Most lead scoring models at growth-stage companies are activity scoring models dressed up as predictive tools. A contact gets +5 for opening an email, +10 for visiting the pricing page, +15 for downloading a whitepaper. The score accumulates until it crosses an MQL threshold, the sales team gets a notification, and the lead turns out to be a student doing competitive research or a freelancer who could never afford the product. Activity scoring tells you who is engaging with your content. It says nothing about who is likely to buy. The distinction matters because a broken MQL definition erodes the sales team's trust in marketing-qualified leads, and that trust, once lost, is hard to rebuild.
The difference between activity scoring and predictive scoring
Activity scoring rewards behavior. Predictive scoring weights behaviors by their observed correlation with revenue outcomes in your historical data. These are not the same thing. A pricing page visit is a useful signal, but a pricing page visit from a 200-person SaaS company with a VP Operations title is a very different signal from the same visit from a Gmail address with no firmographic data. A predictive model incorporates firmographic fit, company size, industry, revenue band, geography, as a multiplier on behavioral signals. The lead who visited the pricing page, works at a growth-stage fintech in Bangalore, and holds a "Head of Revenue" title is not the same MQL as the lead who visited the same page with no identifiable company. One deserves a same-day sales call. The other deserves a nurture sequence while Clay or Apollo enriches the record to confirm fit.
How to build the model from your historical CRM data
The data you need is already in your CRM if your attribution and firmographic fields are reasonably clean. Pull every closed-won deal from the last 18 months and identify the lead source, company size, job title, and the behavioral signals present at the time the lead was generated, not at the time of close. Then pull every lead that converted to MQL but never became SQL or closed. Compare the two populations: what firmographic and behavioral attributes appear significantly more often in the closed-won group? This analysis does not require machine learning or a data scientist. A cross-tab in Google Sheets will surface the top three to five attributes that predict revenue in your specific business.
The two-tier implementation in HubSpot or Zoho
Once you have identified the predictive attributes, implement the score in two tiers rather than a single linear number. Tier 1 is ICP Fit Score: company size, industry, geography, and job title. This is a static score calculated at lead creation from firmographic data, enriched via Clay or Apollo if the CRM does not have the data natively. Tier 2 is Intent Score: behavioral signals weighted by your historical analysis, with higher weights on high-intent signals like pricing page visits, demo request page views without form submission, or return visits within 72 hours of a first session. The MQL threshold fires when Tier 1 is above a minimum fit threshold and Tier 2 crosses the behavioral threshold. Leads who score high on behavior but low on fit go into a nurture sequence.
Tiered scoring by company revenue band: how to design it
Company revenue is often the single strongest fit attribute in B2B, because it stands in for budget, deal size and buying complexity at once. A tiered scoring algorithm based on revenue does not add points in a straight line as revenue grows. It splits annual revenue into bands and gives each band a fixed number of fit points, with the most points on the bands where you actually win and close at a good deal size.
How to choose the bands. Start from your own closed-won and closed-lost deals, not from round numbers. Plot them by the customer's annual revenue and look for the breaks: the size below which buyers cannot afford you, the range where most of your won revenue sits, and the size above which deals slow down or go to enterprise competitors. Five to seven bands are usually enough. Fewer hides real differences; more leaves too few deals in each band to trust the pattern. Keep the band edges identical in your CRM, your enrichment tool and your reports, and define them in the currency you sell in (an India-focused team might use ₹ crore bands rather than converting dollar bands).
A starting points table. The table below is an illustrative starting point for a product whose best customers are mid-market companies. It is not a benchmark and the points are not derived from any dataset. Shift the peak to wherever your own won deals cluster, then recalibrate as described in the next section.
Unknown revenue scores zero, not negative. If enrichment cannot find a revenue figure, give the lead no revenue points and let the other fit attributes and intent decide. A negative score for missing data punishes the record for a gap in your tooling, not for poor fit.
| Annual company revenue | Fit points (out of 25) | How to read it |
|---|---|---|
| $0 to $1M | 0 | Usually below the budget threshold; route to self-serve or nurture |
| $1M to $5M | 5 | Possible fit if other attributes are strong |
| $5M to $10M | 10 | Edge of the core range |
| $10M to $30M | 20 | Core range in this example |
| $30M to $50M | 25 | Peak band in this example: highest win rate times deal size |
| $50M to $200M | 15 | Larger deals, longer cycles and more competition |
| $200M+ | 5 | Enterprise buying process; score up only if you win here |
| Unknown | 0 | Enrich the record; do not penalise missing data |
Illustrative only. Replace the bands and points with figures from your own closed-won data.
Combining revenue bands with fit and intent, and recalibrating from closed-won data
Revenue is one part of the fit score, not the whole score. In the two-tier model above, the revenue band sits inside the ICP fit score alongside employee count, industry, geography and job title. One workable split is to cap revenue at 25 of 100 fit points, so a lead from the perfect revenue band with the wrong industry and a junior title still does not qualify. Intent stays a separate score built from behaviour. The MQL rule then needs both: a minimum fit score and a minimum intent score. The Lead Scoring Logic Architect combines fit and behaviour into one score and a routing decision, and the Lead Grading Calculator grades the fit half on its own, A to F, which is a useful check on whether your revenue bands and other firmographics are weighted sensibly.
Recalibrate the band points from closed-won data. Once you have two or three quarters of opportunities, pull every opportunity created in the period with its revenue band, its outcome and its deal value. For each band, work out:
- Win rate: closed-won opportunities divided by all closed opportunities in that band.
- Average won deal value in that band.
- Expected revenue per opportunity: win rate multiplied by average won deal value.
- Band points: the band's expected revenue per opportunity divided by the best band's figure, multiplied by the maximum revenue points (25 in the table above).
An illustrative example: if the $30M to $50M band wins 30% of opportunities at $40,000, its expected revenue per opportunity is $12,000 and it keeps the full 25 points. If the $200M+ band wins 10% at $90,000, its figure is $9,000, which earns 25 × 9,000 ÷ 12,000, about 19 points, far more than the 5 points in the starting table. That is exactly the kind of correction recalibration is meant to surface.
Two cautions. Merge neighbouring bands when a band has too few closed opportunities to give a stable win rate, because a handful of deals can swing the points either way. And recalculate on the same schedule as the rest of the model, at least every six months, because revenue bands drift as you move upmarket or launch a cheaper plan.
Calibrating the model against actual revenue outcomes
A lead scoring model that is not calibrated against revenue outcomes within 90 days of launch is already drifting from reality. The calibration process is simple: compare the MQL-to-SQL conversion rate before and after implementing the model, and compare it by score band. Leads who cross the MQL threshold in the top score quartile should convert to SQL at a substantially higher rate than leads in the bottom quartile. If the conversion rates are flat across score bands, the model is not predicting anything. Review the closed-won deals from the last quarter and check what score they carried at MQL. If the average score of your closed-won leads is not materially higher than your average closed-lost lead score, the weighting is wrong. Rebuild the model every six months as your ICP, product, and target market evolve.
FAQ
It rewards engagement, not buying likelihood. Points for email opens, pricing page visits and whitepaper downloads accumulate until a lead crosses the MQL threshold, and sales gets a student or a freelancer who could never afford the product. A broken MQL definition erodes sales trust in marketing leads, which is hard to rebuild.
No. Pull every closed-won deal from the last 18 months and every MQL that never became SQL or closed. Compare the firmographic and behavioural attributes present when each lead was generated. A cross-tab in Google Sheets will surface the three to five attributes that predict revenue in your business.
Split annual revenue into five to seven bands that match where your own deals are won, and give each band fixed fit points rather than scaling points with revenue. Put the most points on the band with the highest win rate times average deal value, give unknown revenue zero points, and keep revenue to about a quarter of the fit score. Recalibrate the points from closed-won data every six months.
Calibrate within 90 days of launch. Compare MQL-to-SQL conversion rates before and after, and by score band. Top-quartile leads should convert substantially better than bottom-quartile leads. If rates are flat, or closed-won leads did not score materially higher than closed-lost, the weighting is wrong. Rebuild every six months.