Scoring Model That Rewards the Wrong Buyers
A cybersecurity vendor has a single lead score field combining demographic and behavioral points. Current rules: +10 for 'Job Title contains Director', +5 for 'Company Size > 500', +3 per email open (capped at 15), +20 for pricing page visit, +25 for demo request. The MQL threshold is 100. Sales complains that most MQLs are students and individual researchers who open many emails but never buy, while actual enterprise buyers rarely reach 100. Analytics shows email opens are inflated by an image-proxy prefetch that fires on delivery. The vendor has no separate demographic vs. behavioral scores and no negative scoring.
1. Explain why the current model over-produces poor MQLs and under-produces good ones, referencing the specific point values.
Read the worked answer
Two structural flaws:
- Behavioral points dominate and are gameable. Opens cap at 15, but a pricing visit (+20) plus a demo request (+25) plus opens can push a low-fit student past 100. Demographics add at most 15 (+10 title, +5 size), so fit barely matters. A student who opens 5 emails and visits pricing hits 35 from behavior alone with zero fit points.
- Opens are inflated by prefetch. The image-proxy fires on delivery, so many 'opens' are not human. This inflates behavioral scores for everyone, but disproportionately for high-volume low-intent contacts, pushing them over 100.
Meanwhile enterprise buyers may have titles that don't match 'Director' exactly (e.g., 'Head of Security') and company-size values that don't match '> 500' cleanly, so they earn few demographic points and rely on behavior they may not exhibit at high volume. The single-field model can't separate fit from intent, so the threshold can't be tuned without breaking one group.
2. Propose a restructured scoring approach and the threshold logic, including how to handle the prefetch inflation.
Read the worked answer
Split into demographic (fit) and behavioral (intent) scores, then combine for MQL:
- Fit score: title/role matching (broaden to role families, not exact strings), company size bands, industry. Negative points for student/individual domains.
- Intent score: meaningful actions weighted heavily (demo, pricing, content downloads), opens down-weighted or removed given prefetch inflation.
- Prefetch fix: stop scoring opens, or require a click or a second open before awarding points. This removes the inflation at the source.
- Threshold logic: MQL = fit ≥ X AND intent ≥ Y (two-dimensional gate), rather than a single sum. This prevents a high-intent low-fit student and a high-fit low-intent researcher from both crossing a single line.
- Add negative scoring for disqualifying signals (student email domain, competitor domain).
Justify each change against the observed symptom: students over-produced → negative fit + two-dimensional gate; enterprise under-produced → broaden title/size matching + gate on fit.