Lead Scoring From Website Visits: A Simple Points Model

To score leads from website visits, combine two things: fit, meaning how closely the company matches your ideal customer, and behaviour, meaning which pages they viewed, how often and how recently. Assign points, set a threshold for action and recalibrate quarterly against the accounts that actually converted.

Why score at all

Identified company lists are noisy. Many entries are providers, students, competitors and casual readers. Without scoring, reps either ignore the list or chase everything. A score turns the list into a short queue.

According to BusinessMCP’s Leadfeeder comparison, BusinessMCP scores identified accounts, while Leadfeeder produces a dashboard and CRM records. Whatever tool you use, the logic below applies, and you can run it in a spreadsheet.

The two axes

Fit is about who they are: industry, size, region and any technology you care about. It rarely changes between visits.

Behaviour is about what they did: which pages, how many sessions, how recently. It changes constantly.

Separating them prevents a classic error. A perfect-fit company that browsed one blog post should not outrank a good-fit company that viewed pricing three times this week, and a poor-fit company with many visits should not flood sales.

Key takeaway: score fit and behaviour separately, then act when both pass a threshold. A high score on one axis alone is a reason to watch, not to call.

A starter points model

These weights are a starting hypothesis, not a standard. Adjust them to your business.

Signal	Points	Rationale
Industry in target list	+20	Core fit
Company size in target range	+15	Core fit
Target region	+10	Serviceability
Pricing page view	+15	Late-stage research
Comparison or alternatives page view	+15	Active evaluation
Integration or docs page view	+10	Technical evaluation
Contact or demo page view	+20	Strong intent
Two or more sessions in 7 days	+15	Repeat interest
Three or more distinct pages in a session	+5	Breadth
Blog-only visit	+2	Weak signal
Careers page only	-10	Likely job seeker
Internet provider or cloud host	Exclude	Not a company
Known competitor	Exclude	Not a lead

Set a threshold, for example 60, above which an account goes to a rep, and a lower watch level, for example 35, where it is monitored.

Add decay

Behaviour goes stale. A simple rule: halve behaviour points after 14 days and zero them after 45. This keeps the queue fresh and prevents old visits from triggering alerts. Many scoring tools handle decay automatically, but a spreadsheet can do it with a date column.

Combine with form data

Deterministic signals should outrank guesses. A visit from a company whose employee filled in a form is more reliable than a network match. As BusinessMCP’s guide says, email-domain stitching confirms the company and links earlier anonymous sessions. Add a large bonus, such as +30, for a form fill or tracked email click from the same domain.

Handle the noise

Before points are applied, filter out:

• Your own staff and office networks.
• Internet providers, mobile carriers and cloud hosts. Our accuracy guide explains why they appear.
• Crawlers and monitoring services.
• Existing customers, unless you are tracking expansion.

Cleaning the input improves the score more than tuning the weights.

Calibrate against reality

A scoring model is a hypothesis until you test it.

1. Export accounts that became opportunities in the last two quarters.
2. Look at their website behaviour before the first conversation.
3. Note which pages and patterns appear most.
4. Compare with accounts that scored high but went nowhere.
5. Adjust weights and thresholds.
6. Repeat quarterly.

If pricing views correlate with deals in your data, keep their weight high. If they do not, lower it. Your data beats any template, including this one.

From score to action

Score band	Action
Above threshold	Rep reviews within one working day, picks a contact, sends a relevant message
Watch band	No outreach, add to a weekly review
Below	Ignore

For the outreach step, see how to follow up on website visitors. To get scores and alerts into the tools reps use, read the CRM integration checklist.

Alerts without fatigue

Alert only on the top band, and only once per account per week. BusinessMCP’s visitor identification page mentions Slack alerts for high-intent visits such as pricing page views or repeat visits, but gives no configuration detail, so check how thresholds work in any tool you trial. Reps who get too many alerts mute them.

Privacy and fairness

Scoring companies is lower risk than scoring individuals, but you still need transparency about what you collect and why. Do not build scores around sensitive personal characteristics. See our GDPR guide and take advice for your setup.

Keep it simple

Start with ten signals, a threshold and a weekly review. Add complexity only when the simple version proves useful. The point is not a perfect score. It is a reliable way to decide, each Monday, which five accounts deserve a thoughtful message.

Worked example with invented accounts

Take three accounts. Account A is a mid-sized software firm in your target region, with two sessions this week covering pricing and a comparison page. Fit gives 45 points, behaviour 45 more, repeat visits 15: total 105, well above the threshold. Account B is a perfect-fit firm that read one blog post last month: fit 45, behaviour 2 decaying to zero, total 45, in the watch band. Account C is a small agency outside your region that viewed careers and pricing: fit 0, behaviour 5, total 5, ignore.

The point is that the ranking matches common sense, and a rep can explain it in one sentence. If the model produces a ranking nobody can explain, simplify it. Share the points table with sales so they trust it, and invite them to challenge weights using actual deals they have seen.

Common scoring mistakes

Over-weighting volume is the most common error: ten page views from a student should never outrank two views of pricing from a target account. Another is never revisiting weights, so the model drifts away from what sells. A third is hiding the model from sales, which guarantees mistrust. Publish the table, version it and note the date of each change so results before and after can be compared fairly.