Lead Scoring Model
A lead scoring model is a defined system of rules that assigns numeric points to leads based on attributes and behavior, producing a score that ranks readiness to buy.
Key takeaways
- Fit rules read attributes, engagement rules read actions, and the score is their sum.
- Negative points are as necessary as positive ones for filtering out non-buyers.
- Engagement points should decay, or stale leads stay permanently marked as hot.
- Tier thresholds only carry meaning once lead volume is large enough to separate groups.
- A fit-versus-engagement grid preserves the reason behind a score that one number hides.
In depth
A scoring model is a table of rules: each input has a condition, a point value and a category, and a lead's score is the sum of the rules it matches. Fit rules read attributes such as headcount band, industry or job title; engagement rules read actions such as a quiz completion or a pricing visit. Thresholds then cut the continuous score into tiers. Some teams keep fit and engagement as two separate axes and use a grid instead of one number, which preserves the reason behind a score.
Three choices decide whether the model works: which signals are in it, how heavily each is weighted, and where the tier boundaries sit. Adding more signals feels safer but dilutes the strong ones, because a hundred small positives can outvote the single question that actually predicts revenue. Negative points matter as much as positive ones; without them, a student researching a paper accumulates the same score as a buyer. Decay is the third lever: engagement points that never expire turn every old lead permanently hot.
Most teams start by ranking last year's closed-won deals, listing what those buyers had in common, and turning three or four of those traits into rules with round weights. Refinement comes later, from outcome data. A Pivix scorecard makes the model visible to the respondent as well as the seller: each answer carries its points, categories group related questions, and the resulting tier picks the result page and the follow-up. Publishing the logic this way forces the weights to be defensible.
A rule-based model cannot learn what it was never told, so it misses buying patterns nobody thought to encode. It also needs enough volume for tiers to mean anything; with thirty leads a month, the difference between a 68 and a 72 is noise dressed as precision. Long, multi-stakeholder deals break the single-score assumption, because the person filling in the form is rarely the person who signs. And a model calibrated on one market will misrank leads from another.
Example in practice
How to measure it
The core check is separation: group leads by tier and compare the rate at which each tier converts to a meeting and then to a deal. A working model shows a clean staircase, hot above warm above cold. If two adjacent tiers convert at the same rate, the boundary between them is arbitrary and should be merged or moved.
Watch tier distribution over time as a drift alarm: if the hot share creeps from a tenth of leads to a third without a change in traffic quality, the thresholds have gone soft. Per-rule contribution is the other diagnostic. For each rule, compare conversion among leads that matched it against those that did not, and retire rules that show no gap.
Common mistakes
Weights get set in a workshop where the loudest opinion wins, then never checked against a single closed deal. Six months later the hot tier is full of leads sales refuses to call. Rebuild the model backwards instead: pull the last fifty closed-won and closed-lost records, score them with your current rules, and see whether the two groups separate at all. If they do not, the weights are decoration.
The other failure is one score for everything. A single number is asked to answer both whether you should sell to someone and whether they are ready now, so a perfect-fit enterprise that has done nothing scores like an eager one-person shop. Split the two, either as separate fit and intent scores or as a two-by-two grid, and route each quadrant differently: nurture good fit with low intent, disqualify poor fit regardless of activity.
Frequently asked questions
What is the difference between fit and engagement scoring?
Fit scoring measures how well a lead matches your ideal customer profile using firmographic data, while engagement scoring measures their behavior and intent. The best models combine both so you prioritize leads who are both qualified and interested.
How often should a lead scoring model be updated?
Review your model at least quarterly and whenever your product, pricing, or target market shifts. Recalibrate weights against recent closed-won and closed-lost data to keep the score predictive.
Can a quiz act as a lead scoring model?
Yes. A scorecard quiz assigns points to each answer, totals them, and maps the total to a result tier. That makes the quiz a transparent, self-contained scoring model that runs the moment a lead engages.
How do I choose the point weights in a lead scoring model?
Work backwards from won deals rather than forwards from intuition. List the traits and actions your last several dozen customers shared, keep the three or four that non-customers rarely had, and give those round weights such as ten or twenty. Round numbers are deliberate: they signal the weights are estimates. Then adjust only when outcome data shows a rule failing to separate buyers from non-buyers.
Should a lead scoring model include negative points?
Yes. Without negatives, scores only ever climb, and anyone who browses long enough eventually looks hot. Subtract points for signals that reliably indicate a non-buyer: a free-mail address on a B2B offer, a job title outside your buying committee, a company size below your minimum, or an unsubscribe. Negative rules also stop harmless engagement, like repeated blog reading, from inflating a tier.
How often should scoring weights be recalibrated?
Review quarterly and rebuild whenever something upstream changes: a new pricing tier, a new ideal customer profile, a new ad channel, or a redesigned quiz. Between reviews, watch tier distribution weekly, since a sudden shift usually means the traffic mix changed rather than the model. Recalibration needs a batch of resolved outcomes to learn from, so wait until enough deals have closed.
How many tiers should a lead scoring model have?
Three is the usual answer, because tiers exist to trigger different actions and most teams only have three: call now, nurture, or ignore. Add a fourth only when you genuinely have a fourth play, such as a partner handoff. More tiers than plays creates boundaries nobody acts on and makes each tier too small to measure conversion reliably.
Can a quiz serve as the whole scoring model?
For first-touch qualification, often yes. A scorecard collects fit and intent answers in one session and produces a tier immediately, which is more than most forms deliver. What it cannot see is behaviour after the quiz: email replies, repeat visits, trial usage. Treat the quiz score as the opening balance and let downstream activity add to or subtract from it over time.
Why does sales ignore our lead scores?
Almost always because the score has no visible reason attached. A rep handed the number 84 learns nothing, while a rep told 84 for fifty-plus staff, approved budget and a migration question has an opening line. Show the top contributing rules alongside the number, and let sales mark scores that were wrong. Feeding those corrections back into the weights turns an imposed metric into a shared one.