Scoring glossary: 43 lead and account scoring terms, defined

Scoring has picked up a lot of jargon, and a lot of it gets used loosely. One team’s MQL is another team’s sales-ready lead. “Predictive” can mean a trained model or a spreadsheet with a formula in it. When sales, marketing, and RevOps use the same words for different things, the score stops being trusted before anyone looks at the math.

This glossary sets out what each term means, why it matters, and where teams usually get it wrong. Every definition is short enough to quote. It is grouped into five parts: the models, how they work, the signals behind them, the handoff, and how to keep a model honest.

The models

Lead scoring

Lead scoring is a method for ranking individual leads by how likely they are to become customers, using a points or weighted model built from who they are and what they do. It answers one question for a rep: who should I call first? The common failure is one blended number that mixes fit and activity, so a student who downloads everything outranks a perfect-fit buyer who visited once.

Account scoring

Account scoring ranks companies rather than people. It combines how well the company matches your ideal customer profile with how much activity is coming from everyone there. It matters because B2B deals are bought by groups, and a lead score cannot see that four people from the same company visited your pricing page this week.

Contact scoring

Contact scoring ranks the people inside an account, usually by role, seniority, and personal engagement. Where account scoring tells you which company to work, contact scoring tells you who to talk to there. Most teams only need it once they have more than one or two contacts per target account.

Fit score

A fit score measures how closely a record matches your ideal customer profile, using stable attributes such as industry, company size, region, role, and the systems they run. Fit changes slowly. It says whether someone belongs on your list at all, regardless of what they did last week.

Engagement score

An engagement score measures how active a record is right now, from actions such as page visits, content downloads, event attendance, email clicks, and product usage. Engagement should fade over time. A score that only goes up treats a pricing visit from last year the same as one from yesterday.

Combined score

A combined score puts fit and engagement together, usually as a grid rather than a single number. HubSpot’s lead scoring tool grades fit A to C and engagement 1 to 3, giving nine bands from A1 to C3. Keeping the two visible is the point: an A3 and a C1 need very different plays, and a single blended number would hide that.

ICP score

An ICP score is a fit score built directly from your ideal customer profile. It turns the profile from a slide into a number on every record, with the factors behind it visible. If you have an ICP document but no ICP score, the profile is not doing any work.

Health score

A health score measures how well an existing customer is doing, from usage, support history, satisfaction, and billing. It is the customer-side counterpart to a fit score. It matters for expansion: an unhealthy account should go to a retention play before anyone pitches an upgrade.

Expansion score

An expansion score ranks existing customers by how ready they are to buy more, either a higher tier (upsell) or a different product (cross-sell). It should only apply to accounts that pass a health check first.

How scores work

Weight

A weight is the share of the total score a factor or signal controls, usually expressed as a percentage that adds up to 100 across the model. Weights are where your team’s judgment lives. If you cannot say why company size is worth 30% and not 20%, the weight is a guess, and that is fine as long as you test it.

Points-based scoring

Points-based scoring adds or subtracts fixed points for each attribute or action: +10 for a director title, +5 for a webinar, and so on. It is simple and easy to explain, and it is how most CRM scoring tools work. Its weakness is that points pile up with no ceiling unless you add one, so the score measures volume more than intent.

Score cap

A score cap is the maximum a score or a single signal can reach. A cap on the whole score keeps it on a fixed scale. A cap on each signal, sometimes called events for full credit, stops one repeated action from running up the total. Ten visits to the same blog post should not outscore one demo request.

Disqualifier

A disqualifier is a rule that lowers or zeroes a score because the record should not be worked, whatever else is true. Common ones are competitors, current customers, students, job seekers, personal email domains, and unserved regions. Disqualifiers should subtract, not just score zero, so a disqualified record cannot climb back up through activity.

Negative scoring

Negative scoring takes points away for attributes or actions that signal a poor fit or waning interest, such as visiting the careers page, unsubscribing, or bouncing. It is the broader version of a disqualifier. Used well, it keeps the top of the list clean.

Score decay

Score decay reduces engagement credit as time passes since the activity happened. Without decay, old activity counts forever and the top of the list fills with records that went quiet months ago. Decay can be stepped (credit drops at set intervals) or continuous (credit fades a little every day).

Half-life

A half-life is the number of days it takes for a signal’s credit to fall by half. A 30-day half-life means a pricing visit is worth half as much after a month and a quarter as much after two. Fast-moving signals such as pricing visits deserve short half-lives. Slower signals, such as a webinar on a strategic topic, can hold value longer.

Threshold

A threshold is the score at which a record moves into a new grade or triggers an action, such as becoming an MQL or being routed to a rep. Set thresholds after you have looked at how real records spread across the scale, not before.

Score band

A score band is a range of scores treated the same way, such as fit 7.0 and up with engagement 3.0 to 5.9. Bands exist so each one can have a play. A score without a band and a play is a number nobody knows what to do with.

Rules-based scoring

Rules-based scoring uses factors and weights a person chose and can read. You can explain any score by pointing at the rules. It is easy to trust and easy to tune, but it is only as good as the judgment behind the weights.

Predictive scoring

Predictive scoring uses a statistical or machine learning model trained on past outcomes, usually won and lost deals, to estimate how likely a record is to convert. It can find patterns people miss. It needs enough history to learn from, and it can be hard to explain why a record scored what it did.

AI scoring

AI scoring is a broad label for any scoring that uses machine learning or language models, from a trained predictive model to an AI that reads a company’s website and rates its fit. The label says nothing about whether you can see the reasoning. Ask what the inputs are, and whether a rep can open a record and read why it scored the way it did.

Explainable scoring

Explainable scoring, sometimes called transparent scoring, is scoring where every score shows the factors and weights that produced it. It matters because reps do not act on numbers they cannot explain to a prospect or a manager. A black-box score gets ignored, however accurate it is.

Signals and data

Signal

A signal is any piece of data that changes how likely a record is to buy: an attribute such as industry, an action such as a pricing visit, or an event such as a new executive hire. Signals are the inputs. The score is what you do with them.

Firmographic data

Firmographic data describes a company: industry, employee count, revenue, location, ownership, and growth. It is the foundation of most fit scores because it is stable and widely available.

Technographic data

Technographic data describes the systems a company runs, such as its CRM, marketing platform, or ERP. It is often the factor that separates two companies that look identical on firmographics. A company on a system you replace fits better than one that just signed with your competitor.

First-party intent

First-party intent is buying behavior you observe on your own properties: your website, your product, your emails, your events. It is the most reliable intent data you have, because you know exactly where it came from.

Third-party intent

Third-party intent is research activity observed somewhere else, usually a network of publisher sites, and sold as signals that an account is researching a topic. It is useful for spotting accounts before they reach you. It is also noisier and usually tells you the company, not the person.

Buying group

A buying group is the set of people at an account who influence or approve a purchase, such as the economic buyer, champion, day-to-day users, and technical or security reviewers. Scoring the buying group, not just one contact, shows whether you have reached the people who can actually say yes.

Buying group coverage

Buying group coverage measures how many of the key roles in a buying group you have engaged. An account with one very active contact and no one else is riskier than it looks. Coverage is often a factor in account scoring for that reason.

Product usage signal

A product usage signal is an action inside your product, such as activating a key feature, inviting teammates, or hitting a plan limit. For product-led and freemium motions, usage is usually the strongest engagement signal you have.

Email open inflation

Email open inflation is the overcounting of email opens caused by privacy features that load tracking pixels automatically. Litmus reports that over 50% of email opens now come from Apple Mail with Privacy Protection turned on. Opens should carry little or no weight in an engagement score. Clicks and replies carry the email signal instead.

Data completeness

Data completeness is the share of records with a value in the fields your model reads. A factor that is blank on half your records cannot carry much weight. The safest rule is that unknown values add nothing to a score but do not subtract either.

The handoff

MQL (marketing qualified lead)

An MQL is a lead that marketing judges ready for sales attention, usually because its score crossed a threshold. The term is only useful if sales and marketing agree on what crossing that threshold means and what happens next.

SQL (sales qualified lead)

An SQL is a lead that sales has reviewed and accepted as a real opportunity worth pursuing. The rate at which MQLs become SQLs is one of the clearest tests of whether your score works.

PQL (product qualified lead)

A PQL is a user or account that has shown buying intent through product usage, such as reaching an activation milestone or hitting a plan limit, rather than through marketing activity. PQLs matter most in product-led and freemium motions.

Speed to lead

Speed to lead is the time between a lead raising a hand, such as a demo request, and a rep responding. Scoring decides the order of work, but a high-intent request should route immediately, whatever its band.

Routing

Routing is the set of rules that decide which rep or team gets a record, usually based on territory, segment, and score band. A good score with bad routing still ends up in the wrong queue.

Play

A play is the specific action attached to a score band, such as “rep outreach within a day” for fit and engagement both in the top band or “nurture, no rep time yet” for mid fit with engagement under 3.0. Plays are what turn a score into behavior.

Keeping the model honest

Conversion by band

Conversion by band is the rate at which records in each score band become opportunities or customers. It is the simplest test of a scoring model: the top bands should convert clearly better than the rest. If they do not, the weights are wrong.

Model drift

Model drift is the slow loss of accuracy as your market, product, or data changes after a model was built. Every scoring model drifts. The fix is a regular review, usually quarterly, not a rebuild every time a number looks odd.

Backtesting

Backtesting means running a scoring model against past records whose outcomes you already know, to see whether it would have ranked the winners near the top. It is the fastest way to test a new model before you switch it on.

False positive

A false positive is a record that scored high but was never going to buy, such as a researcher, a student, or a competitor. A high false positive rate is how reps learn to ignore a score. Most of the fix is disqualifiers and capping repeated activity.

False negative

A false negative is a record that scored low but became a good customer. Looking at recent wins that came from low bands is one of the best ways to find a signal your model is missing.