What Is Lead Scoring?

A lead is a contact who has left their details, usually through a form. How they get there is covered in the lead generation entry. Lead scoring starts after that and sorts. It measures in two directions.

The first is the profile, often called fit: does the contact match the offer? HubSpot's help describes fit scores as qualifying records based on demographic information such as age, job title, company size or annual revenue. Salesforce calls this grading in Marketing Cloud Account Engagement: prospects are graded based on profiles, and the more criteria someone matches, the better the grade. The scale runs from A+ to F.

The second direction is behaviour, often called engagement: how much does the contact interact with the company? HubSpot gives as examples visiting the website, subscribing to a newsletter, clicking a call to action or opening a marketing email. In Salesforce, the score measures implicit interest by assigning points to activities such as opening an email, attending a webinar or submitting a form.

The two values can be combined. HubSpot uses levels from A1 to C3 for this, where the letter stands for fit and the number for engagement. According to HubSpot, A means high fit and C low fit, while 1 means high engagement and 3 low engagement. A contact with low fit but high engagement therefore gets C1.

In the classic points model, the team sets the rules itself. Each attribute and each action gets a value, including negative ones. HubSpot's example is adding points for a good-fit industry and size and subtracting points for a region the company does not operate in. An example with made-up values: marketing manager role +15, company with more than 50 employees +10, pricing page visit +10, located outside the delivery area −30.

Predictive scoring lets a model work out the weights by learning from past cases. HubSpot states that its AI score is trained on contacts and needs at least 50 of them, 25 converted and 25 not converted. Salesforce writes that Einstein Behavior Scoring uses machine learning to uncover the most influential behaviour signals and assigns a score from 0 to 100. These are the vendors' statements about their own products.

A 2025 case study by Laura González-Flores and colleagues separates the two approaches. In their description, traditional models are based on rules, points and scorecards and usually lack formal statistical validation. At the B2B software company they studied, the machine learning model identified high-quality leads considerably better than the previous method. That is one company, not a general finding.

The handover to sales needs a threshold. In HubSpot, thresholds can be set that sort contacts into high, medium and low. Salesforce describes the threshold as the score at which a prospect's activity suggests a potential customer. There is no generally correct number. Marketing and sales set it together and check it against the results of earlier handovers, as the lead nurturing entry describes.

Points for behaviour should decay over time, or a download from two years ago counts as much as one from yesterday. HubSpot offers score decay for engagement and combined scores, at intervals of 1, 3, 6 or 12 months. Salesforce describes decay as reducing a prospect's score for inactivity.

Lead scoring decides what sales spends its time on. Without a score, sales calls contacts in the order they arrived, including those who will never buy. With a threshold set too low, immature leads reach sales; set too high, ready buyers sit waiting. Both can only be measured by the outcome: how many handed-over leads sales accepts, and how many become customers.

Scoring individual people also touches data protection law. The GDPR defines profiling in Article 4(4) as automated processing of personal data to evaluate personal aspects relating to a natural person, in particular to analyse or predict aspects such as their economic situation, interests or behaviour. Under Article 21(2), where personal data are processed for direct marketing, the data subject may object at any time, including to profiling to the extent that it is related to such direct marketing. If they object, the data may no longer be processed for those purposes (Article 21(3)). Article 22(1) gives the right not to be subject to a decision based solely on automated processing that produces legal effects or similarly significantly affects the person. Whether a given scoring falls under this needs a legal assessment case by case.

• Keep profile and behaviour points separate, so it stays visible whether a lead fits or is merely active.
• Agree criteria and threshold in writing with sales before the model goes live.
• Use negative points too, for example for job seekers, competitors or regions outside your offer.
• Let behaviour points decay.
• Check the model regularly against actual deals and adjust the weights.
• Clarify with your data protection officer on what basis you score people, and apply objections to direct marketing in the scoring too.

• Counting behaviour only, so that students or competitors with many downloads end up at the top.
• Copying thresholds from examples in vendor help pages without testing them against your own data.
• Never letting points decay.
• Training a predictive model on too few converted cases.
• Setting up the model once and never comparing it with sales results again.