Rodok
Abderrazzak AbouyaalaAI lead scoring uses a model to rank leads by how likely they are to become customers, based on patterns in your past wins and losses instead of point rules you write by hand. It needs data to learn from, so small teams usually start with a simple rules-based score, then add AI as their deal history grows. The best scores combine three things: fit, signals and engagement.
Most articles on lead scoring are written for marketing teams with thousands of form fills. This one is for a sales team of 2 to 20 that wants a short list of who to contact first. It covers how scoring works, the calculation step by step, a worked scorecard you can copy into a spreadsheet, the mistakes that make scores useless, and how a buying signal feeds in.
What is lead scoring?
Lead scoring is a way to give every lead or account a number or a label (hot, warm, cold) so reps know where to spend their time. HubSpot, for example, builds scores from points added or subtracted per rule, and sorts records into High, Medium and Low bands (HubSpot).
A good score answers one question: who should I contact first today? A bad score is a number nobody trusts, so reps ignore it.
The three ways to score leads
| Method | How it works | Needs | Best for |
|---|---|---|---|
| Rules-based | You assign points by hand: +15 for a VP title, +10 for 50 to 200 employees | A clear ideal customer profile | Any team, from day one |
| Predictive | A model finds what your past customers have in common and scores new leads against it | Enough closed deals and leads to learn from | Teams with a few hundred leads and many wins |
| AI (signal-aware) | A model, often a language model, reads signals like posts and news and judges fit in context | A source of signals and a definition of fit | Teams that act on timing, not just attributes |
The boundaries blur. "Predictive" and "AI" are often used for the same thing. The practical difference is where the information comes from: your own history, or fresh signals read in context.
How much data does predictive scoring need?
More than most small teams have. HubSpot's AI scores need at least 50 contacts, 25 converted and 25 not, and are only available on Marketing Hub Enterprise (HubSpot). Salesforce's Einstein cheat sheet says a model on your own data needs at least 1,000 lead records with at least 120 converted, and falls back to a shared model below that (Salesforce). Thresholds change, so check the current docs.
If you have 30 customers, start with rules.
The AI lead score calculation method, step by step
Whatever the vendor, a predictive or AI score is built in roughly the same way:
- Collect outcomes. Mark each past lead as won, lost or never progressed.
- Pick the inputs. Attributes of the person and company (title, industry, size, region) plus behavior (replies, meetings, site visits).
- Train the model. It looks for combinations that show up more often in won deals than in lost ones.
- Score new leads. Each gets a probability of converting, usually shown on a 0 to 100 scale or in bands.
- Retrain regularly. Your market, product and buyers change, so a model trained on last year's wins drifts.
Two traps hide in step 1. If your past leads came mostly from one channel, the model learns that channel, not what makes a good buyer. And if your sales team never contacted certain kinds of leads, the model can't know they would have bought.
A worked scorecard: fit, signal and engagement
Here is a rules-based score you can build in a spreadsheet in an hour. It has three parts, each capped so no single part can dominate. The numbers are an example. Replace them with what your own wins show.
| Part | What it measures | Max | Example rules |
|---|---|---|---|
| Fit | Is this the kind of buyer we win? | 50 | In target industry +15, 20 to 200 employees +10, VP, Head or founder +15, in your markets +5, uses a tool you work with +5 |
| Signal | Is there a reason to talk now? | 30 | Started a target role in the last 30 days +15, hiring for a related role +10, posted about the problem you solve +10 |
| Engagement | Have they responded to us? | 20 | Replied +15, accepted a connection +5, visited pricing page +10 (if you track visits) |
Add the three parts. Then set bands:
- 70 and above: hot. Contact today.
- 40 to 69: warm. Contact this week, or wait for a second signal.
- Below 40: nurture. Don't spend rep time yet.
Three leads, scored
All names are made up.
| Lead | Fit | Signal | Engagement | Total | Band |
|---|---|---|---|---|---|
| Dana Whitfield, VP of Sales at Northgate Software (120 people, SaaS, US). Started 9 days ago; Northgate is hiring 2 SDRs | 45 | 25 | 0 | 70 | Hot |
| Priya Raman, Director of Operations at Halden Logistics (1,800 people, outside your target). Started 3 weeks ago; visited your pricing page | 10 | 15 | 10 | 35 | Nurture |
| Marcus Lee, founder of a 6-person consultancy. No signals; replied to a newsletter | 20 | 0 | 15 | 35 | Nurture |
Read what the score does here. Dana has no engagement at all, yet she ranks first, because she is a strong fit and has two fresh signals. Priya has a signal and some interest but the wrong profile (right region and tools, wrong company size, industry and seniority), so she waits. Without the signal, Dana would be 45 and warm. The signal is what moves her to the top of today's list.
Add decay
Signals go stale. A new role is most useful in the first 30 to 90 days. Halve a signal's points after 30 days and drop them to zero after 90. If Dana's signals were 40 days old, her signal part would be about 12 and her total 57, warm instead of hot. Without decay, last quarter's leads clog the top of the list.
How a buying signal feeds a lead score
A buying signal is an event that suggests a lead may be ready to talk: a new role, a hiring push, a funding round, a post asking for recommendations (15 examples here). In a scorecard, signals are the part that changes week to week. Fit rarely changes. Signals do.
Two ways to feed them in:
- Add points. Each signal type gets a weight and a decay, as in the scorecard above.
- Use judgment. Read the signal in context. A "VP of Sales hired" signal means one thing at a 15-person startup you can serve and another at a company that's too big or in the wrong segment.
This is where AI earns its place. Matching "VP of Sales" in a title field is easy. Judging whether a post about "struggling to scale outbound" comes from a company you can help is a reading task. For how signals work in selling, see signal-based selling and intent signals.
Lead scoring mistakes that make scores useless
- Scoring on opens and clicks. Email opens are unreliable and say little about intent. Give weight to replies and meetings.
- No negative scores. Subtract points for the wrong things: a student email, a competitor domain, a company size you can't serve, "unsubscribed."
- Fit and interest in one number. A tiny company that clicked ten times and a perfect-fit VP who hasn't clicked once need different treatment. Keep fit and engagement as separate parts, as HubSpot does with its fit, engagement and combined scores.
- Never checking against outcomes. Every quarter, look at your closed-won deals and ask what score they had. If half were "nurture," the rules are wrong.
- No decay. Old signals look like current ones.
- Too many rules. Ten rules you understand beat fifty you don't. Start small.
- No action attached. A score only works if each band triggers something: hot gets a call today, warm enters a sequence, nurture waits. See lead qualification for what to do after the score.
Lead scoring tools
| Tool | What it does | Price |
|---|---|---|
| HubSpot | Manual scoring of contacts, companies and deals; AI scores for contacts on Marketing Hub Enterprise | Scoring needs Marketing Hub or Sales Hub Professional or Enterprise; check their site (docs) |
| Salesforce Einstein | Predictive scoring on your lead and conversion history | Check their site (cheat sheet) |
| A spreadsheet | Rules-based score you control | Free |
| Rodok | Judges each signal for fit, marks it hot or warm, adds an AI write-up; the CRM holds the rest of the record | Pro $79/month for 3 seats |
If you use a spreadsheet, add one column per part of the score and one for the band. If you use a CRM, put the score in a field so reps can sort by it. Rodok's lead enrichment fills in the company and person details that the fit part needs.
Frequently asked questions
What is lead scoring?
Lead scoring ranks leads by how likely they are to buy, using points or a model. Points come from fit (title, industry, size), signals (new role, hiring) and engagement (replies, meetings). The score tells reps who to contact first.
What is AI lead scoring?
AI lead scoring uses a trained model to estimate each lead's chance of converting, instead of fixed point rules. It learns from past won and lost deals, and some tools also read signals such as news or posts in context. It works best when you have enough history to learn from.
What is the AI lead score calculation method?
Collect past outcomes, choose inputs like title, company size and behavior, train a model to find what won deals have in common, then score new leads with a probability that is usually shown on a 0 to 100 scale. Retrain regularly so it doesn't drift.
Do I need AI for lead scoring?
Not at first. With fewer than a few hundred leads, a rules-based score you understand beats a model with nothing to learn from. Add AI when you have enough closed deals, or when you want signals read in context.
How do you build a lead scoring model for a small team?
Write down your ideal customer, assign points for fit, signals and engagement, cap each part, set hot, warm and nurture bands, and test it against your last 20 won deals. Adjust the weights until your best customers would have scored as hot.
What is predictive lead scoring?
Predictive lead scoring is a type of AI scoring that finds patterns in your historical data and uses them to rank new leads. HubSpot and Salesforce both offer it, with minimum data requirements, so check the current docs for your plan.
