What is lead scoring?
Lead scoring is a way to rank potential customers by how likely they are to buy. Each lead earns points for signals such as fit with your ideal customer, engagement with your brand and a stated need. The total, often on a 0-100 scale, tells sales who to contact first and who to nurture.
Without a score, every lead looks the same in a list. Reps pick whoever is on top, or whoever looks familiar, and the person who asked for a quote this morning waits behind a newsletter subscriber from last spring.
A score fixes the order, not the outcome. It will not close a deal for you. It makes sure your best hour of the day goes to the people most likely to say yes.
Lead scoring criteria: explicit, implicit and intent data
Every lead scoring model draws on some mix of three kinds of data. Most teams know the first two well. The third is the strongest signal and the easiest one to miss.
Explicit data is what you know about the person: role, company size, industry, location. It answers “could they buy?”. Implicit data is what they do: pages viewed, emails opened, a webinar attended. It answers “are they paying attention?”.
Intent data is what they say they need, in their own words: “we are looking for a new IT support provider before the end of the quarter”. It answers the question that matters most: “do they want to buy now?”. A perfect-fit company that never mentioned a need is a target. A person who asked for a recommendation yesterday is a lead.
| Data type | What it is | Example | What it tells you |
|---|---|---|---|
| Explicit (fit) | Facts about the person and company | Head of marketing, 40-person agency, Berlin | Whether they could buy |
| Implicit (behaviour) | Actions they take around your brand | Opened three emails, visited the pricing page | Whether they are paying attention |
| Intent (stated need) | What they say they need, in their own words | “Can anyone recommend a payroll tool? We switch in October.” | Whether they want to buy now |
How to build a lead scoring model in six steps
You do not need a data team to build your first lead scoring model. A spreadsheet, a list of your recent closed deals and one honest conversation between sales and marketing are enough.
Here is the order that works for most small teams. Expect the first version to be wrong in places. That is fine, because step six exists to fix it.
Step 1: Define your ideal customer
Industry, size, role, region. Fit it on one page.
Step 2: Study won and lost deals
What did buyers have in common before they bought?
Step 3: List the signals
Sort them into fit, behaviour and intent.
Step 4: Assign points
Give stated intent the most weight. Add minus points too.
Step 5: Set thresholds
Decide which total means “call today”.
Step 6: Review every month
Compare scores with real outcomes and adjust.
Example points: what to reward and what to subtract
Keep the scale small and the logic obvious. A new rep should be able to read your model and predict a score in their head. Here is an example set of lead scoring criteria for a B2B service business:
- +30: a direct request (“can anyone recommend…”, “we are looking for…”).
- +20: a deadline or date (“before Q4”, “by the end of the month”).
- +15: a budget mentioned or already approved.
- +15: a match with your ideal customer (role, company type, region).
- +10: switching away from a current supplier or tool.
- −20: offering the same service (a competitor, not a buyer).
- −15: the signal is more than a month old.
Why minus points and time decay matter
Negative points matter as much as positive ones. Without them, competitors, recruiters and job seekers float to the top simply because they use all the right words.
Time decay matters too. Interest from three months ago is history, not pipeline. If old signals keep their points forever, your hot list slowly fills with people who have already bought from someone else.
Lead scoring examples for three kinds of business
The same idea looks different depending on what you sell. What stays constant is that stated intent carries the most weight, and fit decides whether that intent is worth your time.
Here is how the strongest signals shift from one business to another:
- A B2B agency: the heaviest points go to a request for a contractor, a named deadline and a company that matches your client profile. A founder asking “who built your site?” outscores a marketer who liked three of your posts.
- An online store supplier: switching signals matter most. “Our current supplier keeps missing deliveries” is worth more than any number of catalogue views.
- A local service, such as a dental clinic or a tutor: location and timing decide everything. “Need a dentist near Kings Cross this week” is hot; “dentists are so expensive now” is cold, even though it names the same service.
What is a good lead score? Score bands and next actions
There is no universal “good” lead score. A 70 in one model can be a 40 in another. What matters is that each band on your scale maps to one clear action, and everyone on the team knows it.
Three bands are enough for most teams. With more than five, people stop remembering what each one means and start ignoring the labels.
86A request with a date and a budget
- 80–100HotA direct request, often with a date or budget. Reply the same day.
- 50–79WarmA real problem, still comparing options. Send a note that fits their case.
- 0–49ColdAn opinion or broad interest. Do not pitch; keep watching the topic.
Keep an eye on how big each band gets
If half your leads are hot, your model is too generous and “hot” no longer means “drop everything”. If nobody is ever hot, you are probably scoring behaviour and missing intent.
A healthy hot band is small enough that someone can answer all of it the same day. If it keeps overflowing, raise the threshold before you hire another rep.
And do not be afraid of the warm band. It is usually the biggest one, and it is where a careful, useful reply works best: competitors have not shown up yet, and the person is already looking for a fix.
Rule-based vs predictive vs AI intent lead scoring
Once you know what to score, the next question is who does the scoring: you, a statistical model trained on your history, or a language model that reads messages. Each approach has its place.
| Rule-based | Predictive | AI intent scoring | |
|---|---|---|---|
| How it works | You set points for each signal by hand | A model learns weights from past won and lost deals | A language model reads what the person wrote and judges the need |
| What it needs | A clear ideal customer profile | A long history of closed deals in your CRM | Text: requests, posts, chat messages |
| Strengths | Transparent, cheap, quick to change | Finds patterns people miss | Understands meaning, not keywords; works on brand-new leads |
| Weak spots | Only as good as your guesses; drifts out of date | Hard to explain; weak with little history | Needs a reason next to the score to be trusted |
| Best for | Small teams, new products | Mature sales teams with lots of data | Chats, communities and inbound messages |
When predictive lead scoring falls short
Predictive lead scoring sounds like the obvious upgrade, but it learns from the past. If you closed only a handful of deals last year, it has almost nothing to learn from. And it cannot say anything about a person who has never touched your website or your emails.
AI intent scoring works the other way round. It does not need your history, because it reads the need straight from the message. That is why it suits Telegram, where most of your future buyers have never heard of you.
How AI lead scoring reads a single message
An AI intent model does not count keywords. It reads the whole message the way a sharp sales rep would and asks a few questions: is this person asking for something, or just talking? Is there a date, a budget or a switch? Does the request match the buyer you described?
Then it does the part people skip when they are tired: it lowers the score for ads, for people offering the same service and for opinions dressed up as questions. What comes out is a number and a one-line reason, such as “asks for a contractor, deadline in two weeks, matches your niche”.
The reason is what makes the number usable. You can agree with it in a glance, or spot that the model misread a joke and move on.
Language is one more advantage. People write in English, Russian or Ukrainian, mix them in one sentence or post a screenshot, and the request still means the same thing.
Score what people say, not only what they click.
Scoring conversations, not form fills
Classic lead scoring was built for websites. Someone downloads an e-book, opens three emails and visits the pricing page, and the points add up. It works, but it is indirect: you are guessing intent from clicks.
In public Telegram chats people skip the guessing. They write, in plain words, what they need, when and sometimes for how much. The catch is volume. The same chat is full of jokes, complaints, ads and news, and a keyword filter cannot tell them apart.
Look at the thread below. Four people are talking about catering. Only one of them is ready to book it this month.
Is it just me, or did every caterer put their prices up after the summer?
We made the team lunch ourselves once. Forty sandwiches at 7am. Never again.
Need a caterer for our 60-person team offsite on 14 November, vegetarian-friendly, budget around €2k. Who have you used?
Buying intentHot · 89
We are a catering company in Amsterdam, DM us for 10% off your first order
What the score sees in that thread
A keyword alert would ping you four times. An intent score puts Hanna near the top because her message has a direct request, a date, a headcount and a budget. Sofie and Tom are just chatting. Ruben scores low because he is selling, not buying.
This is also why a score needs its reason. “89: asks for a caterer, event on 14 November, 60 people, budget stated” takes two seconds to check. A bare “89” asks you to take it on faith.
Example search — sign in free to run it
Lead scoring vs lead qualification (and where BANT fits)
The two terms often get mixed up, but they answer different questions. Lead scoring is automatic and relative: it ranks a whole list so you know where to start. Lead qualification is a human check on one lead: is this a real opportunity worth a sales conversation?
BANT is a popular qualification framework: Budget, Authority, Need and Timeline. It works well as a checklist once you are talking to someone. It is also a handy source of scoring signals, because budget, need and timeline often appear in the very first message.
So the flow is simple: score to sort, then qualify the top of the list. Scoring saves your time. Qualification saves your pipeline from wishful thinking.
- Scoring: many leads, automatic, gives you an order.
- Qualification: one lead, a conversation, ends in yes or no.
- BANT: a checklist that feeds both.
Common lead scoring mistakes
Most scoring models fail quietly. Nobody complains; reps just stop looking at the number. These are the usual reasons.
- Points for every email open
- No minus points
- Scores never decay
- A bare number with no reason
- Set once, never reviewed
- Most weight on stated intent
- Minus points for sellers and job seekers
- Old signals lose points
- Every score shows why
- A monthly check against won deals
Activity is not intent
The biggest trap is rewarding activity instead of intent. Someone who opened ten newsletters is a fan, not necessarily a buyer. Someone who wrote “we need this by Friday” is a buyer, even if they have never seen your site.
The second trap is the bare number. If a rep cannot see why a lead scored 85, they will not trust the 85, and they drift back to picking leads by gut feeling.
Who should own the lead scoring model
A model with no owner drifts. Pick one person, usually the head of sales or the founder in a small team, who changes the points and signs off on the thresholds.
Once a month, pull the leads that turned into deals and the hot leads that went nowhere. If the winners kept scoring 60 and the losers 90, your weights are upside down. Fix one thing at a time, so you can tell which change helped.
And agree on the words. When the whole team means the same thing by “hot” and “warm”, handing a lead from one person to another stops being a lottery.
Hot, warm and cold leads explained
Hot, warm and cold are just names for score bands, but each one carries its next action. Here is what each one looks like in a real chat, and what to do with it.
- PSHot
Priya S.
@priya_builds
90intent scoreSaaS founders London· public group · 1h ago
Looking for a videographer to film our product launch event on 3 November in Shoreditch. Budget is around £4k.
Why · Direct request with a fixed date and a stated budget.
- LRWarm
Luca R.
@luca_ships
63intent scoreEtsy and Shopify sellers· public group · 5h ago
Our stock lives in three spreadsheets and we oversold twice this month. What do you all use for inventory?
Why · Clear pain and an open question, but no timeline yet.
- NWCold
Nina W.
@nina_people
22intent scoreHR managers Europe· public group · 1d ago
Honestly, every HR platform demo looks the same to me at this point.
Why · An opinion about the category, no request or need.
- Hot lead (80–100): asks directly, often with a date or a budget. Reply the same day and mention their exact problem.
- Warm lead (50–79): describes a real pain but is still comparing. Send a short, useful note that fits their case.
- Cold lead (0–49): an opinion or broad interest. Do not pitch. Keep watching the topic.
Cold does not mean worthless
A cold message still tells you where your topic is being discussed. Today's cold commenter can post next month's hot request, and you want to be watching that chat when it happens.
Warm leads deserve patience. Answer the question they actually asked, share something useful and skip the sales pitch. When they are ready to choose, you will already be part of the conversation.
What does a good first reply to a hot lead look like? To Priya from the founders' chat: “Saw you need the launch filmed on 3 November. We have shot product launches in East London at short notice, happy to send two short cuts.” Short, specific and clearly written after reading her message.
Where to start this week
If you have no scoring at all, do not start with software. Take your last closed deals, write down the first thing each buyer said to you, and your strongest intent signals will be right there in black and white.
If many of your buyers talk in Telegram chats, you can skip the manual reading. AI lead scoring for Telegram reads public chats 24/7, gives each person a 0-100 score with a short reason and keeps the source message attached. No Telegram account is needed on your side, and the free plan includes 50 ranked leads a month.
Want the wider picture first? Read what counts as a lead, how lead generation works and where scored leads sit in your sales funnel.
Frequently asked questions
What is lead scoring in simple terms?
Lead scoring gives each potential customer a number, often from 0 to 100, based on how well they fit your ideal customer, how engaged they are and whether they have said they need what you sell. The higher the score, the sooner sales should reach out.
How do you build a lead scoring model?
Define your ideal customer, study your recent won and lost deals, list the signals that came before a purchase, assign points (with minus points for competitors and old signals), set thresholds for each band and review the model every month against real outcomes.
What is a good lead score?
It depends on your model, so there is no universal number. What matters is that each band maps to one action. A common setup is 80–100 hot (reply today), 50–79 warm (send a tailored note) and 0–49 cold (keep watching).
What is the difference between lead scoring and lead qualification?
Lead scoring automatically ranks a whole list so you know where to start. Lead qualification is a human check on one lead, often with a framework like BANT, to decide whether it is a real sales opportunity. Score first, then qualify the top of the list.
What is predictive or AI lead scoring?
Predictive lead scoring learns point weights from your past won and lost deals, so it needs a lot of history. AI intent scoring reads what a person actually wrote and judges how ready they are to buy, so it works on new leads from chats and communities too.