Lead Scoring Models: How to Choose and Build One That Sells

Lead Scoring Models: How to Choose and Build One That Sells

A lead scoring model is a set of rules that gives every lead a number, so your sales team knows who to call first. The best models for B2B businesses score two things separately: how well the lead fits your ideal customer, and how close they are to buying. Then they prove the score against closed revenue in your CRM, not against form fills.

That last step is where most models fall down. A score that has never been checked against actual sales is a guess with a number on it. This guide covers the main model types, how to build one, how to set thresholds, and how to test it so it earns its place.

TL;DR

• Start with a simple rules-based (points) model. Move to predictive scoring only when you have enough clean CRM history to train it.
• Score fit (who they are) and intent (what they do) on separate scales. A high-fit, low-intent lead needs nurturing. A high-fit, high-intent lead needs a call today.
• Set your threshold from real conversion data by score band, not from a gut feel, and review it with sales regularly.
• If your ads feed the model, the model is only as good as your conversion tracking. Fix tracking first, then score.
• Not every business needs scoring. If you can call every lead the same day, spend your time on follow-up speed instead.

Table of Contents

• What a lead scoring model actually is
• The main types of lead scoring models
• Fit and intent: score them separately
• How to build a B2B lead scoring model, step by step
• Setting thresholds and passing leads to sales
• Testing the model against closed revenue
• When lead scoring is not worth it
• Where your ads fit: scoring is only as good as the lead source
• Best practices that keep a model honest
• Our perspective
• Not sure your leads are worth scoring?
• FAQ
• Sources

What a lead scoring model actually is

Lead scoring ranks prospects by assigning values to their attributes and behaviour. Salesforce describes it as a way to indicate lead quality and likelihood of conversion so sellers can prioritise effort. The model is the rulebook behind the number: which attributes earn points, how many, and what happens when a lead crosses a line.

G2 notes that scores often run from 0 to 100, but the scale matters less than whether the score reflects the lead’s attributes and how well they suit your product. Spotler’s Australian guide adds that scales can also be letter grades, traffic lights or a simple yes/no. Pick whatever your sales team will actually read.

Two things a score is not. It is not a qualification decision, and it is not a forecast of revenue. It ranks leads. Someone still has to decide who gets a call, who gets nurtured and who gets dropped.

The main types of lead scoring models

A systematic review of 44 lead-scoring studies, published in PubMed Central, splits models into two families: traditional and predictive. Traditional models include filters, rules or points, and scorecards. Predictive models use classification, clustering or regression.

Manual or automated

Spotler points out that scoring can be manual or automated. Manual scoring gives you more control but takes more time and can be less objective. Automated scoring can combine things like website visit frequency and promotional email opens without anyone touching it.

In practice, we see most Australian SMEs do best with an automated points model inside their CRM or marketing platform, built from rules a human has checked against won deals.

Why predictive is not the default

Predictive scoring sounds like the upgrade, but it depends on data. One vendor’s implementation documentation, from SegmentStream, recommends at least six months of historical data for training, clean sales and lead records, and capturing the Google Client ID. If your CRM is patchy, a predictive model will learn the patches.

A points model you can explain in one sentence beats a predictive model nobody trusts.

Fit and intent: score them separately

The most common design mistake is mixing everything into one number. A student who downloads five guides and opens every email can outscore a managing director at your ideal customer who visited your pricing page once. Combined scores hide that.

A cleaner approach is to run two scores.

Fit (explicit) criteria

Fit is who the lead is. G2 lists inputs such as job title, industry, company size and employee headcount. For B2B, Twilio adds revenue, growth rate, locations, technology stack and relevant job title as firmographic factors.

For an Australian business, add the filters that matter locally: are they in a state or region you can serve, are they a registered business rather than a consumer, and do they sit in a revenue band you can profitably work with.

Intent (behavioural) criteria

Intent is what the lead does. G2 names web visits, form submissions, content downloads, webinar registrations, email activity and social engagement. Not all actions are equal. A pricing page visit or a request for a demo says far more than a blog read or an email open.

The point of splitting the scores is the action it creates. A high-fit, low-intent lead goes into nurture. A high-fit, high-intent lead gets a phone call today. A low-fit, high-intent lead might be a competitor, a student or a job seeker, and should not eat your sales time.

Add negative scores

Points should go down as well as up. Subtract for personal email domains on a B2B offer, for roles that never buy, for locations you do not serve, and for leads who unsubscribe. Also let engagement points fade. A lead who was active eight months ago is not the same lead today.

How to build a B2B lead scoring model, step by step

1. Define the outcome. Decide what “good” means: a booked sales call, an opportunity created, or closed-won revenue. Closed-won is the strongest, because it is the one your bank account sees.
2. Pull your closed-won and closed-lost deals. Export them from your CRM with the attributes you captured at lead creation.
3. Describe your ideal customer. G2 recommends starting with a target customer profile covering demographics, company size and location.
4. Find what separates winners from losers. For each attribute (industry, role, source, page visited), calculate the close rate. G2 recommends this: calculate close rates for attributes, then assign point values accordingly.
5. Assign points. Give more points to attributes with higher close rates. Keep the model small: eight to twelve criteria are easier to maintain than forty.
6. Set negative scores and decay. Penalise poor-fit signals and reduce engagement points over time.
7. Set a threshold and a routing rule. Decide what score triggers a handoff and who owns the lead.
8. Test it, then review it. More on this below.

A worked example (hypothetical)

Imagine a Melbourne commercial cleaning company that sells to office managers. This is an illustration, not a client result.

• Fit: facilities or operations manager title (+20), office of 50 or more staff (+15), Melbourne metro (+10), residential email address (-20).
• Intent: requested a quote (+30), visited the pricing or services page twice or more (+15), downloaded the checklist (+5), opened an email (+1, capped).

A facilities manager at a 120-person Richmond office who requests a quote scores 75 before anything else. A student who downloads the checklist scores 5. Nobody needs a machine-learning model to see who should get the call.

Setting thresholds and passing leads to sales

A threshold is the score at which a lead becomes sales-ready. Apteco’s Australian guide describes it simply: points are awarded for information and behaviours, and reaching your threshold makes a lead qualified for sales contact.

Do not pick the number by feel. Bucket your historical leads by score band (0 to 20, 21 to 40, and so on) and look at how many in each band became opportunities and customers. The threshold usually sits where conversion jumps. If the 60 to 80 band closes at six times the rate of the 40 to 60 band, 60 is your line.

Then make the handoff concrete: who gets the lead, how fast they must act, and what happens if they do not. For a deeper look at that handoff, see our guide to the MQL vs SQL handoff for Australian marketers. For connecting marketing scores to wider sales qualification frameworks, this MQL to SQL acceptance playbook is worth a read.

The qualification step still matters after the score. Salesforce ANZ’s qualification guide suggests probing goals, plans, challenges, timeline, budget and authority. A score tells you who to call. The conversation tells you whether they are real.

Testing the model against closed revenue

This is the part most guides skim. The academic review above describes the sensible approach: define a measurable outcome, use historical CRM data to compare conversion rates by score band and by individual criterion, test thresholds and weights on later cohorts, watch data quality and false positives, and review the model regularly with sales.

Practically, that means:

• Test on leads the model has not seen. Build the weights on last year’s leads, then check them against the most recent quarter.
• Check each criterion on its own. If “downloaded the ebook” adds 10 points but those leads never close, remove it.
• Watch false positives. High scores that never convert are the quickest way to lose sales trust.
• Review quarterly with sales in the room. The people on the phones know which leads waste their time.

There is a reason to be humble about all this. A research report on assessing sales opportunities found that 79% of respondents had low or very low confidence in their lead scores, and 62% said scores were reliable only 40% to 60% of the time. The same report found 49% of surveyed firms used a formal methodology at all. Scores are directional. Use them inside a broader qualification process, not instead of one.

When lead scoring is not worth it

Apteco cautions that scoring may be unnecessary if you can comfortably manage your daily lead volume or have a short sales cycle. It earns its keep when teams must prioritise a large pool of prospects.

If you get 15 enquiries a week and your team calls every one within an hour, scoring will not fix your results. Speed to lead, a better quote process and better lead quality at the source will. That last point brings us to advertising.

Where your ads fit: scoring is only as good as the lead source

If you buy leads through Google Ads or Meta Ads, scoring sits downstream of a bigger question: are the ads bringing in the right people in the first place? A model can rank a bad pile of leads, but it cannot turn it into a good one.

Here is what we see when we audit accounts that feed a CRM:

• Platform conversions count form fills, not customers. The ad platform reports a “lead” the moment someone submits a form. Your CRM knows which of those became revenue. The gap between the two is where the wasted spend hides.
• Brand searches flatter the numbers. People searching your business name were already coming to you. Keep brand exclusions on acquisition campaigns, and report brand and new demand separately, so you can see what the ads create.
• The same person gets counted twice. A lead who fills in a form, then calls, then returns via retargeting can show up as three conversions. Count each lead once.
• Broad targeting brings the wrong enquiries. We do not use broad match keywords on Google Ads. Exact and phrase match, tight negative keywords and regular search term reviews keep out the students, job seekers and price-shoppers who drag your scores and your sales team’s morale down.

The fix is a closed loop. Capture the click ID (GCLID for Google, fbclid for Meta) when the form is submitted, store it against the lead in your CRM, and send the outcome back to the platform when the lead qualifies or closes. Our guide to closed-loop attribution with GCLID and fbclid walks through it, and our pipeline attribution audit shows how to tie spend to closed-won deals.

Once that loop runs, your lead score becomes a bidding signal as well as a sales tool. The platform learns which clicks produce high-scoring leads and paying customers, not just form fills. If your tracking is shaky, fix Google Ads conversion tracking before you build anything on top of it. For the wider picture on B2B campaigns, read our B2B Google Ads strategy guide.

Best practices that keep a model honest

• Keep it simple enough to explain. If sales cannot say why a lead scored 82, they will stop using it.
• Agree the definition with sales first. G2’s view is that scoring creates value only when marketing and sales are aligned.
• Capture the data you score on. You cannot score job title if the form does not ask for it. Keep forms short, but ask for the two or three fields that separate a good lead from a poor one.
• Review weights from conversion data. Treat every weight as a hypothesis until closed revenue backs it.
• Clean your CRM. Duplicates, missing sources and unlogged outcomes will quietly break any model.
• Separate sources. Score and report paid search, paid social, organic and referral leads separately. A 60 from a Google Ads lead and a 60 from a referral rarely mean the same thing.

Our perspective

Most lead scoring projects we see start in the wrong place. The business picks a tool, loads in a long list of point values from a template, and never checks any of it against what closed. Six months later, sales ignores the scores and marketing blames sales.

We work the other way around. First, we make sure tracking is sound: one lead counted once, the click ID stored in the CRM, and the outcome flowing back. Then we look at which sources, keywords and audiences produce deals, not just enquiries, and cut the spend that is not pulling its weight. Only then does a score add something, because it is built on revenue rather than opinions.

A simple model tested against closed revenue will beat a sophisticated one nobody has validated. And if you are not sure your leads are any good, scoring them harder is the wrong answer. Look at where they come from.

— Jesse Lees

Not sure your leads are worth scoring?

Start with the source. Our $99 AUD Google Ads audit checks your tracking, brand and non-brand split, match types, negatives and wasted spend, and shows whether your account is producing real customers or just form fills. If tracking is the problem, our website tracking service maps it to CRM revenue. We have no 12-month lock-ins, and we will say upfront if we cannot help.

FAQ

What does a lead score mean?

A lead score is a number that estimates how likely a lead is to become a customer, based on who they are and what they have done. A higher score means a higher priority for sales follow-up. It is a ranking tool, not a guarantee, so it works best alongside a conversation that confirms need, budget and timing.

What is the best lead scoring model for a small business?

A rules-based points model, with fit and intent scored separately, suits most small and mid-sized businesses. It is easy to explain, quick to build in a CRM and simple to adjust. Predictive models make more sense once you have a large lead volume and months of clean historical data.

How often should I update my lead scoring model?

Review it at least quarterly with your sales team, and sooner if your offer, market or ad channels change. Compare conversion rates by score band against closed deals, then adjust weights and thresholds where the data disagrees with the model.

Do I need lead scoring if I only get a few leads a week?

Probably not. If your team can follow up every enquiry quickly, a score adds little. Put your effort into response time and lead quality at the source, and revisit scoring when volume grows past what the team can handle.

Sources

• What Is Lead Scoring? | Salesforce
• Lead Scoring: What It Is, Models, Best Practices (G2)
• What is lead scoring? A beginner’s guide (Spotler Australia)
• The state of lead scoring models and their impact on sales performance
• Lead Scoring conversions (SegmentStream documentation)
• The Complete Guide to Lead Scoring Models (Twilio)
• How to start lead scoring (Apteco Australia)
• Learn To Qualify A Lead (Salesforce ANZ)
• Assessing Sales Opportunities (research report)

Recommended

• Professional Services Lead Generation: Stop Chasing Lead Volume
• Close More Revenue: MQL vs SQL Handoff Playbook for Australian Marketers