Lead scoring for website conversations: a simple model

TL;DR

• Website conversations carry better scoring signals than page views, because visitors tell you what they need, how big their team is and when they want to start.
• Score two things separately: fit (who they are) and intent (what they are trying to do now). Subtract points for clear non-buyers. Act on the combination with three or four simple thresholds.
• Keep the rules on one page that sales can read, let intent fade over time, and recalibrate against real outcomes after your first 50 to 100 leads.

Why score conversations at all

Traditional lead scoring adds points for clicks, page views and email opens, then guesses at intent. A conversation on your website is different: the visitor says what they want in their own words. "We're 40 people, moving off spreadsheets, need it live before January" is worth more than ten page views.

Scoring those conversations answers one practical question: what should happen next with this visitor? Offer a meeting now, send a useful follow-up, or simply answer and let them go. Without a score, every conversation either goes to sales, which drowns the team, or none do, which loses buyers.

It also matters because of how B2B buying works now. Forrester's 2026 research found that a typical decision involves 13 internal stakeholders and nine external participants (Forrester), and Gartner found 69% of buyers prefer to validate AI-generated insights with a sales rep (Gartner). Many website conversations are research by one member of a group, and a few are the moment a buyer wants a person. Scoring is how you tell them apart. It is the numeric side of lead qualification.

The model in one picture

The total runs from 0 to 100. Keep fit and intent visible as separate numbers as well, because a perfect-fit company with no intent needs nurturing, not a sales call, and a hot visitor from a company you cannot serve needs a polite answer, not a meeting.

HubSpot's scoring tool uses the same split. It distinguishes fit scores, which "qualify records based on their demographic information through property values," from engagement scores based on "actions and interactions," and its combined labels run from A1 to C3, where the letter is fit and the number is engagement (HubSpot).

Fit points: who they are

Use the facts your sales team already uses to decide whether a deal is worth pursuing. Four or five signals are enough. An example for a B2B software company selling to teams of 20 to 500:

Cap fit at 50. If a fact is unknown, give it zero rather than guessing. Many conversations will score on two or three fit signals; that is fine.

Intent points: what they are doing now

Intent comes from what the visitor asks and says. This is where conversations beat forms, because most of these signals arrive without a single qualifying question.

Cap intent at 50. Note that a security question is a strong signal on its own: people rarely ask about a DPA out of curiosity.

Negative signals

Subtract points, or route the conversation elsewhere entirely, when the visitor is clearly not a prospect:

• Job seeker: route to the careers page. Minus 50.
• Student or researcher: answer, no sales follow-up. Minus 30.
• Vendor or agency pitching you: minus 50.
• Existing customer with a support issue: route to support. Do not score as a new lead.
• Competitor research: if it is obvious, minus 30.

Routing these correctly is as valuable as spotting the hot leads. It keeps your sales team's queue clean and gets the job seeker and the customer to the right place faster.

Thresholds that trigger action

A score is only useful if each range leads to a specific next step. Start with four:

For hot leads, speed matters more than anything else on this page. Our guide to speed to lead covers why and how to respond within minutes, not days.

Three conversations, scored

All three are illustrative examples, not real customers.

1. The operations lead. "I run operations at a logistics company, about 60 people, and we schedule in spreadsheets. Do you integrate with our accounting tool? We'd want to start in Q1." Fit: team size 20, industry 10, role 10 = 40. Intent: integration 10, timeline 15 = 25. Total 65, warm. The agent answers the integration question from your docs and offers a call. If the visitor accepts, add 25 for intent: 90, hot.

2. The researcher. "What's the difference between your product and a CRM? I'm writing a report." Fit: unknown, 0. Intent: 0. Negative: student or researcher, minus 30. Cold. The agent answers well and points to a comparison page. No sales follow-up.

3. The security reviewer. "Do you have a DPA and where is data hosted? We're 300 people in fintech." Fit: team size 20, industry 10 = 30. Intent: security 10. Total 40, warm. A security question from a company this size often means an evaluation is underway. Send the documents and offer a call with someone who can answer follow-ups.

Collecting signals without interrogating visitors

The best scoring data comes from helping, not from asking. A few rules:

• Let visitors lead. Their questions are intent signals. A visitor who asks about pricing and integrations has told you more than any form.
• Ask two or three questions at most, after a useful answer, and only the ones your rules actually use. Team size and timeline are usually enough.
• Never ask what you can see. If the visitor already said "we're 60 people," do not ask for company size.
• Use AI to extract, not to decide. A language model is good at turning "about 60 of us, maybe more by spring" into a team size bracket. Keep the scoring rules themselves simple and readable, so sales trusts them. An AI SDR that books meetings on an opaque score will book the wrong ones.

For the question design, see how to qualify inbound leads without annoying visitors and conversational lead generation.

Let intent fade

Intent is perishable. A visitor who asked about pricing yesterday is warmer than one who asked three months ago. Give intent points an expiry: halve them after 30 days and drop them after 90, while fit points stay. HubSpot's tool works on a similar idea: its score decay "automatically reduces an individual event's score based on how long ago a scored event occurred," at intervals of 1, 3, 6 or 12 months.

Calibrate against outcomes

Your first point values are guesses. Treat them that way.

Should you replace the rules with a learned model? Not early. HubSpot's AI-built scores need at least 50 contacts in the sample, including 25 that converted and 25 that did not, before they can be generated. Most B2B sites take months to collect that many conversation leads with known outcomes. A rule set your sales team understands will do better until then.

Privacy: scoring is profiling

Scoring people to decide how to market to them is profiling. Under GDPR Article 21, people in the EU have the right to object at any time to the use of their data for direct marketing, including profiling related to it, and after an objection the data must no longer be used for that purpose. In practice:

• Score business facts, such as company size, role and timeline. Do not score sensitive personal data.
• Say so in your privacy notice: that conversations may be used to prioritize follow-up.
• Honor objections in your CRM, so a visitor who opts out of marketing is not scored into a sales sequence.

Handing scored leads to a person

A score is the start of a human handoff, not the end. Send the person who follows up the score, the reasons for it ("team of 60, Q1 start, asked about the accounting integration") and the full transcript. Reasons build trust in the model; a bare number does not.

The simplest lead scoring model that works has two axes, a handful of signals on each, a few negative rules and thresholds that each trigger a specific action. Write it on one page, let the conversations fill it in, and let real outcomes adjust it. That beats a complex model nobody on the sales team believes.