How to Build a Lead Scoring Model
A practical fit-plus-intent lead scoring framework for inbound sales teams: criteria, point tables, routing thresholds, decay and how AI changes the inputs.
· 9 min read
Lead scoring has one job: decide which conversation a rep should have next. Everything else, the dashboards, the AI, the CRM fields, exists to make that decision better. This guide builds a scoring model from scratch in a way a sales manager can maintain, then shows where AI changes the inputs without changing the method.
Two axes, not one number
A single score hides the most useful distinction in the pipeline: a lead can be a great fit with no urgency, or urgent and a terrible fit. Score the two separately.
- Fit is whether you can and want to serve them. It comes from facts about the buyer: location, size, category, role. It changes slowly.
- Intent is whether they want to buy now. It comes from behaviour: what they asked, how fast they replied, which page they came from. It decays quickly.
High fit and high intent is a rep’s next call. High fit and low intent is a nurture sequence. Low fit and high intent is a polite decline or a referral. Low fit and low intent is noise. Four quadrants, four different actions, and no single number could have told you which.
Choose fit criteria you can actually observe
Start from closed deals. Pull the last fifty wins and the last fifty losses and list the facts you knew about each at the moment they entered the pipeline. The facts that separate the two groups are your fit criteria. Typical ones:
- Serviceable geography (branch, city, delivery zone, micro-market).
- Size band (team size, family size, bed count, order volume).
- Category match (the product line or treatment they asked about is one you sell).
- Buyer role (decides, influences, researches).
- Channel of arrival, because some channels consistently bring better-fit buyers.
Keep it to five or six. A model with fifteen fit criteria is a model nobody updates.
Intent signals by channel
Intent lives in behaviour, and behaviour looks different on each channel. Define the signals per channel rather than pretending a web visit and a WhatsApp reply mean the same thing.
| Channel | Strong intent | Weak intent |
|---|---|---|
| WhatsApp / chat | Names a timeline, asks about availability or price, replies within minutes | “Send details”, no reply to a question |
| Voice | Asks to book, mentions a deadline, calls back | Hangs up before the second question |
| Web form | Pricing or demo page, specific product selected | Generic contact page, no product chosen |
| Email / campaign | Replies to a campaign, clicks a booking link | Opens only, unsubscribes |
Build the point table
Give every criterion a small number of possible answers and a point value for each. Points are not science; they encode your judgement so that it is applied consistently. A workable first version:
| Axis | Criterion | Best → worst |
|---|---|---|
| Fit | Geography | 25 → 10 → 0 |
| Fit | Size band | 25 → 15 → 5 |
| Fit | Category match | 25 → 10 → 0 |
| Fit | Role | 25 → 12 → 5 |
| Intent | Timeline stated | 40 → 20 → 5 |
| Intent | Asked about price / availability / booking | 30 → 0 |
| Intent | Reply speed to a question | 20 → 10 → 0 |
| Intent | Source page or campaign | 10 → 5 → 0 |
Each axis sums to 100, so a lead reads as “fit 75, intent 40” and everyone in the room knows what that means.
Thresholds and routing tiers
Three tiers are enough for most teams. Attach an action and an owner to each; a tier without an action is decoration.
- Hot: fit above 60 and intent above 60. Route to a rep now, with the conversation attached. Target a human reply within minutes.
- Warm: fit above 60, intent between 30 and 60. Automated follow-up on a cadence, then a rep call once intent rises or after a set number of touches.
- Nurture or decline: everything else. Content sequence for low intent, a courteous no for low fit.
Decay and recency
Intent is perishable. A buyer who asked about pricing three weeks ago and went silent is not the same lead as one who asked this morning. Reduce intent points on a schedule, for example by half every seven days without a reply, and let any new reply reset the clock. Fit does not decay; a buyer in your service area stays in your service area.
Where AI changes the inputs
The framework above is the same with or without AI. What AI changes is how many of the inputs you actually have.
- Unstructured conversations become structured fields. “Budget around 1.2Cr, shifting next month” in a WhatsApp thread becomes a budget band and a timeline without a rep typing them in.
- Every lead gets asked. Manual qualification skips leads when the team is busy, which means the score is missing exactly when volume is highest. An agent asks the five questions on every conversation.
- Reply speed and language become signals. The agent can log how quickly a buyer answered and what they asked, both of which correlate with intent and neither of which a form captures.
This is the difference between how a conversational qualification tool scores and how a CRM that scores on form fields and email opens does; the Synergon vs HubSpot comparison goes into that in detail.
Calibrate against outcomes, monthly
A scoring model is only as good as its last calibration. Once a month, take every lead that closed or was lost and look at the score it had when it entered the pipeline. If closed deals are spread evenly across tiers, the model is not discriminating and a criterion is wrong. If warm leads close as often as hot ones, the intent threshold is too high. Adjust one thing at a time and write down why.