Knowing who is on your site is no longer the hard part. The hard part is turning that identity into a score your sales team trusts and that reliably predicts revenue. Most website visitor lead-scoring breaks here and drifts away from real pipeline value.
Post-identification scoring is everything that happens after you resolve a visitor to a person or account. Your intent model is how you estimate their likelihood to buy. The lead score is the number your sales and marketing teams actually act on.
Build a Post-Identification Baseline That Matches Reality
Before you worry about decay, drift, and retraining, you need a baseline model that reflects how your buyers actually behave once they are identified. That baseline starts with three input groups: identity (match rate by segment), behavior (what they do on-site), and context (where, when, and how they visit).
Many teams see 10–25% match rates on cold, first-touch traffic and 60–80% on known audiences like email or retargeting. Track match rate by channel so you know which traffic your model can truly score.
Behavior should stay simple at the start: session count and repeat visits, high-intent paths like pricing or demo, depth (scroll percent or time on key pages), and recent activity, not just lifetime history.
A practical way to build a baseline: start with a clear target (qualified opportunity or closed-won), then use a straightforward model like logistic regression or a compact tree model with six to 10 features. You might find accounts where at least two visitors view pricing, then a customer proof page, then return within seven days convert to qualified opportunity at 18%, versus a 4–6% median.
Stop Score Decay From Quietly Killing Good Leads
Even the best score gets old. Intent cools, priorities shift, teams change. Score decay is the math you apply so a great visit yesterday doesn't look the same as that same visit from last month.
You want decay rules that follow your actual sales motion:
- High-intent behaviors like demo, pricing, or deep product docs hold value longer
- Light behaviors like a single blog visit fade quickly
- Enterprise cycles usually need longer decay windows than fast SMB cycles
A simple framework: use a longer half-life for strong intent events (14–30 days), a short half-life for weak signals (3–5 days), and recalculate decay parameters on a regular schedule using recent closed deals.
When a visitor views pricing twice in two days and gets an SDR touch within one day, B2B SaaS teams often book meetings on 35–40% of those accounts. When that same pattern sits untouched for 5–7 days, meeting rates fall toward 10–15%. Front-load the score, then let it slide down quickly over the next week.
Time Retraining to Your Sales Cycle and Drift to Reality
Retraining cadence is how often you refresh or rebuild your intent model with recent data. A full retrain re-learns the whole model; a parameter refresh keeps the structure but updates weights, thresholds, and score bands.
A practical rhythm many teams follow: monthly to refresh calibration and check score bands, quarterly to retrain on the last 2–4 quarters and test new features, and yearly to revisit what "qualified" means with sales and RevOps.
Alongside retraining, watch for model drift. Data drift happens when traffic mix changes; concept drift happens when the core buying pattern changes. Simple drift checks: track conversion from each score band to opportunity and closed-won (watch for >20–30% relative change over a month), watch for sudden drops in top-band performance, and mark launch dates so you can compare before and after.
Turn SDR and CRM Feedback Into a Closed Loop
The fastest drift detector on your team is not a dashboard. It is the SDR who lives in the queue every day. Wire their feedback directly into your scoring and retraining: create clear disposition codes (no budget, no authority, wrong persona, timing off), make sure reps tag every high-score contact they touch, and feed those codes back into your model training data.
If many high-score leads get tagged "no authority," that pattern might really be a researcher track. If a big share of high-score accounts are already customers or partners, you probably need a separate upsell or partner score and cleaner exclusion rules.
Run a Quarterly Scoring Health Check
All of this works best as a simple habit, not a one-time project. A straightforward rhythm:
- Monthly: glance at score distributions, band conversion, and a sample of SDR notes
- Quarterly: retrain, recalibrate thresholds, adjust decay, rerun drift checks
- Annually: reset definitions of "qualified" with sales and marketing leaders
Teams that run this kind of health check often see steady gains. Tightening thresholds and retraining quarterly might move meeting rates on "high" leads from 18% to 22% over two quarters, without adding more spend.
Turn Anonymous Traffic Into Qualified Sales Conversations
If you are ready to turn more of your traffic into pipeline, we can help you prioritize the visitors most likely to buy. Our website visitor lead scoring approach shows you which accounts are engaging, how they behave, and when they are sales ready. Book a demo to build a scoring model that aligns with your goals.