Lead scoring: how to build a lead scoring model

Lead scoring is a points system for leads. You decide which facts about a lead make it more or less likely to become a customer, give each one points, and add them up. A Director of Operations at a 300-person company who asked for a demo ends up near the top. A student on a Gmail address who downloaded one ebook ends up near the bottom. Sales works the list from the top.

Scoring and qualification get used as if they were the same thing. A score ranks leads. Qualification is the yes or no decision about whether a lead is worth a salesperson's time, made partly from the score and partly from a conversation. Our lead qualification how-to covers that decision, with BANT and the questions to ask.

Lead grading is a third word you'll meet, mostly from Salesforce's Account Engagement (formerly Pardot). There, the score measures interest from what a prospect has done, and the grade, from A+ to F, measures how well they match the profile you sell to. HubSpot, for one, shows both: fit and engagement scores, apart or combined.

Fit scoring, also called explicit scoring, reads what the lead or the data tells you about them: job title, seniority, company size, industry, country, the tools they use. It answers "is this the kind of person and company we sell to?" Fit barely changes over time. A VP at a 500-person logistics firm is the same fit next month.

Engagement scoring, also called implicit or behavioral scoring, reads what the lead has done: pages visited, emails opened and clicked, forms filled, webinars attended, a demo booked. It answers "are they interested right now?" It needs tracking (a marketing automation tool, website analytics) and it goes stale, which is why engagement points usually decay.

A combined model keeps the two apart and reads them together. HubSpot does this with labels from A1 to C3: "The letters refer to the fit score values where A is high-fit and C is low-fit. The numbers refer to engagement score values, where 1 is high engagement and 3 is low engagement." An A3 is the right company that hasn't done much yet, so marketing nurtures it. A C1 is very active but the wrong fit, so it may be a student or a competitor doing research.

Everything above is rules-based: a person picks the rules and the points. Predictive scoring, sold as AI lead scoring, works the other way round. A model looks at which past leads converted and which didn't, finds the patterns itself, and scores new leads against them. It can find signals nobody thought to write a rule for. It needs history to learn from, though: Salesforce's Einstein Lead Scoring wants at least 1,000 leads created in the last 200 days, at least 120 of them converted, and HubSpot's AI scores need at least 50 contacts, 25 converted and 25 not. Below that, Einstein falls back to a global model built from many Salesforce customers' anonymous data, which knows nothing about your market.

So for most small and mid-sized teams, start with rules. You can explain every point to a sales rep who asks why a lead scored 85, and you can change a rule the day you learn it's wrong. Move to predictive once you have a year or two of clean conversion data.

Every scoring model assumes the fields it reads are right. Often they aren't, and a bad record scores as well as a good one. Salesforce says as much about its own model: "Encourage reps to add as much data to their leads as possible. When leads have more data, Einstein Lead Scoring generates better insights." The same goes for rules you write yourself.

Say a demo request comes in from a Head of Revenue at a 400-person software company. It scores 90 and goes straight to a rep, who calls and finds the company has been a customer for three years. The person filled in the form with a personal address, so the CRM made a new contact instead of matching the one it had. Nothing in the score could have caught that. A lookup against the CRM before scoring would have.

By hand, that means an export of your CRM's contacts and companies, a lookup of each new lead's email and domain against it, an email checker, and a duplicate check on the list. Our CRM data quality how-to covers the wider cleanup.

Run these checks first:

Seven steps. The first two are research, the middle three are the model, and the last two decide what the score does and keep it honest.

Here are B2B examples for a company selling software to mid-sized businesses. The points are a starting place, not a standard: change them to fit what your won deals showed.

Fit criteria:

Engagement criteria, from the same company:

Nadia Ferreira, Head of Sales Operations at Larkspur Freight (220 employees, logistics, US), requested a demo after visiting the pricing page twice and opening two emails, from nadia.ferreira@larkspurfreight.example. Fit: Head of +25, 220 employees +20, logistics is an industry with customers +15, US +10, no phone 0, so 70. Engagement: demo request +30, pricing page +15 (counted once), two email opens +2. Total 117, so the lead goes to sales the same day.

Compare a lead with the same demo request from a student at a university on a Gmail address: demo request +30, student -30, free email domain -15, so -15. The demo request alone didn't carry them, which is the point of negative scoring.

Negative points do as much work as positive ones. Without them, a lead that does a lot of reading looks like a buyer. Common B2B negatives:

Score decay takes engagement points away as time passes, so a lead who was busy in March doesn't still look hot in September. HubSpot's version: "Score decay intervals can be set to every 1, 3, 6, or 12 months", and each event decays on its own, so a 10-point form fill set to lose 50% a month is worth 5 points a month later. If your tool has no decay, a simpler rule works: only count actions from the last 90 days.

Fit doesn't decay. A Director at a 300-person company is the same fit in a year, unless they change jobs. What goes stale there is the data: people move roles and companies grow, so re-check fit fields on old leads before you score them again.

Each CRM has scoring built in, but on different plans and for different records. Facts from each vendor's own help pages, read October 21, 2026 (Salesforce on October 20).

HubSpot: lead scoring is listed for Marketing Hub Professional and Enterprise and Sales Hub Professional and Enterprise; Starter isn't listed. Contacts take scores with Marketing Hub, companies with Marketing Hub or Sales Hub, and deals with Sales Hub, where scores are combined only. Engagement and fit scores are labeled High, Medium and Low, combined scores A1 to C3, and rules can "add or subtract points". AI scores need Marketing Hub Enterprise.

Salesforce: Einstein Lead Scoring is "available in Performance and Unlimited Editions, and for an extra cost in Enterprise Edition", needs the data volumes in the models section above, and "reanalyzes your lead data every 10 days and refreshes your scores". Rules-based points and A+ to F grades are in Account Engagement, a separately licensed product. Its default points include +50 for a form submission, +1 for a page view and -5 for a form error.

Pipedrive: Scores are "available on Premium and higher plans", and "currently, you can only choose deals", not leads in the Leads Inbox or people. Criteria fall into three groups worth fixed points (Highly positive +25, Slightly positive +10, Negative -10), you can "set up to 10 criteria in total", and only deal and activity fields can be used. The result is shown on a scale of 0-100.

Side by side:

If your plan has no native scoring (HubSpot Starter, Salesforce without Einstein or Account Engagement, Pipedrive below Premium, or Pipedrive for anything before it's a deal), you have two routes. Keep a score in a number field you set with a formula or a workflow your plan allows, or score the leads before they reach the CRM.

CRM scoring starts once the record exists. By then a junk lead is already a contact, maybe with a task on a rep's list. Scoring at intake means the score is set while the lead is on its way in, so a lead that scores 1 can be held back before it becomes a record anyone has to deal with. It also covers what CRM scoring often misses: event lists, bought lists and spreadsheets someone imports.

In Operelio, that's a trail's Scorecard. You give each rule points, from -100 to 100, and the total falls into a score from 1 to 5, at the point bands you choose. Rules can match words in the job title, check that a column has a value, check that the email verified, spot a free email domain, or test any column with the same conditions as the filter step, such as "company size is at least" 50. Bridgeant can suggest title words and columns from your column names, never from the values in your cells, and you approve each rule.

Each score has an action: keep it and send it on, send it to review, where someone says yes or no, hold it in Operelio, drop it, or hand it to the trail's routing rules. By default 1 and 2 are held, 3 waits for review, and 4 and 5 go on. Routing rules can also send a lead on by its score, for example 5 to a senior rep and 4 shared out in turn.

Lists and single leads are scored the same way, with one difference in what runs before the score. A list can go through Verify emails, Company Active Check and Deduplicate first. A single lead sent to a folder's webhook (from Zapier, Make or your own code), straight from Typeform, or by email runs the trail's own steps and the Scorecard but none of those three, so an email verified rule adds points only on lists that ran Verify emails. Typeform can send each response straight to a folder, and each response becomes one lead, so a test response from Typeform makes a real lead too. A test run on one of your own files shows what the Scorecard did with the rows, how many were kept, sent to review, held or dropped, before the trail is live.

For single leads into HubSpot or Salesforce, each score can set the lifecycle stage or the Lead Status, and if you turn deals on, one is made for scores 4 and 5 by default, never below 3. A HubSpot lifecycle stage only moves forward, and a Salesforce Lead Status is set only on a Lead the push creates. Pipedrive takes lists from a trail, not single leads yet.

What the Scorecard doesn't do: it scores the columns a row arrives with, so it's fit scoring plus whatever a column carries, such as a Source or Event column. It doesn't see page views, email opens or form history, it has no decay and no predictive model, and it doesn't write the number into a CRM field. So the honest split is this: score fit at the door with the Scorecard, and let your CRM's own scoring add engagement once the lead is in.

Usually not at first. If your CRM plan includes scoring and your leads arrive through forms the CRM tracks, its own tool is the right place: it sees the engagement, and the score sits on the record where reps work.

A separate tool earns its place in three cases. Your plan has no scoring and the upgrade costs more than the problem. A lot of your leads arrive as files and inbox emails, which CRM scoring sees only after they've been imported. Or you want bad leads stopped before they become records, not ranked low once they're in. In the first case a spreadsheet model is a fine start; in the other two, score at intake.