Lead scoring: what it is, how it works, and how to build a model
A plain-English guide to lead scoring for B2B teams: the definition, the two halves of every score, the types of model, a worked example, and a seven-step build.
Lead scoring is a way to rank your leads by how likely they are to become customers, so your team knows who to work first. Each lead gets a number built from two kinds of evidence: how closely the person and their company match the customers you win (fit), and what they have done recently that suggests interest (engagement). Done well, it turns a long, unsorted list into an ordered one, with a reason behind every position.
What is lead scoring?
Lead scoring is the practice of assigning each lead a score based on weighted factors, such as company size, industry, job role, and recent activity, so the leads most likely to buy rise to the top. Marketing, sales, and RevOps use the score to decide who to call first, who gets routed to sales, and who stays in nurture.
A lead scoring model has three parts:
- Factors: the things the model looks at, such as company size, job level, or pricing page visits.
- Weights: how much each factor counts toward the total.
- Rules for action: the thresholds and bands that turn a number into a next step.
If you cannot list those three things for your current score, you do not have a model yet. You have a number.
Why lead scoring matters
Most B2B teams have more leads than time. Without a score, reps work the newest lead, the loudest one, or the one with a familiar logo, and the best-fit buyer can sit untouched for a week. A working lead score fixes three things:
- Speed. Reps reach the right leads first, while interest is still current.
- Alignment. Marketing and sales agree, in writing, on what a good lead looks like before any handoff.
- Focus. Low-fit leads go to nurture instead of a rep’s calendar, and disqualified leads stop wasting time at all.
It also matters more than it used to. Buying is a group decision: Forrester found that 95% of B2B purchases involve three or more people. And more of the work that acts on a score, such as routing, sequencing, and prioritizing, is now done by automation and AI agents. A score that is wrong, or that nobody can explain, gets acted on faster and more often than ever.
How lead scoring works: fit and engagement
Every useful lead score is built from two different questions. Keeping them apart is the single most important design choice you will make.
| Fit score | Engagement score | |
|---|---|---|
| The question it answers | Does this lead match the customers we win? | Is this lead active right now? |
| Typical factors | Company size, industry, region, job role, seniority, systems they run | Pricing page visits, demo requests, webinar attendance, email replies, product use |
| How fast it changes | Slowly, when the company or role changes | Quickly, with every visit and every week that passes |
| Should it decay? | No | Yes, older activity should count for less |
| What it tells a rep | Whether the lead belongs on the list at all | Whether now is the time to reach out |
When fit and engagement are blended into one number, they hide each other. A student who reads every blog post can outscore a director at a perfect-fit company who visited your pricing page once. Kept apart, the two scores tell you something useful together:
- Fits and is active: work it now.
- Fits, but not active yet: the right company at the wrong moment. Nurture it, or reach out with a reason to talk.
- Active, but does not fit: someone interested who probably will not buy. Answer the question, then move on.
- Neither: leave it in nurture.
Many CRMs combine the two in a grid rather than one number. HubSpot’s lead scoring tool, for example, can show fit as a letter and engagement as a number, giving combined bands from A1 to C3.
A lead scoring example
Here is a demo lead scored on fit, on a /10 scale, with every factor and weight visible. One factor subtracts: the lead works at an agency that resells software, a pattern this team has learned rarely buys for its own use.
- Company size (200 to 1,000 employees) 30% 8.7
- Industry match 25% 9.1
- Role and seniority 25% 7.4
- Runs HubSpot 15% 6.3
- Agency or reseller domain 5% -3.2
The math is simple on purpose. Each factor’s value times its weight gives its share of the score: 8.7 at 30% is 2.61, 9.1 at 25% is 2.275, 7.4 at 25% is 1.85, and 6.3 at 15% is 0.945. That adds to 7.68. The disqualifying factor subtracts 3.2 at 5%, or 0.16, for a fit score of 7.5.
The same lead’s engagement score reads its recent activity:
| Signal (last 90 days) | Counts toward engagement | Why |
|---|---|---|
| Visited the pricing page twice | Strongly | The clearest sign of buying research |
| Attended a product webinar | Moderately | Real time spent, but early stage |
| Replied to a sales email | Strongly | A person, not a pixel |
| Opened six marketing emails | Not at all | Opens are inflated by privacy features (see below) |
| Read three blog posts four months ago | Barely | Old activity, mostly decayed |
A rep looking at this lead sees a 7.5 fit, recent pricing and reply activity, and one small subtraction they can explain on the call. That is a lead worth working today, and the rep knows why.
Types of lead scoring models
Lead scoring models differ mainly in who chooses the factors and weights, and whether you can see them.
| Rules-based | Predictive | AI scoring in the CRM | Transparent, human-built | |
|---|---|---|---|---|
| Who sets factors and weights | Your team | A model trained on past outcomes | The CRM vendor’s model | Your team, with AI assisting |
| Can a rep see why a lead scored what it did? | Yes | Rarely | Partly | Yes, on every record |
| Needs a lot of history? | No | Yes | Yes, for the AI version | No |
| Easy to change? | Yes | Retrain the model | Within the tool’s options | Yes |
| Main risk | Only knows what you tell it | A number nobody can explain | Plan limits and data minimums | Needs someone to own the model |
Rules-based scoring is the classic approach: “director or above adds points,” “competitor domain sets fit to zero.” It is readable and easy to change, but it is only as good as the rules you write.
Predictive scoring trains a statistical or machine learning model on which leads converted in the past. It can find patterns people miss, but it needs enough clean history, and it can be hard to explain any one score. Salesforce’s Einstein Lead Scoring, for example, needs at least 1,000 leads and 120 conversions to build a model from your own data.
AI scoring inside the CRM brings predictive scoring into the tool your team already uses. It is convenient, but features depend on your plan, and the AI version still needs history to train on.
Transparent, human-built scoring keeps what works in each: people choose and own the factors and weights, AI helps analyze data and suggest a starting point, and every score shows its reasons on the record.
Lead scoring criteria: what to score
Start with fewer factors than you think you need. Three to six fit factors and a handful of engagement signals beat a long list, because every extra factor dilutes the ones that matter.
Fit factors (who they are):
- Company size, by employees or revenue, in the band where you win
- Industry or sub-industry
- Region or market you serve
- Job function and seniority of the person
- Systems the company runs, such as the CRM you integrate with
- Company stage or recent changes, such as funding or new leadership
Engagement signals (what they do):
- Pricing page and comparison page visits
- Demo or contact requests
- Replies to sales emails, and meetings booked
- Webinar or event attendance
- Product sign-ups and usage, for product-led motions
- Repeat visits from several people at the same company
Disqualifiers (what subtracts):
- Competitors and their employees
- Students and job seekers
- Personal email domains, if you only sell to businesses
- Companies outside your market or below a minimum size
- Existing customers, in a model meant for new business
A note on email opens: do not score them. Litmus reports that over 50% of email opens come from devices with Apple Mail Privacy Protection turned on, which loads tracking pixels whether or not anyone reads the email. Score clicks and replies instead.
Your ideal customer profile is the source for your fit factors. If you have not turned it into weighted factors yet, start with our five-step framework for building an ICP scoring model.
How to build a lead scoring model in seven steps
- Pull your last 12 to 24 months of won and lost deals. Look for what the wins have in common that the losses do not. This is your evidence, and it beats opinion.
- Agree on definitions with sales. Write down what a qualified lead is, in plain words, before you assign a single weight. Most scoring arguments are definition arguments.
- Pick three to six fit factors and weight them. Weights add up to 100%. Give the most weight to what most clearly separates wins from losses, and write down why.
- Pick your engagement signals and set decay. Decide how fast each signal fades. A demo request stays relevant longer than a blog visit. Cap how much any one signal can add.
- Add disqualifiers. Decide what subtracts, and what removes a lead from the list entirely.
- Test it on real records before you switch it on. Score a sample of past leads and check that the ones that became customers land near the top. If they do not, change the weights, not the threshold.
- Set thresholds and bands, then the play for each. A threshold is a cutoff that fires an action, such as routing to sales. A band is a range treated the same way, such as every lead scoring 8 and up. Give each band a clear next step and a response time.
Lead scoring best practices
- Keep fit and engagement separate, and combine them in a grid rather than one blended number.
- Let engagement decay. Without decay, a lead who was active once last year outranks one who is active today.
- Cap repeat signals. Ten blog visits should not outscore one pricing page visit.
- Score the buying group, not just the person. Several engaged people at one company are a stronger signal than one very busy contact. That is where account scoring takes over from lead scoring.
- Show the reasons on the record. Reps trust a score they can check, and they can use the reasons on the call.
- Treat missing data as zero, not a penalty. An empty field should add nothing. If a heavily weighted factor is often empty, fix the data or lower the weight.
- Review conversion by band every quarter. Change one or two weights at a time, and keep a note of what changed.
Most broken models fail in the same few ways, and our guide to where teams get ICP scoring wrong covers the fit side in detail.
How to tell if your lead scoring works
There is one test that matters: do the leads you score highest convert better than the rest? Measure conversion by band, such as the share of each band that reaches a meeting, an opportunity, and a closed deal.
| What you see | What it usually means | What to do |
|---|---|---|
| Top bands convert clearly better than lower ones | The model works | Keep reviewing quarterly |
| Every band converts about the same | The weights do not separate good leads from bad | Go back to your won and lost deals and reweight |
| Sales rejects most of the leads you route | The threshold is too low, or fit is underweighted | Raise the threshold or the weight on fit |
| Good leads sit in low bands | A key factor is missing, or its data is empty | Add the factor or fix coverage |
| Scores keep creeping up across the board | No decay, or no caps | Add decay and cap repeat signals |
Lead scoring vs. related terms
| Term | What it means | How it relates to lead scoring |
|---|---|---|
| Lead grading | A rating of fit only, often a letter | Some tools split scoring (engagement) from grading (fit). Account Engagement, formerly Pardot, grades prospects from A to F alongside a separate score |
| Account scoring | Ranks companies instead of people | Rolls up fit and the activity of everyone you know at the account |
| Contact scoring | Ranks the people inside an account | Tells a rep which person at a strong account to call |
| Lead qualification (such as BANT) | A rep’s judgment after talking to the lead | Scoring decides who to talk to; qualification decides what happens after |
| MQL | A lead marketing judges ready for sales | Usually defined by crossing a lead score threshold |
Where BoostIdeal fits
BoostIdeal is a scoring engine for HubSpot teams, built for the part of lead scoring that most tools leave out: a score your whole team can read, check, and change.
- You build the model and set the weights. AI helps you analyze your data and get to a first draft faster, and a person decides the logic.
- Every score shows its reasons. The factors and weights behind each score write back into HubSpot, so reps, managers, and the workflows and agents acting on the score all see the same logic.
- It scores any record. Leads, contacts, accounts, and custom objects, and one object by the data on another, such as a contact by the fit of its account.
- It runs several models side by side, such as fit next to engagement, so you can compare them before anything acts on a score.
- Scores stay current. They recalculate hourly by default, and in near real time with webhook sync.
- It takes minutes to stand up. You can also start from a CSV.
When it is not the right fit. If you are not on HubSpot, or you want a vendor to hand you a number you never need to question, another approach will suit you better.
Frequently asked questions
What is lead scoring in simple terms?
Lead scoring gives each lead a number that shows how likely it is to become a customer. The number comes from who the lead is (fit) and what they have done recently (engagement), so your team can work the most promising leads first.
What is a good lead score?
There is no universal number. A good score is one that predicts conversion in your own data. Check what share of each score band became customers in the past, and set your thresholds where conversion clearly rises.
What is the difference between a fit score and an engagement score?
A fit score measures how closely a lead matches the customers you win, using stable facts such as company size, industry, and role. An engagement score measures recent activity, such as pricing page visits and email replies, and should fade as that activity gets older.
What is negative lead scoring?
Negative scoring subtracts from a lead’s score for signs it will not buy, such as a competitor’s domain, a student email, or a company outside your market. It keeps activity alone from pushing a bad-fit lead to the top.
How many factors should a lead scoring model have?
Usually three to six fit factors and a handful of engagement signals. More factors dilute the ones that matter and make the score harder to explain.
Is predictive lead scoring better than rules-based scoring?
Not automatically. Predictive scoring can find patterns people miss but needs enough clean history and can be hard to explain. Rules-based scoring is easy to read and change. A strong approach uses AI to inform the model and keeps the final logic readable and owned by a person.
How often should you update a lead scoring model?
Review conversion by band every quarter, and change one or two weights at a time. Revisit the whole model when your market, product, or pricing changes.
What is the difference between lead scoring and account scoring?
Lead scoring ranks individual people. Account scoring ranks companies, combining company fit with the activity of everyone you know there. Because most B2B purchases involve several people, many teams use both.
Do you need software for lead scoring?
You can design a model in a spreadsheet, and that is a good way to start. To score every lead continuously and act on the score, you need it in your CRM, either through the CRM’s own scoring tool or a dedicated scoring tool that writes back to it.
The takeaway
Lead scoring is not about a clever formula. It is about a team agreeing on who a good lead is, writing it down as factors and weights, and checking every quarter that the leads it ranks highest actually buy. Keep fit and engagement apart, let old activity fade, subtract what disqualifies, and make every score show its reasons. A score your team can read is a score your team will use.