Lead scoring: how to prioritize the leads that close

When you have ten leads a week, you call all of them. When you have a hundred, someone has to decide who gets called first, and that decision is usually made by gut feel, whoever shouted loudest or whatever is at the top of the inbox.

Lead scoring replaces that guesswork with a number. This guide explains how scoring works, how to build a simple points model you can trust, where to set thresholds and how to keep the model honest over time.

What lead scoring is (and isn’t)

A lead score is a single number, often 0 to 100, that estimates how likely a lead is to become a customer. It’s a prioritization tool, not a verdict. A low score means “call later,” not “never call.”

Scoring is useful when:

• Leads arrive faster than your team can respond to all of them.
• Reps spend hours on leads that rarely close.
• You want automations to treat hot and cold leads differently.

It’s less useful when you have very few leads, or when every lead gets a call within minutes anyway. Don’t add a scoring model just because the software has one.

Fit vs engagement signals

Most scoring models combine two kinds of signals.

Fit answers “Is this the kind of customer we win?” It’s based on who the lead is:

• Service or product they need
• Location or service area
• Budget range or project size
• Company size or role (for B2B)
• Timeline to buy

Engagement answers “How interested are they right now?” It’s based on what the lead has done:

• Requested a quote or a callback
• Replied to a text or email
• Booked or attended an appointment
• Came from a high-intent source (a “get a quote” form rather than a newsletter signup)
• Visited your pricing page or opened a quote

A lead with great fit and no engagement is a nurturing candidate. A lead with high engagement and poor fit (outside your area, wrong service) is a polite “no” or a referral. The leads you want first have both.

Build a simple points model

Start with your own data, not someone else’s template. Pull your last 30 to 50 won deals and a similar number of lost or dead leads, and look for the traits that show up far more often in the winners. Those are your signals.

Here’s an example model for a hypothetical home remodeling contractor:

Signal	Type	Points
Requested an on-site estimate	Engagement	+25
Project budget above $15,000	Fit	+20
Wants to start within 60 days	Fit	+15
Inside primary service area	Fit	+15
Replied to first text or email	Engagement	+10
Came from a referral	Engagement	+10
Homeowner (not a renter)	Fit	+5
Outside service area	Fit	−30
“Just getting prices” with no timeline	Engagement	−10
No reply after three attempts	Engagement	−15

A few rules of thumb:

• Keep it to 6–12 signals. More rules rarely mean better predictions; they mostly make the score harder to explain.
• Weight by evidence. If 80% of your won jobs requested an on-site estimate and only 20% of lost ones did, that signal deserves a lot of points.
• Use negative points. Disqualifiers matter as much as positives. A lead outside your service area shouldn’t outrank a local one because they clicked a lot.
• Cap the total at 100 so scores are easy to compare.

A high-ticket closer might use a different list (stated goal, budget confirmed, decision maker on the call, attended the first appointment), but the structure is the same.

Don’t forget time

Interest fades. In a 2011 Harvard Business Review study of 1.25 million online leads at 42 US companies, firms that tried to contact a lead within an hour were nearly seven times as likely to qualify it as firms that waited even an hour longer, and more than 60 times as likely as firms that waited 24 hours or more. The research is over a decade old, but the lesson for scoring still holds: brand-new inbound leads should get a fast first touch regardless of score, and engagement points should fade if nothing happens for weeks.

Make scores explainable

A score nobody understands is a score nobody uses. If a rep sees “72” and has no idea why, they’ll ignore it and go back to gut feel.

Explainable scoring shows the factors behind each number, for example: +25 requested estimate, +15 inside service area, +10 replied to text, −10 no timeline given. That helps in three ways:

1. Reps trust it. They can see the logic and agree or disagree.
2. Reps use it in the conversation. “I see you’re hoping to start within two months. Let’s talk about scheduling.”
3. You can fix it. When a score looks wrong, you can see which signal caused it.

Explainability also depends on clean data. A score built on missing or outdated fields will mislead you. In Salesforce’s 2024 State of Sales survey of 5,500 sales professionals (vendor research), only 35% said they completely trust the accuracy of their organization’s data. Required fields on your forms and a habit of updating records after calls do more for scoring than any clever formula.

Set thresholds and actions

A score only matters if it changes what someone does. Define two or three bands and the action for each:

Score	Label	Action
70–100	Hot	Call within the hour, high-priority task, owner notified
40–69	Warm	Call same day, add to standard follow-up sequence
0–39	Cool	Automated nurture messages (with consent), review monthly

Set the first thresholds by looking at where your past won deals would have scored, then adjust after a month. If your “hot” band contains half of all leads, it’s too wide. If it rarely contains a deal that closes, your signals need work.

Scoring also pairs naturally with lead routing: hot leads can go to your most experienced closer, while warm leads rotate across the team.

Avoid bias in your model

Scoring models learn from your history, and history has blind spots. A few guardrails:

• Score the job, not the person. Use signals tied to need, budget, timeline and engagement. Avoid proxies like name, age, neighborhood demographics or language that can quietly filter out good customers and may create legal risk.
• Watch for self-fulfilling loops. If low-scored leads never get called, they’ll never close, and the model will “prove” it was right. Call a sample of low-scored leads each month to test it.
• Check who gets filtered out. If one source, area or customer type always scores low, ask whether that reflects real buying behavior or just how you’ve treated them in the past.
• Keep a human in the loop. Reps should be able to override a score with a note explaining why.

Review the model every quarter

Markets, offers and seasons change, so your model should too. Once a quarter:

1. Pull leads created in the last 90 days and their outcomes.
2. Compare average scores of won vs lost leads. If they’re close, the model isn’t separating them.
3. Check each signal: does it appear more often in won deals? If not, reduce or remove it.
4. Look for new patterns, such as a new lead source that converts unusually well.
5. Write down what you changed and why, so next quarter’s review has context.

Common lead scoring mistakes

• Copying a generic template. A SaaS company’s scoring model won’t fit a roofing company.
• Scoring only engagement. Lots of clicks from someone outside your area is still a poor lead.
• Black-box scores. If reps can’t see the factors, they won’t trust the number.
• Set and forget. A model built two years ago is scoring a business that no longer exists.
• Letting score replace speed. Every new inbound lead deserves a fast first reply; the score decides how hard you chase after that.

For qualification questions that feed your fit signals, see how to qualify leads, and for how AI can help triage, see our post on AI lead qualification.

How Exa helps

Exa includes explainable lead scoring that shows the factors behind each score, so reps can see why a lead ranks where it does instead of trusting a bare number. Scores sit alongside the rest of the lead record: source attribution with UTM parameters and click IDs, custom fields for your qualification answers, tags, notes and an activity timeline.

You can put scores to work with automations, which can use the lead’s score (along with status, source, tags and custom fields) as a condition, then assign an owner, create a high-priority task, notify someone or send a message. The AI assistant can summarize a lead with highlights, risks and next steps, and answer questions like “Which high-scoring leads haven’t been contacted this week?”

Final thoughts

Lead scoring works best when it’s simple, explainable and tied to action. Build it from your own wins and losses, keep it to a handful of signals your team understands, give every new lead a fast first touch and check the model every quarter. A score your reps trust is worth far more than a sophisticated one they ignore.