Lead scoring: how to build a model your sales team will trust

Lead scoring gives each lead, usually a person who has engaged with you, a number that reflects how likely they are to become a customer, so sales time goes to the most promising leads first.

Points come from two kinds of evidence. Fit says whether the person and their company look like your customers: industry, company size, role. Engagement says what they have done: asked for a meeting, visited the pricing page, replied to an email, attended a webinar. When a lead passes a set score, it moves from marketing to sales. The stages that follow, from marketing qualified lead to sales qualified lead, are covered in how to qualify leads.

Most lead scoring models fail the same way: they reward attention instead of need. A student who downloads every guide outscores a director who asked one pointed question. The model below is built to stop that.

• Key takeaway: gate on company fit before you count any engagement.
• Key takeaway: weight actions by how often they led to deals, not by how busy they look.
• Key takeaway: engagement should cap and fade; fit should not.

A lead scoring model adds points for fit and engagement, subtracts or disqualifies for warning signs, and compares the total with a threshold that decides who sales calls first.

Many tools keep fit and engagement apart. HubSpot’s lead scoring tool, for example, offers fit, engagement and combined scores, and labels combined scores A1 to C3: the letter reflects fit, from A (high) to C (low), and the number reflects engagement, from 1 (high) to 3 (low). Two axes are easier to act on than one total, because the fix is different: a low-fit lead needs rerouting, while a high-fit, low-engagement lead needs outreach.

The model below scores out of 100: up to 30 points for company fit, 20 for role and 50 for engagement, after a gate that removes companies you cannot serve.

Fictional sample. The seller is a fictional, unnamed plant-automation consultancy selling to manufacturers. Copy the structure and replace the criteria with your own.

The handoff rule: 60 or more, with role fit of at least 12, goes to sales the same working day; 40 to 59 gets a researched, personal follow-up from an SDR; under 40 stays in marketing. Now four fictional leads, scored first by engagement alone and then by the full model.

Engagement alone ranks the leads A, B, C, D, and the first call goes to a company you cannot serve. The full model ranks B and D at the top and keeps A and C out of the sales queue. Decay matters too: if lead B’s activity were 45 days old, its engagement would halve to 14 and its score would fall to 46, a follow-up rather than a same-day call.

Lead D shows the limit of any lead score. Ruth sits one point under the line, yet Halvorsen Castings has a county permit dated Sep 12, 2026 for a 140,000 sq ft expansion in Dayton. Nothing in her clicks says that. A dated event at the company is account timing, which is what account scoring adds.

Build a lead scoring system from your own won and lost deals: find the traits and actions that came before wins, weight them in proportion, set a threshold and review it quarterly.

1. Export a year of leads with their outcome: became an opportunity, or did not.
2. Write your company fit rules as must, prefer and exclude, and use the must rules as the gate.
3. For each action, compare the share of leads who took it and became opportunities with the share across all leads.
4. Give points in proportion to that lift. An action with little or no lift earns little or nothing, however common it is.
5. Add disqualifiers and decay, then cap engagement so many small actions can’t outweigh one strong one.
6. Set the threshold by sales capacity: how many new leads a week the team can follow up the same day.
7. Agree the handoff in writing: who follows up, by when, and how sales returns a lead with a reason.
8. After a quarter, compare scores with opportunities and wins, and re-weight.

In this sample, a guide download converts below the base rate, so it earns a token score and a cap. Many teams score downloads highly because they are common; the data says they are a sign of interest, not intent. The weights in the example model above follow this pattern.

AI lead scoring, also called predictive lead scoring, learns the weights from your past conversions instead of having a person set them, and updates them as new outcomes arrive.

It needs history to learn from. HubSpot, for example, requires a Marketing Hub Enterprise subscription to build contact lead scores with AI, and a minimum sample of 50 contacts, 25 converted and 25 not, to generate one (Knowledge Base, updated Sep 2, 2026). A newer form uses AI agents that read about each company, such as its news, filings and job posts, and score it against written rules; that works before anyone has engaged, and is closer to account scoring than lead scoring.

The decision rule: start with rules. Switch to a predictive model only when you have the history it needs, and run both side by side for a quarter before you trust it. A predictive model trained on your past can’t score the market you never sold to, and it can mark a new segment as poor simply because you never tried it.

Lead scoring ranks people by fit and engagement; account scoring ranks companies by fit, timing and reach, so it works before anyone at the company has engaged.

B2B purchases are made by groups. Forrester reported in January 2026 that a typical buying decision includes 13 internal stakeholders and nine external influencers. A lead score sees one of those people at a time. Three modest leads from the same company in the same month can matter more than one hot lead, so roll engagement up to the company as well. The account model, with its weights, tiers and worked numbers, is in account scoring, and a copyable version is the account scoring matrix template.

The best lead scoring tool is usually the one already holding your lead data, your CRM or marketing automation platform; add a predictive or research tool only for a gap it can show it fills.

Whatever the tool, ask the same three questions: can a seller see why a lead scored what it did, can your team change the weights, and can you compare scores with outcomes? Intent-based scores deserve one more: what the score is built from, which intent data vs buying signals explains. For predictive platforms side by side, see 6sense alternatives.

Most lead scoring mistakes make the score louder than the evidence: points for activity that never led to deals, no decay, and no way for a seller to see why a lead ranked high.

• Scoring engagement before fit, so the busiest leads reach sales whatever their company.
• Giving points for email opens, which many mail apps trigger automatically.
• No cap, so ten small actions outweigh one meeting request.
• No decay, so last spring’s webinar still counts today.
• One combined number, which hides whether fit or engagement is missing.
• A threshold set once and never checked against won deals.
• Scores sales can’t explain, so reps ignore them and work their own lists.

Use this checklist when you build a lead scoring model and again at each quarterly review.

• Company fit rules are written as must, prefer and exclude, and the must rules gate the score.
• Role fit separates the person who owns the problem from people who influence it.
• Every engagement point traces to a conversion rate from your own deals.
• Engagement is capped, and points fade after a set number of days.
• Disqualifiers remove competitors, students, job seekers, vendors and anyone who asked not to be contacted.
• The threshold matches how many leads sales can follow up the same day.
• Engagement from several people at one company rolls up to the account.
• Sellers can see the reasons behind every score.
• Sales returns rejected leads with a reason, and someone reads them monthly.
• Scores are compared with opportunities and wins every quarter, and the weights change when they disagree.

SalesOne scores accounts, not leads: S1 deep research agents check each company against your ideal customer profile, score every qualified account on fit, timing and reach, and show the reasons beside the score.

That covers the companies a lead score can’t see, the ones that have not engaged yet. S1 agents read SEC filings, press releases, company sites, trade press and public records across 1,000+ data sources. In October 2026 production runs they checked 1,051 companies and kept 249; all 802 companies turned down kept the reason they failed. All 375 buying signals link their source, and 98% are dated to the day. About four people are mapped per account, 95% with a dated source.

A review pass, on by default, re-checks every buying signal against its source before delivery, and nothing is researched twice for a client. A person approves every message, and every approval is recorded with who gave it. Keep your lead score for inbound; use the account score to decide which companies are worth the work.