Lead scoring is a numerical system that ranks contacts by their likelihood to buy, based on who they are and what they have done. The four implementation steps are: define firmographic fit criteria, map behavioural signals to point values, set thresholds that route leads to sales or nurture, and connect the model to sequences that act on each band. When it is built correctly, sales only talks to leads that are ready.
TL;DR. A B2B scoring model combines firmographic fit — company size, industry, job title — with behavioural signals: pages visited, content downloaded, emails clicked. Score thresholds route leads to sales, ongoing nurture, or disqualification. The right 90-day nurture frequency is lower than most teams run: one to two emails per week for the first month, dropping to bi-weekly after day 45. AI scoring adjusts weights automatically as conversion data accumulates; rules-based scoring uses fixed weights and is easier to maintain under 500 leads per month. Your model is working when SQL-to-close rates improve and sales stops calling the leads unqualified.
How do you actually implement lead scoring?
Start with your closed-won data, not a blank scoring sheet. Pull the last 50 to 100 deals you closed and identify the common attributes — company size, job title, pages visited before the first sales call, content consumed. Those are your leading signals. Any model built without this analysis is a guess that will need to be torn down in three months.
Step one: define fit. Score every lead on firmographic criteria before touching behavioural data. Assign points for company size in your ICP (50–500 employees = 20 points, 500+ = 15, under 50 = 0). Score job title by buying power (VP or C-suite = 25 points, Director or Manager with budget authority = 15, individual contributor = 5). Score industry fit (exact ICP match = 15 points, adjacent = 5, out of scope = 0). A perfect firmographic score before any behaviour is around 60 points.
Step two: layer in behaviour. Map content and site actions to point values based on how strongly each correlates with a closed deal in your data. Pricing page visit = 20 points. Demo request = 40 points plus an immediate sales alert. Gated content download = 10 points. Three or more blog visits in one session = 5 points. Email open = 2 points; email click = 5 points.
Step three: set thresholds. Three bands cover most B2B models. Sales-ready: 70 points or above — route to CRM, trigger SDR task. Active nurture: 30 to 69 — enrol in the relevant sequence, continue tracking. Disqualify: below 30 with no activity in 30 days — suppress from active sends.
Step four: connect to your CRM and marketing platform so the score recalculates on every new event. Threshold crossings should fire in real time, not on a nightly batch. A lead that hits 70 points at 11am should be in a sales queue by 11:05am.
What signals belong in a B2B scoring model?
The signals that belong in a scoring model are the ones that correlate with revenue in your data. The signals that do not belong are the ones that feel important but have not been validated. A common mistake is over-weighting email opens. Opens indicate deliverability, not intent. A pricing page visit or a repeat visit to your ROI calculator is worth ten opens.
Positive behavioural signals to weight heavily: demo or trial request (40–50 points), pricing page visit (20–25 points), case study or ROI content download (15 points), three or more pages in a session (10 points), webinar attendance (10 points). Moderate signals: email click (5 points), blog visit from a non-branded keyword (3 points), retargeting ad click (3 points). Light signals: email open (2 points), single organic blog visit (1 point).
Negative signals matter as much as positive ones. Score down for: job title outside buying authority (−10 points), company size outside ICP (−15 points), personal email domain (−20 points), unsubscribe from nurture (immediate disqualify). A lead with 80 raw points from email opens but a personal Gmail address and a job title of ‘student intern’ is not sales-ready. Negative scoring prevents those from polluting your SQL queue.
Decay scoring keeps the model honest over time. If a lead has not engaged in 30 days, reduce their score by 10 percent. At 60 days of silence, halve it. Leads who were warm three months ago and have gone cold should not sit at 65 points in your CRM, blocking a better lead from the queue.
What’s the right nurture frequency over 90 days?
Most teams over-send. A 90-day nurture sequence should run at a frequency the lead controls, not a cadence your team finds convenient. Send when you have something worth saying. In the first 30 days, that is usually one to two times per week — welcome, value delivery, educational content. After day 30, drop to weekly. After day 45, bi-weekly is usually sufficient unless the lead re-engages.
A practical 90-day cadence for a B2B lead that downloaded a gated asset: Day 1–2 — immediate delivery email plus a one-line confirmation. Day 4 — one supporting resource directly related to what they downloaded, no product pitch. Day 7 — a short outcome example relevant to their industry. Day 14 — a direct question (‘Are you evaluating this for a specific project?’) — this is a behaviour trigger, not a broadcast. Day 21 — a how-to piece or tool. Day 30 — soft CTA: book a call or take the assessment. Days 31–60 — weekly, alternating between education and soft CTA. Days 61–90 — bi-weekly, progressively more conversion-focused.
Triggered emails should break the cadence. If a lead visits the pricing page on day 18, they do not need the day 21 how-to email — they need a sales alert and an offer to talk. The cadence is a floor, not a ceiling. High-intent behaviour should always override the scheduled sequence.
Remove leads who have not engaged with a single send in 45 days from active nurture. Send one re-engagement email and suppress them if there is no response. A clean list outperforms a large one.
How does AI scoring differ from rules-based?
Rules-based scoring assigns fixed weights: every pricing page visit is worth 20 points, for every lead, regardless of company size, industry, or prior behaviour. It is transparent, easy to audit, and easy to fix when something breaks. For teams processing under 500 leads per month, it is the right choice. Build the rules model first, run it for 90 days, and use the conversion data it generates to decide whether a predictive model is worth the investment.
AI-based scoring builds a predictive model from historical conversion data. Instead of fixed weights, it identifies which combinations of signals — firmographic profile, content consumed, sequence of actions — most strongly predict a closed deal in your specific pipeline. A mid-market company that downloads your pricing guide and then attends a webinar might score 20 points higher in a predictive model than an enterprise doing the same actions, if your data shows mid-market converts faster.
The calibration difference shows up at scale. A rules model is adjusted manually — someone reviews it quarterly and updates weights based on sales feedback. A predictive model updates continuously as new conversion data flows in. But this requires clean data: instrumented CRM fields, consistent lead source tracking, and closed-won/lost data tagged to the right lead record. If your data is messy, a predictive model will learn your mess.
AI scoring also handles conditional signal combinations that rules logic cannot maintain at scale. A pricing page visit from a lead who has opened 15 emails and downloaded three assets is a different signal from a first-touch pricing page visit. A rules model treats them identically. A predictive model learns the difference automatically.
How do you know your model is working?
The primary signal is SQL-to-close rate. If leads your scoring model marks as sales-ready are converting to pipeline at a higher rate than your pre-model baseline, the model is calibrated correctly. Track this monthly. A well-tuned model should improve SQL-to-close rate by 15 to 30 percent within 90 days of activation.
Secondary signals: sales complaint rate and speed-to-follow-up. If sales is consistently marking scored leads as unqualified, the thresholds are set too low — tighten the sales-ready threshold or the firmographic fit criteria. If scored leads are sitting uncontacted for more than two hours, the routing is broken, not the scoring.
Watch for model drift every quarter. As your content mix changes and new campaigns go live, signal weights need to be reviewed. A pricing page that drove high-intent leads six months ago may now attract a different audience because you changed the content on it. Schedule a quarterly calibration: pull last quarter’s closed-won deals, check which signals they shared, adjust weights where the data no longer matches your original assumptions.
The model is not working if MQL volume stays the same but pipeline does not improve. That usually means one of three things: thresholds are wrong and too many unqualified leads are hitting sales-ready, nurture sequences are not generating the intent signals that cross the threshold, or the data plumbing is broken and scores are not updating in real time. Diagnose in that order. If you would rather have someone build, calibrate, and maintain this system than figure it out yourself, that is what our AI marketing automation agency does — book a 30-min scope call to see how we scope it.