Lead Scoring That Predicts Revenue: A Checklist

Most lead scoring models in Indian B2B companies are decoration. Someone assigns 1-100 points in the CRM, sales ignores the number within three weeks, and the field sits empty again. The people who pay for this are founders and sales ops leads already spending on Google Ads, IndiaMART, expos and WhatsApp campaigns, who still cannot say which leads repaid the spend. A score that predicts revenue is not harder to build — it is just built in a different order.

Start from closed-won deals, not from personas

Your first model should be reverse-engineered from deals you have already closed, because that is the only evidence you have of what revenue actually looks like. Export 12-24 months of closed-won and closed-lost deals from your CRM, even a messy one, and lay them side by side. For each deal, note the firmographics (industry, headcount band, city, GSTIN on file), the source, and the lead's behaviour before the first call. The question you are answering is deliberately narrow: what did the winners have that the losers did not? Almost always it is two or three things, not fifteen. Write those down before you open a spreadsheet.

Score only signals your systems already capture

A signal belongs in the model only if your CRM or website records it automatically today, because a field a salesperson has to fill will be filled badly or not at all. Good candidates, roughly in order of usefulness:

• Page depth on pricing and product pages, tracked per lead rather than per session
• A verified GSTIN or a corporate email domain on the enquiry form, which filters out students, job seekers and competitors
• Whether the lead replied on WhatsApp inside 24 hours of the first message
• Source quality: referrals and inbound demo requests usually outperform purchased lists
• Deal-stage velocity, once a lead is actually in the pipeline

If a signal you believe in is not captured anywhere, fix the plumbing before you build the model. Adding event tracking to an existing site or B2B portal typically takes a couple of weeks of engineering, and it is cheaper than guessing.

Weight by won-versus-lost ratio, not by opinion

Assign each signal a weight proportional to how much more often it appears in won deals than in lost ones, because that ratio is the only thing separating a real signal from a plausible-sounding one. Concretely: for each signal, take the share of won deals that had it and divide by the share of lost deals that had it. A signal present in 60% of wins and 20% of losses has a ratio of 3 and deserves real points. A signal present in 45% of wins and 40% of losses is noise — drop it, however sensible it sounds in a meeting. Cap the model at five or six signals and 100 total points so the number stays readable on a phone screen.

An illustrative example, not a template: 30 points for a verified GSTIN plus corporate domain, 25 for three or more pricing-page visits, 20 for a WhatsApp reply inside 24 hours, 15 for a referral source, 10 for a named decision-maker. That shape is common; your weights should not be copied from it.

Backtest on a holdout month before go-live

Run the score against the most recent month of closed deals, data the model has never seen, because that tells you whether it separates revenue or merely looks tidy in a slide. Sort those leads by score and check two things: whether the top band (say 70+) contains a disproportionate share of closed revenue, and whether any large deal sits in the bottom band. One straggler is normal; a pattern of big deals scoring low means you are scoring the wrong people. If the top 20% of leads by score is not carrying at least half the closed revenue, the weights need another pass. Do this before a single salesperson sees the number.

Expect a rough but usable model somewhere around 200-400 historical deals. Below 100, skip scoring for now and run a short qualification form instead.

Make the score decide who gets called today

A score changes revenue only when it drives routing and response time, so tie each band to an action with a named owner and a clock. A typical setup: 70+ goes to a senior rep and gets a call within 30 minutes during business hours; 40-69 enters a nurture sequence with a call within 24 hours; below 40 stays in email or WhatsApp drip and never reaches a rep's queue. Store the score as a single field on the lead record with a one-line reason, such as "GSTIN verified, 4 pricing visits", because reps trust a number they can argue with. Then review quarterly, since page-visit patterns shift and ad sources decay. A model that was 70% right in January can be guessing by July.

Where to start

Pull last quarter's closed-won deals this week, list the three things they shared before the first call, and check whether your CRM actually records all three. If it does not, fix that first — the scoring can wait a fortnight. Maysan Labs builds this kind of scoring, routing and tracking into CRMs and B2B portals for Indian teams; if you would rather see the finished plumbing first, browse our case studies or look at how we scope custom builds.