an AI SDR, should do differently when a lead's engagement score crosses a specific point: move from a nurture sequence into a direct outreach attempt, alert a rep for same-day follow-up, or flag the lead for a different message track entirely. Leaving the score as a passive number on a dashboard that nobody checks defeats the purpose of making it dynamic in the first place.

Examples of thresholds that change what happens next:

• Move the lead from a nurture sequence into a direct outreach attempt when the engagement score crosses your first threshold.
• Alert a rep for same-day follow-up when the score jumps because of a reply or a click on a pricing page.
• Flag the lead for a different message track when its behavior suggests the current sequence is not the right fit.
• Let the score decay so a lead that goes quiet drops back below the threshold and leaves the rep's task list.
• Write each threshold down with its exact action and owner, so the score never sits as a passive number on a dashboard.

The most common failure is scoring on activity your platform happens to track easily rather than activity that actually correlates with a real conversation. Website visits from an anonymous IP that might not even be the named contact, for instance, get weighted heavily just because the data exists, while a genuinely warm signal like a prospect replying to ask a clarifying question gets the same weight as a one-word out-of-office reply. Audit what the score is actually rewarding every quarter or so, not just when it's first built.

If an AI SDR is running part of your outbound motion, its conversations generate the same kind of engagement signal a human rep's outreach does, a reply, a question, a scheduled call, and that signal belongs in the same scoring model rather than a separate one nobody reconciles. Keep one engagement score per prospect regardless of which channel or which sender, human or AI, produced the activity. Splitting them creates a blind spot where a prospect engaging heavily with an AI SDR's sequence doesn't surface as hot to the human rep who'd take the next call.