B2B lead scoring: a practical model for deciding which contacts deserve a sales call

Monday morning, the salesperson opens the CRM and finds forty new contacts. There’s the student who downloaded a guide for their dissertation, the supplier who wants to sell you something, and the buyer from a perfect-fit company who has looked at the pricing page three times. Without a criterion, they call them in order of arrival, or call none of them at all.

The short answer is this: B2B lead scoring is there to provide that order. Each contact gets a fit score (how closely they resemble the ideal customer: role, industry, size, geography) and a behaviour score (how far they’re moving towards a purchase: pricing page, demo request, webinar, clicks). Above a certain threshold the contact goes to sales; below it, they stay in nurturing. The thresholds are calibrated on closed deals.

Below you’ll find the model we use as a starting point in our marketing automation projects, with the scores, thresholds and rules for handing over to sales.

What is B2B lead scoring and what is it for in practice?

Lead scoring is a points system that ranks contacts by their likelihood of becoming customers, so that salespeople spend their time on those most likely to sign. In B2B it’s especially useful when there are more leads than sales can follow up properly, or when marketing and sales disagree about what a “good lead” is.

A score on its own doesn’t sell anything. Its value lies in three decisions it makes automatic:

• who to call today, because they have the right profile and are showing interest;
• who to leave in nurturing, because they have the right profile but are still far from a decision;
• who to ignore, because they’ll never buy (competitors, job applicants, students, suppliers).

If your company receives ten enquiries a month and the salesperson calls them all back within an hour, you probably don’t need a scoring model: you need more volume. Lead scoring becomes useful when there are choices to be made.

Fit and behaviour: why you need two separate scores

A good model keeps the fit score (who the contact is) separate from the behaviour score (what they do), because they answer different questions and a single total muddles them. The owner of a target company who hasn’t done anything yet and a student who has downloaded five ebooks can end up with the same number, yet they deserve opposite treatment.

Explicit scoring: fit with the ideal customer

Fit is calculated on the data the contact gives you (forms, LinkedIn, CRM enrichment) and on company data. In B2B, four criteria almost always matter:

• role: decision-maker, technical influencer, end user, or no role in the buying process;
• industry: those where you already have customers and case studies carry more weight than “possible” ones;
• size: number of employees or revenue, within the range where your product makes sense;
• geography: where you have a sales network, support and logistics.

We often add the type of email address (business or personal) and, for those selling machinery or plant, the installed base or the technology used.

Implicit scoring: the behaviour that signals interest

Behaviour is measured through actions tracked on the website, in emails and at events. They aren’t all worth the same: what makes the difference is how close they are to a purchase.

• strong signals: a demo or quote request, a visit to the pricing page or configurator, a return to the site after a call;
• medium signals: attending a webinar, downloading a technical data sheet, reading a case study from their own industry;
• weak signals: clicking a newsletter, reading a blog article.

We keep email opens at zero, or close to it. Since Apple introduced Mail Privacy Protection, announced in June 2021 with iOS 15, many opens are recorded automatically even if nobody has read the message, inflating the score without saying anything about interest.

Negative scores and decay: how to stop the score from lying

Negative points take people who will never buy off the radar; decay lowers the score of those who have gone quiet. Without these two rules the score only ever goes up, and after a year half the database looks “hot”.

Examples of negative points we almost always include:

• a visit to the careers page, or a job application;
• an email address on a competitor’s domain;
• the role of student, sales consultant or agency;
• a company outside the area served or below the minimum size;
• an unsubscribe from emails, or a hard bounce.

Decay applies to behaviour only. A purchasing director’s role doesn’t expire; a visit to the pricing page eight months ago does. Tools handle it differently: in HubSpot lead scoring (Professional and Enterprise plans) you can set a percentage reduction at regular intervals, for example 50% every three months; in ActiveCampaign you can make a rule’s points expire after a certain period, so the action stops counting all at once. If your CRM doesn’t support it, a monthly automation that resets or halves the behaviour points of contacts inactive for 60 or 90 days will do the job.

A complete B2B lead scoring model, with scores

The following model is designed for a company that sells a technical product or service to other businesses, with a sales cycle of a few months. It’s a starting point: the weights need adapting to your market and then correcting with data.

Take a company that produces automation systems for the food industry, sells in Italy and German-speaking countries, and works well with customers of between 50 and 500 employees.

Area	Criterion	Points
Fit	Decision-making role (owner, managing director, technical or production director)	+15
Fit	Influencer role (technical office, purchasing, maintenance)	+8
Fit	Student, consultant, agency	-15
Fit	Primary industry (food, beverages)	+10
Fit	Secondary industry (packaging, pharmaceuticals)	+5
Fit	Off-target industry	-10
Fit	50-500 employees	+10
Fit	10-49 employees	+4
Fit	Fewer than 10 employees	-10
Fit	Italy, Germany, Austria, Switzerland	+5
Fit	Outside the area served	-20
Fit	Business email address	+5
Behaviour	Demo or quote request	+40
Behaviour	Visit to pricing page or configurator (once every 7 days)	+15
Behaviour	Attending a webinar	+15
Behaviour	Registering for a webinar without attending	+5
Behaviour	Downloading a technical data sheet	+10
Behaviour	Reading a case study from their own industry	+8
Behaviour	Clicking an email (maximum 3 times a month)	+3
Behaviour	Opening an email	0
Behaviour	Visiting the careers page	-30
Behaviour	Unsubscribing from emails	-20
Decay	No activity for 60 days	-50% of behaviour points

Maximum fit: 45 points. Behaviour has no theoretical ceiling, but the frequency limits (one pricing visit a week, three clicks a month) prevent a contact who is very active on the newsletter from overtaking someone who has asked for a quote.

Two examples to show how it behaves. The production director of a cheese-maker in Emilia with 120 employees reaches 45 fit points; they attend a webinar and look at the pricing page: 30 behaviour points. High fit and medium behaviour, so in the matrix below they’re an MQL. A consultant from a training company in Spain starts with a fit below zero (consultant, off-target industry, outside the area served) and can click every newsletter they like: the salesperson will never see them on the list.

How do you set the MQL and SQL thresholds?

The MQL (Marketing Qualified Lead) threshold indicates when a contact is on-target and active enough to deserve sales’ attention; the SQL (Sales Qualified Lead) comes later, when the salesperson has verified that there’s a real project. The first thresholds are estimates, then they’re corrected by looking at the scores of closed deals.

With two separate scores, it’s better to think in combinations rather than a single number. This matrix is the one we propose as a first version:

Fit / behaviour	Low (under 15)	Medium (15-34)	High (35 or more)
High (30-45)	Targeted nurturing	MQL	Priority MQL, call the same day
Medium (15-29)	Nurturing	Intensive nurturing	MQL
Low (under 15)	No action	Generic nurturing	Manual check

A demo or quote request is the exception: if fit is at least medium, the contact goes straight to sales, regardless of anything else. Nobody wants a company asking for a quote to wait because it “doesn’t have enough points yet”.

Calibrating thresholds on closed deals

After three to six months you have enough data to check whether the thresholds are right. The procedure is simple:

1. Export from the CRM the closed opportunities of the last 12-24 months, won and lost, with the score the contact had at the moment of handover to sales.
2. Divide the contacts into score bands and calculate the percentage of won opportunities for each.
3. Look for where the percentage changes gear. If almost nobody below 40 behaviour points closes and people above it do, that’s where the threshold is.
4. Check the criteria one by one: if almost all your won customers come from an industry you marked as “secondary”, increase its weight.
5. Do the same with the lost deals: a criterion that often appears among those who said no deserves fewer points, or negative points.

If there are few closed opportunities, the model remains largely a hypothesis. That’s fine, as long as you say so and review it more often. To connect the score to what happens next, right up to the invoice, we’ve written a guide on following the journey from lead to contract, from CRM to revenue.

The handover to sales: the SLA that makes the score useful

The SLA (service level agreement) between marketing and sales sets out what happens when a contact crosses the threshold: who takes it, how quickly they make contact and how they report back on the outcome. Without this agreement the score becomes a number nobody looks at.

Speed matters. Research published in Harvard Business Review, based on 1.25 million leads received by 42 US companies (29 B2C and 13 B2B), found that those who tried to contact the lead within an hour were almost seven times more likely to qualify it than those who waited even one hour longer (The Short Life of Online Sales Leads). The study dates from 2011, but the dynamic hasn’t changed: someone who asked for a quote and gets a call the next day has often already received another offer.

A starting SLA might look like this:

• demo or quote request: first contact within one working hour;
• priority MQL: contact within the same working day;
• MQL: contact within two working days;
• mandatory outcome within five days: accepted (becomes an SQL), rejected with a reason, or to be recontacted on a set date;
• marketing’s commitment: an agreed minimum number of MQLs a month, with quality measured by the share accepted by sales.

The rejection reason is the most valuable data point in the whole system. “Company too small”, “no budget this year”, “already a customer of a competitor on a multi-year contract”: if the same reason keeps coming back, it should become a criterion in the model.

What to do with contacts who aren’t ready yet: automations and nurturing

Contacts below the threshold shouldn’t be thrown away: they should be followed up with a lead nurturing journey that brings them back to sales when their score rises. This is where lead scoring and marketing automation work together.

The automations we set up most often:

• high fit, low behaviour: a content sequence for the industry and role (case studies, data sheets, webinar invitations), with a personal email from the salesperson after a few weeks;
• medium fit, rising behaviour: an alert to the salesperson when the contact returns to the pricing page, even if they haven’t crossed the threshold yet;
• rejected by sales with the reason “not now”: automatic return to nurturing and a reminder to the salesperson on the date given;
• MQL not worked within the SLA: a notification to the sales manager;
• contact inactive for six months: a reactivation campaign, then removal from the list if there’s no response.

We’ve collected examples of complete flows in our guide to B2B lead nurturing, and the basics of workflows, triggers and segments are explained in marketing automation: what it is and how it works.

The implications for lead generation, CRM and campaigns

A scoring model also changes the way you measure marketing: the number of leads matters less, while what counts is how many become accepted MQLs and then opportunities. It’s the same logic as the metrics for managing B2B leads, with an extra level of detail.

In practice:

• campaigns: you can compare sources (Google Ads, LinkedIn, trade fairs, organic) by share of MQLs rather than by cost per lead. A campaign that brings few contacts, almost all with high fit, is worth more than one that fills the CRM with students;
• offline conversions: if you send the “MQL” or “SQL” event back to Google Ads or Meta instead of a simple form submission, the algorithms learn to look for contacts similar to the good ones;
• CRM: you need clean fields for role, industry and size. A form with “Industry: other” as the first option makes half of the fit score useless;
• reports: every month we look at how many MQLs were generated, how many were accepted, how quickly they were worked and how many became opportunities.

Predictive scoring and AI in CRMs: what HubSpot, ActiveCampaign and Salesforce do today

Predictive scoring uses machine learning to estimate the likelihood of conversion from your customer history, instead of hand-written scores. It works well when you have plenty of clean data; with only a few deals a year, a rule-based model remains more reliable and easier to explain to the salesperson.

The situation in October 2026 in the three CRMs we come across most often:

CRM	Rule-based scoring	Predictive scoring	What to know
HubSpot	Fit, engagement and combined scores, with thresholds and decay (Marketing or Sales Hub Professional and Enterprise)	“Likelihood to close” and “Contact priority” properties (Enterprise plans)	The estimated probability refers to closing within 90 days, and contacts are grouped into four priority tiers
ActiveCampaign	Contact and deal scores, with positive and negative points and expiry	Win Probability on deals	The probability is calculated on the won deals in each pipeline, and irrelevant factors can be excluded
Salesforce	Custom rules or via Account Engagement	Einstein Lead Scoring	A minimum history is needed: at least 1,000 leads created in the last 200 days and at least 120 converted; otherwise it uses a global model built on anonymised data

Sources: HubSpot’s documentation on predictive lead scoring, ActiveCampaign’s page on Win Probability and Salesforce’s page on the Einstein Lead Scoring requirements, which is included in Sales Cloud Einstein (Performance and Unlimited editions, paid add-on in Enterprise).

Our advice: start with a rule-based model, run it for six months, then add predictive scoring alongside and compare. If the predictive and rule-based scores point to the same contacts, good; if they diverge, the comparison tells you which signals you were ignoring. Predictive scoring has a practical limitation: it explains very little about why a contact has a given score, and salespeople tend not to trust a number they don’t understand.

Lead scoring mistakes to avoid

The mistakes we see most often aren’t technical: they concern maintaining the model and the relationship with sales.

• Too many criteria. Nobody remembers thirty rules weighted from 1 to 7, and nobody knows how to correct them. Ten or fifteen criteria are almost always enough.
• A score that’s never reviewed. A model set up when the CRM was launched and never touched again describes the market as it was then. A review every six months, based on closed deals, is the minimum.
• Sales don’t use it. If sales weren’t involved in defining the criteria, they’ll carry on working their own way. Involve them from the start, showing them the ranked lists and asking where the score gets it wrong.
• A single total. Adding fit and behaviour together hides the worst cases, such as the very active off-target contact.
• No negative points and no decay. The whole database turns hot and the score loses all meaning.
• Empty fit data. If role and industry are missing for 60% of contacts, the model works on behaviour alone. Make the key fields mandatory on high-value forms and use enrichment where possible.

Checklist for setting up the model in a month

1. Bring marketing and sales together and define the ideal customer jointly: role, industry, size, geography.
2. List the actions that currently precede a request for a quote (pages, content, events).
3. Check that the CRM collects that data and that website tracking records it against the contact.
4. Write the first version of the model: 10-15 criteria, negative points, decay.
5. Apply it to your history and check whether your current customers would have qualified as MQLs.
6. Agree the SLA: contact times, outcomes, rejection reasons.
7. Set up the automations for alerts, nurturing and returns.
8. After three months, first review of the thresholds on real data; then every six months.

Want to set up lead scoring with your sales team?

The model works when marketing and sales build it together. That’s why we set it up through a workshop: one or two sessions in which we define the ideal customer, buying signals, thresholds and SLA, then configure it in your CRM and connect it to B2B lead nurturing flows. It’s part of our work on marketing automation and data analysis.

If you want to find out whether it makes sense in your case, book a call with Federico or write to us via the form. We’ll look at volumes, CRM and sales process and tell you frankly whether it’s worth starting with scoring or with something else.

Frequently asked questions

How many leads do you need for lead scoring?

For a rule-based model there’s no minimum: you just need more contacts than the salesperson can follow up properly. Predictive scoring needs higher volumes; Salesforce, for example, requires at least 1,000 leads in 200 days and 120 conversions.

What’s the difference between an MQL and an SQL?

An MQL is a contact who, according to the score, is on-target and active enough to be passed to sales. It becomes an SQL when the salesperson has spoken to them and verified that there’s a project, a contact person and a timeframe.

How often should the lead scoring model be reviewed?

The first review after three months, then at least every six, comparing scores with won and lost opportunities. It should also be reviewed when your product, markets or website change.

Does lead scoring work with only a few contacts a month?

With a few dozen contacts a month, the salesperson can assess them one by one. In that case it’s worth using at least fit and negative points, to filter out off-target contacts, and leaving the rest until volumes grow.

Does predictive scoring replace rule-based scoring?

No, it usually works alongside it, and only when the history is large and clean. With only a few closed deals a year, a rule-based model is more reliable and easier to explain to sales.