Lead scoring can become a proxy for activity when an agency changes the model, tracking, lifecycle rules, or routing. More page views, clicks, or email events may raise a score while sales sees no stronger buying evidence. Diagnose the first break between the score, buyer intent, stage ownership, and commercial outcome.
1. State the decision and change window
Write what may change: score weights, threshold, routing, campaign, agency scope, data source, stage, or budget. Name the agency transition date, audience, offer, sales cycle, owner, capacity, and outcome. Preserve the old model, effective date, configuration, and reports before editing anything.
Do not compare a score from one model with a score from another as if the number had a stable meaning.
2. Define buyer evidence
List observable evidence for problem, fit, authority, urgency, service area, budget, stakeholder involvement, requested next step, and explicit acceptance. Separate evidence supplied by the buyer from passive activity inferred by a platform. A score should not reward repeated visits when the person is researching support, recruiting, or an irrelevant service.
The Salesforce lead implementation guide can prompt ownership, qualification, conversion, and disposition questions. Local definitions and sales evidence govern; a score is not a stage.
3. Reconstruct the score formula
Export rules, weights, thresholds, decay, exclusions, negative signals, account fit, recency, frequency, and data sources. Mark events that changed after the agency transition: page view, email click, webinar attendance, form submit, source, firmographic field, or CRM stage. Check whether a tracking migration doubled activity or removed negative signals.
Use a rule ledger with version, owner, effective date, sample record, and reason for change. Keep unknown and unavailable fields visible instead of assigning a default that raises the score.
4. Compare scored cohorts
Select pre-change and post-change cohorts matched by source, offer, geography, device, sales owner, and maturity. Compare score distribution, buyer evidence, response, acceptance, opportunity, disposition, and capacity. Inspect high-score rejects and low-score accepts; both reveal where the model is misaligned.
Do not judge recent records before the sales cycle matures. A score that predicts a later opportunity may be useful even when early response is slow; a score that only predicts activity is not a buying-intent signal.
5. Check identity and event quality
Trace score inputs to page, form, email, ad, CRM contact, account, and owner. Check duplicate profiles, anonymous-to-known joins, shared devices, bot activity, imported events, consent, deletion, and recency. The GA4 event guidance can structure event names and parameters; it does not validate a score’s commercial meaning.
Record whether the agency changed tags, event names, attribution, or enrichment. A score increase after tracking changes may be a data artifact.
6. Check threshold and routing consequences
Map score threshold to queue, SLA, owner, priority, nurture, rejection, and compensation. Record workable volume, response capacity, territory, and escalation. If the threshold sends activity-rich but unqualified records to sales, a scoring fix alone will not repair the handoff.
Keep a manual review queue for borderline and unknown records. Do not silently lower the threshold to meet a lead target or raise it to hide a backlog.
7. Test alternative explanations
List offer mismatch, audience mix, agency targeting, tracking duplication, data enrichment, seasonality, response delay, CRM stage change, sales capacity, and real demand shift. State which observation distinguishes each. The agency change may coincide with a site release or campaign change rather than cause the score problem.
If the score feeds ad optimization, review the Google Ads offline conversion imports FAQ for qualified/converted definitions, timing, and duplicate questions. It does not validate the local score threshold.
8. Run a scoring review
Select a fixed sample of high-score rejects, low-score accepts, recent open records, mature opportunities, and unknowns. Have sales and marketing independently identify the buyer evidence, fit, urgency, next action, stage, and capacity state. Log disagreements rather than averaging them away. Compare the result with the score version that existed when the record was routed.
If a rule is changed, run it on a historical sample without rewriting production history. Check whether the new threshold would overload a queue, exclude a legitimate segment, or change an advertising conversion signal. Keep a manual review route for borderline cases and a suppression path for privacy or consent changes. Define what evidence would make the model less trusted and who can pause it.
Treat the scoring model as a decision aid. It may prioritize work, but a person or an explicit business rule must own acceptance and rejection.
9. Apply the score gate
| Gate | Required evidence | Hold if | | — | — | — | | evidence | buyer action, fit, urgency, next step | score is activity-only | | formula | version, weights, decay, exclusions | rule changed without history | | identity | profile, account, consent, duplicate | inputs cannot be reproduced | | cohort | matched, mature, pre/post, owner | recent data is treated as final | | routing | threshold, queue, SLA, capacity | sales receives unworkable volume | | outcome | accepted, opportunity, disposition | score is called revenue | | stop | owner, pilot, restore, review | model cannot be reversed |
Choose repair the model, separate activity and intent, change routing, run a bounded pilot, keep manual review, or hold. Preserve the old and new rules, cohort, sample, agency change log, CRM evidence, rejected explanations, owner, and next review date. Keep this diagnostic local and non-indexable until current CRM, analytics, privacy, agency, overlap, and editorial review are complete; it does not guarantee qualified pipeline.
How did this article land?
Choose one reaction. You can change it anytime.
