When marketing works, sales gets more leads than it can call properly. Some are ready to buy, many are researching, and some will never be a fit. Without a way to tell them apart, salespeople work through leads in the order they arrived, and the best buyer of the week waits behind a student doing an assignment. Lead scoring fixes the order.
What is lead scoring?
Lead scoring gives each lead a number or grade that reflects how likely it is to become a customer, based on who the lead is and what they have done. Sales uses the score to decide who to contact first, how quickly and in what way. Low-scoring leads are not discarded; they go into slower nurturing until they show more interest.
Rules-based or predictive scoring?
| Rules-based scoring | Predictive scoring | |
|---|---|---|
| How it works | Points for chosen traits and actions, such as +10 for a director, +15 for a pricing page visit | A model learns which traits and actions preceded won deals in your history |
| Data needed | Little; can start on day one | A meaningful history of closed won and lost deals |
| Strengths | Simple, transparent, easy to adjust | Finds patterns people miss, and updates as buyers change |
| Weaknesses | Based on opinion; drifts out of date | Needs clean data; harder to explain each score |
| Best for | Newer pipelines and smaller volumes | Businesses with steady lead volume and good CRM records |
Most businesses should start with rules, agreed between sales and marketing, and move to predictive scoring once the CRM holds enough reliable history. The rules exercise is valuable in itself: it forces everyone to agree what a good lead looks like.
What signals feed a lead score?
| Signal type | Examples |
|---|---|
| Fit (who they are) | Industry, company size, job role, location, budget range given on the form |
| Intent (what they did) | Pricing page visits, return visits, case studies read, calculator or assessment completed |
| Source | Search ad, organic search, referral, LinkedIn, WhatsApp, event |
| Engagement | Email replies and clicks, meeting booked, response to the first call |
| Negative signals | Student or job-seeker answers, personal email for a business service, out-of-area location, competitor domain |
How does a predictive lead scoring model work?
- Gather historical leads with their traits, behaviour and final outcome: won, lost or never qualified.
- Clean the data: remove duplicates, standardise fields such as industry and company size, and fix missing sources.
- Train a classification model, often logistic regression or gradient-boosted trees, to predict the chance each lead becomes a customer.
- Test it on leads the model did not see during training, to check it predicts well rather than memorising the past.
- Score new leads as they arrive and write the score back into the CRM.
- Retrain on a schedule, because markets, campaigns and offers change.
Some CRMs, including HubSpot and Salesforce, offer built-in predictive scoring on higher subscription tiers. A custom model makes sense when you need to combine data the CRM does not hold, such as product usage, payment history or offline sales.
How do you put lead scores to work?
- Route the highest-scoring leads to a senior salesperson with an alert, and set a same-day response target.
- Send mid-range leads into a short sequence of useful follow-ups, with a call once they engage.
- Place low-scoring leads in long-term nurturing, such as a newsletter, rather than a sales queue.
- Show the reasons behind each score, such as "pricing page visited three times", so salespeople trust it and open the call well.
- Send scores back to ad platforms as conversion values, so bidding favours the kind of leads that close. See our guide to marketing attribution.
How do you know if lead scoring works?
Compare outcomes by score band. If the top band converts to customers at a clearly higher rate than the middle, and the middle beats the bottom, the score is doing its job. Track response times for top-band leads, and check a sample of low-scoring leads every month: if good buyers keep appearing there, the model or rules need changing.
What are common lead scoring mistakes?
- Scoring on what is easy to track, such as email opens, rather than what predicts a sale.
- Training a model on a messy CRM, where lost deals were never closed off and sources are missing.
- Setting rules once and never revisiting them.
- Using the score to ignore leads instead of to order them.
- Building scoring before agreeing, with sales, what a qualified lead is. Our guide to B2B inbound marketing covers that agreement.
What does it take to set up predictive lead scoring?
The groundwork is usually the bigger job: consistent CRM records, lead sources captured from every channel, and outcomes recorded for every deal. At Lulonke, this sits within our data and growth intelligence work, which starts with a diagnostic audit; the initial architecture is typically R65,000 to R220,000. See AI growth intelligence and our guide to what a growth intelligence layer does.
Frequently asked questions
What is predictive lead scoring?
Predictive lead scoring uses a machine learning model, trained on your past won and lost deals, to estimate how likely each new lead is to become a customer. Sales uses the score to decide who to contact first.
How much data do I need for predictive lead scoring?
There is no fixed number, but the model needs a meaningful history of closed deals, both won and lost, recorded consistently with lead source and outcome. With too little history, start with rules-based scoring.
What is the difference between lead scoring and lead grading?
Many systems separate fit, often shown as a grade such as A to D, from interest or engagement, shown as a score. Combining both tells sales whether a lead is the right kind of customer and whether they are ready to talk.
Does HubSpot have predictive lead scoring?
HubSpot and Salesforce both offer built-in predictive scoring on higher subscription tiers. A custom model is useful when you need to combine data from systems outside the CRM.
Is lead scoring allowed under POPIA?
Scoring leads on information they gave you and their activity on your own website and emails is common practice, provided you collect only what you need, explain it in your privacy notice and protect the data. Take legal advice for automated decisions with significant effects on people.
