VISUOSOFTS · DIGITAL MARKETING
Your sales team has finite hours and an infinite list of leads. Lead scoring is how you stop treating those as equal: a simple, data-driven ranking that tells every rep who to call first, who to nurture, and who to leave alone. Teams that adopt it typically see the same headcount produce more pipeline — because the hours move to the leads most likely to close.
This guide builds a lead scoring model from scratch: what to score, how to weight it, where to set thresholds, and the feedback loop that keeps the model honest as your market changes.
WHAT TO SCORE — FIT, BEHAVIOUR, AND NOTHING ELSE
Every scoring model has two ingredient families. Fit: who they are — sector, company size, role seniority, geography, and any firmographic signals that correlate with your wins. Behaviour: what they do — pages visited (pricing pages score heavily), content downloaded, emails engaged with, demo requested, repeat visits in a short window. Fit predicts can they buy; behaviour predicts will they buy soon.
Start with what you can actually observe. A five-field form gives you limited fit data — enrich it or accept the constraint rather than scoring on guesses. Behavioural data is usually richer than teams expect: someone who visited pricing three times this week is broadcasting intent. List every signal available in your CRM and marketing tools before designing the model; the model can only be as good as its inputs.
WEIGHTING FROM DATA, NOT OPINION
The classic failure is a scoring model built in a meeting room: "let's give 10 points for a demo request because that feels important." Build it from your win data instead. Pull your last 100 customers and 100 lost leads, and compare: which attributes and behaviours actually distinguished them? If 70% of customers visited the pricing page but only 20% of lost leads did, pricing visits deserve serious weight. If job title barely differed, it deserves little.
Keep the maths simple enough to explain in one minute — a 0–100 scale with clear point values per signal. Complexity does not improve accuracy at SMB scale; it just makes the model unmaintainable. Document why each signal scores what it does, with the data cited. When sales asks "why is this lead only a 40?", the answer should be a sentence, not a shrug. For the broader qualification context, see our lead qualification playbook.
THRESHOLDS THAT MEAN SOMETHING
Scores are useless without thresholds that trigger action. The standard three: marketing-nurture (below the sales-ready line — automated touches only), sales-ready (routed to a rep within the hour), and priority (top-tier score plus hot behaviour — immediate personal outreach). Set the sales-ready threshold from the data: the score above which close rates justify a salesperson's time.
Calibrate with a simple test: take last quarter's leads, apply the model retroactively, and check the close rate by score band. If the top band does not close at multiples of the bottom band, the model is wrong — adjust weights, not thresholds. Thresholds set by wishful thinking ("we need 50 sales-ready leads a month, so the line goes here") just relabel the same junk. Let the data draw the line.
DECAY AND NEGATIVE SCORING — KEEP IT FRESH
Intent has a half-life. A lead that downloaded your guide six months ago and went silent is not still "warm" — but most models keep the points forever, so the queue fills with the walking dead. Build decay in: behavioural points fade after 30, 60, 90 days of inactivity. A lead should have to stay engaged to stay highly scored.
Negative scoring is equally important: subtract for disqualifiers and junk signals — personal email domains for B2B, job titles like "student", unsubscribes, email bounces, competitors researching you. Explicit disqualification rules ("existing customer enquiring about support → route to support, score zero for sales") prevent the absurdity of your hottest "lead" being someone with a billing question. Review the negative rules quarterly; junk evolves.
THE FEEDBACK LOOP — SALES GRADES THE MODEL
A scoring model without feedback rots. The loop: sales dispositions every worked lead (good, poor, and why), marketing reviews dispositions weekly, and the model is adjusted monthly in the early days, quarterly once stable. When sales marks ten "sales-ready" leads as poor in a row, that is not sales being difficult — it is the model misfiring, and the data will show which signal is lying.
Make feedback frictionless: a two-click disposition in the CRM, not a form. And close the loop visibly — "you told us pricing-page visitors were duds; we checked, you were right, we reweighted" builds the trust that keeps feedback flowing. The model is a living instrument, not a monument. Teams that treat it as finished watch it decay within two quarters as campaigns, markets and buyer behaviour move on.
ROLLOUT WITHOUT REVOLT
Sales teams are rightly sceptical of scoring — most have been burned by a model that sent them rubbish labelled "hot". Earn trust in stages: run the model silently for a month, then show reps the retroactive results ("had we worked top-scored leads first last quarter, you'd have hit quota two weeks earlier"). Pilot with two willing reps, let their numbers do the talking, then roll out.
Keep the human override — reps can pull any lead forward, but they log why, and overrides get reviewed. Override patterns are diagnostic gold: if everyone overrides the same way, the model is missing something real. Position scoring as leverage, not judgement: it does not replace their instinct, it aims their hours. Talk to us and we will build your first model from your actual win data — usually a two-week project with a visible pipeline impact within the quarter.
LEAD SCORING — FAQs
How many signals should our model use?
Eight to fifteen well-chosen signals beat fifty noisy ones. Start with the strongest fit attributes and the clearest behavioural intents, prove the model works, then add sophistication. Every signal must earn its place with evidence it predicts.
Do we need predictive/AI scoring?
Not at SMB scale. Rules-based scoring built from your win data is transparent, maintainable and usually just as accurate on smaller datasets. Consider predictive models when you have thousands of leads and the data science to maintain them.
What is a good score threshold for sales-ready?
Whatever score your data says separates closers from the curious — there is no universal number. Retroactively score last quarter's leads and find the band where close rates justify sales effort. Start conservative and lower the bar only with evidence.
How often should we update the model?
Monthly for the first quarter, quarterly thereafter, plus ad-hoc reviews when you launch new campaigns or enter new segments. Also revalidate annually against a full year of data — markets drift and models must drift with them.
What if sales ignores the scores?
Find out why before forcing it — usually the model is wrong, the thresholds are off, or nobody showed them it works. Silent-run the model, prove it retroactively, pilot with volunteers, and make the scores impossible to ignore by building them into the daily call list.
