Lead Scoring Model: How to Build a Rule-Based Score Your Reps Trust

A lead scoring model is a set of weighted rules that turns what you know about a lead into one number, so reps call the best leads first. A rule-based model adds fit points for the buyer and engagement points for what they have done, against a threshold agreed with sales. In Senitix CRM, it runs as a scoring profile.

This guide builds one: where criteria and weights come from, a worked scorecard, the threshold, what a rep needs to see, and the quarterly back-test. If your team has not agreed what a qualified lead is, settle that first with one of the lead qualification frameworks. A score works on top of that agreement, not instead of it.

What goes into a lead scoring model?

Every rule-based model uses three kinds of lead scoring criteria:

• Fit criteria describe the company and the person: industry, employee count, annual revenue, region, job title. They answer “would we want this customer?” and rarely change.
• Engagement criteria describe what the lead has done: requested a demo, attended a webinar, replied to a rep. They answer “are they looking now?” and they go stale.
• Negative criteria subtract points for signs a lead is unlikely to buy, such as a personal email address or a business too small for your product.

Keep fit and engagement readable as two parts, even when the CRM shows one number. A high-fit lead with no engagement is a prospecting target; a low-fit lead with heavy engagement is often a consultant doing research. HubSpot draws the same line in its scoring knowledge base: the tool keeps separate fit, engagement and combined scores, labeling the combined score from A1 to C3, where the letter is the fit grade.

Rule-based vs. predictive lead scoring: which should you build first?

A rule-based model adds up points you chose. A predictive model learns its weights from past leads and their outcomes, using statistics or machine learning. Both produce a number; they differ in what they need and what you can explain.

• Rule-based scoring works from the first week, and anyone can read why a lead scored 65. Its weakness: the weights are an opinion until you test them.
• Predictive scoring can find patterns nobody guessed, but it needs enough converted and lost leads to learn from, and it is harder to tell a rep why one lead outranks another.

For most small and mid-size B2B teams, rules come first. A few quarters of back-tested rules show which signals matter and leave you with consistently filled fields, which any learned model needs anyway. Our guide to AI in a CRM covers what a learned score can and cannot tell you, and automation vs. AI covers the wider choice between writing a rule and training a model.

How do you build a lead scoring model?

Six steps, each with an owner and a written output:

1. Pull last year’s leads with their outcomes. Owner: sales ops. Export the last twelve months of leads with source, fit fields and result (converted, won, lost, disqualified). Output: one row per lead.
2. Find the attributes that separate buyers. For each candidate criterion, compare its won-deal rate with the rate across all leads. HubSpot’s lead scoring guide describes this method: an attribute that closes well above your overall rate earns points in proportion. Output: a ranked list.
3. Choose six to ten criteria. Three to five fit, two to four engagement, one or two negative. Drop a field that is blank on most old leads: start recording it and test it next quarter. Output: the draft scorecard.
4. Give each criterion a coarse weight. Three levels, strong, moderate and negative, are easier to explain than a scale of one to a hundred. Output: points per criterion.
5. Agree on the threshold with sales. Owner: the sales manager and the marketing lead. Output: a written rule for what happens above and below the line.
6. Publish the scorecard and open a change log. Owner: sales ops. Output: one page every rep can read, and a dated record of every weight change and why.

Example: a lead scoring scorecard for a B2B SaaS team

Example: a hypothetical 12-rep B2B SaaS team in Austin, Texas, sells scheduling software to outpatient healthcare practices. Four of its reps are SDRs working inbound leads. The weights are illustrative, not a benchmark; yours come from step two.

The positive criteria add up to 100, so the score reads as a share of the ideal lead. Two leads:

• Lead A: an operations director at a physical therapy group with 180 employees and 12 locations, who attended a webinar last month. Score: 25 + 10 + 10 + 10 + 10 = 65.
• Lead B: a practice manager at a single-location dental office with eight employees, who requested a demo from a personal Gmail address. Score: 25 + 10 − 10 + 25 − 10 = 40.

Lead B shows the common trap: the clearest intent on the card, yet a lower score than Lead A. Agree up front that anyone who asks for a demo or a price gets a same-day reply whatever the score: points decide which of ten Monday demo requests gets called first, not whether it gets called.

How do you set a lead scoring threshold sales will accept?

Start from rep capacity, not from the score. The threshold is the line above which a lead gets a rep’s time within an agreed window, so the leads above it must fit the conversations the team can hold.

Example: on top of their follow-ups, the Austin team’s four SDRs can each work about 20 new leads a week, 80 in total. Marketing brings in around 240 leads a week. Scored with the new card, about a quarter of last year’s leads land at 60 or above, roughly 60 a week, which leaves room for hand-raisers below the line. The threshold is 60.

Then write the agreement down:

• Above the threshold: first contact within one business day, an outcome recorded within five.
• Below the threshold: the lead stays in marketing’s nurture program. A rep can pull any lead up by hand, and notes why.
• Hand-raisers: demo and pricing requests get a same-day reply.
• Changes: weights and threshold are reviewed once a quarter, after the back-test, not whenever someone dislikes a score.

When reps keep pulling up the same kind of lead, the model is missing a criterion.

What does a lead score need to show a rep?

Reps ignore a number they cannot explain. A score earns trust when the rep can answer four questions from the lead record:

1. Which criteria fired. “Outpatient healthcare +25, demo request +25, personal email −10” is worth more than “40”.
2. When it last changed. Tell reps when the score recalculates (on every change, overnight, or on refresh), since a lead that replied yesterday but still shows last week’s score reads as broken.
3. How much is fit and how much engagement. A 60 made of fit is a prospect to research; a 60 made of engagement is a call to make today.
4. What it is not. A score ranks leads. A 70 is not a 70-in-100 chance of closing, and it does not belong in the forecast.

A useful test: when a rep can guess a lead’s score before opening it, the model is understood. When a rep is surprised, a field is wrong or a weight needs a conversation.

How do you back-test a lead scoring model each quarter?

Once a quarter, score last quarter’s leads with the current model, group them into bands and count how many in each became deals. Conversion should climb band by band.

Example (illustrative numbers), the Austin team’s leads from last quarter:

• 80 to 100: 200 leads, 60 became deals, 3 in 10.
• 60 to 79: 580 leads, 116 became deals, 1 in 5.
• 40 to 59: 900 leads, 75 became deals, 1 in 12.
• Below 40: 1,440 leads, 29 became deals, about 1 in 50.

The bands rise and the threshold sits where conversion jumps, so the model holds. Then run three checks:

1. Do the bands rise? If 60 to 79 converts no better than 40 to 59, the threshold is misplaced or the middle criteria are noise.
2. Does each criterion still earn its points? Repeat step two. A criterion that no longer separates buyers loses its weight.
3. What did the model miss? Read the won deals that scored below the threshold. An attribute several of them share is your next criterion.

Change two or three weights a quarter at most, logged with the date and reason. Change everything at once and the next back-test cannot say which change helped.

What mistakes make reps stop trusting a lead score?

• Scoring what reps do. Points for “call logged” rise because a rep dialed, so the score rewards whoever calls most.
• Letting engagement pile up. Last year’s webinar is not interest today. Give every engagement criterion a time window, or use decay where your tool offers it; HubSpot’s knowledge base, for one, lets engagement points decay every 1, 3, 6 or 12 months.
• Scoring fields nobody fills in. If annual revenue is blank on most leads, a revenue criterion ranks research effort instead of fit.
• Counting one thing twice. Employee count and annual revenue often measure the same company size.
• Silent changes. A weight changed without notice makes yesterday’s scores incomparable with today’s, and reps notice first.

How do you build a rule-based lead score in Senitix CRM?

In Senitix CRM, lead scoring is rule-based. An administrator builds a scoring profile for leads, contacts or deals from conditions on the record’s fields: standard fields such as source, industry, annual revenue, number of employees and title, and custom fields such as a picklist, a number or a checkbox. Each criterion is one or more conditions, weighted highly positive, slightly positive or negative, and the result shows on the record as a score from 0 to 100.

A lead is scored when it is created and rescored when it is updated, and leads not yet converted are rescored when an administrator changes the active profile. One lead profile is active at a time, so a new version of your scorecard replaces the old one rather than running beside it.

Two things follow. A profile reads fields, not a feed of website visits or email opens, so engagement counts once it is recorded on the lead: a “Demo requested” checkbox, a webinar picklist, a status field a rep resets when a lead goes quiet. A time window like “in the last 90 days” lives in that field, since a condition compares a field with a value rather than a rolling clock. And the score follows the rules you wrote, not a prediction: Senitix AI drafts email, summarizes records and suggests a next action for a person to confirm, and never sets a score.

The lead features in Senitix CRM list what a lead record holds; plan details are on the pricing page.

Frequently asked questions

Do you need a separate lead scoring model for each product or segment?

Only when the buyer is different. A second product sold to another department, or an add-on sold to existing customers, may need its own criteria. Otherwise keep one model and add segment criteria such as industry or company size: two scores on one lead leave reps unsure which number counts. In Senitix CRM, one lead profile is active at a time, so segments are criteria inside it, not a second model.

Can you score leads without marketing automation data?

Yes. Fit criteria need only the fields on the lead: industry, size, title, source. Engagement can be recorded by reps as fields instead: a demo request, an event attended, a reply received. You lose automatic signals such as page visits, but gain a score every rep can explain, and the quarterly back-test works the same way.

What is the difference between a lead score and a deal probability?

A lead score ranks people and companies who have not become deals yet, to decide who gets a rep’s time first. A deal probability estimates how likely an open deal is to close, usually set by stage, and it feeds the forecast. They answer different questions, so keep them in different fields: see what a CRM is and how it works for how leads, contacts and deals fit together.