Lead Scoring with Firmographic and People Data: Which Fields Matter

Six fields carry most of a B2B lead score: company headcount, industry, the lead's seniority, their job title, their location, and the company's technology stack. Engagement and intent signals refine the score; fit — those six fields — decides whether the lead is worth scoring at all. And fit has a property most scoring guides skip: it expires. In a Reverse Contact benchmark across technology industries — 100 million profile records observed over three months — 4.8% changed job or company, roughly 18% annualised. A score built on last year's CRM is scoring one person in five who no longer holds the job the score assumes.

Three families of signal, one of which you do not own

Every lead scoring model combines three kinds of evidence.

• Engagement — what the lead did with you: pages visited, emails opened, a demo booked. You own this data; it is as fresh as your event pipeline.
• Intent — what the lead is doing in the market: hiring, raising, researching a category. Partly yours, partly bought, usually noisy.
• Fit — who the lead is and where they work: the six fields above. You do not generate this data. It comes from outside, and it is only as current as the last time someone looked.

Most scoring models spend their design effort on engagement, because it is the part the team controls and the part the marketing tool makes visible. But engagement without fit is noise: a student who reads every blog post scores higher than a VP who read one. Fit is the gate. The rest of this article is about the gate, field by field, from the point of view of the data — where it comes from, how fast it goes stale, and what happens when it is wrong.

Field by field

Company headcount

Source: declared by the company (public profiles, filings) or observed (counted from public employee profiles). Decay: slow for stable companies, fast for anything growing or shrinking — a company that doubles in eighteen months crosses two size bands. When it is wrong: the lead lands in the wrong segment, gets the wrong offer, and is routed to the wrong rep. Headcount is also the field where declared and observed values diverge most; see the section on freshness below.

Industry

Source: declared classification, often one of several incompatible taxonomies. Decay: slow, except for pivots. When it is wrong: usually because two taxonomies were merged without mapping — "Software" in one becomes "Information Technology & Services" in another and a filter silently excludes half your market. Normalise before you score.

Seniority

Source: derived from the job title. Decay: tied to job changes — a promotion changes seniority without changing employer, and your CRM will not notice unless something re-reads the profile. When it is wrong: a manager scored as a director gets a sales conversation they cannot advance; a director scored as a manager gets a nurture sequence and leaves.

Job title

Source: the person's own declaration. Decay: the fastest of the six — this is the field the 4.8% per quarter is about. When it is wrong: everything derived from it (seniority, function, persona) is wrong too. Title is also where naïve matching fails hardest: substring rules that match "Head of Data" also match "Head of Data Entry". Score on normalised function and seniority, not on the raw string.

Location

Source: declared, often at city level. Decay: moderate; remote work has made it less predictive of territory than it was. When it is wrong: territory assignment and compliance rules (which legal basis applies, which rep may contact) misfire.

Technology stack

Source: observed from public signals — job postings, integrations, website technology. Decay: moderate; stacks change at renewal cycles. When it is wrong: you pitch an integration the company no longer needs, or miss one they just adopted.

The score expires: what 4.8% a quarter means for you

The benchmark figure deserves a careful reading. Across technology industries, on a sample of 100 million records observed for three months, 4.8% of people changed job or company. Annualised, that is around 18% — close to one in five — and it is a floor for the decay of your fit data, not a ceiling: it counts only moves, not promotions, title changes within the same company, or companies that changed size band.

Three consequences follow.

• A fixed-date re-score is always late. If you refresh fit data once a year, on average a record is six months stale, and 9% of your leads are being scored on a job they left. Quarterly refresh halves that. Neither removes it.
• The right trigger is the event, not the calendar. When a person's title or company changes, that is the moment to re-score — and often the moment they become a lead again, in a new company, with a new budget. This is what webhook delivery is for: you subscribe to changes and re-score the records that moved, instead of re-reading the ones that did not.
• Freshness is a field. Every record Reverse Contact returns carries a last-updated timestamp. A score that ignores it treats a profile verified yesterday and one verified fourteen months ago as equally reliable. They are not, and the timestamp should discount the score.

This is also why the fit layer belongs in an API you call, not a file you imported. A file is a snapshot. A call is the current state.

From score to routing

A score is a number; routing is what the number does. The two are usually designed by different people and drift apart. Three rules keep them aligned.

• Route on the same fields you score on. If headcount drives the score, headcount drives the segment and the rep. Two sources of truth for company size guarantee that a lead scores as enterprise and routes to SMB.
• Set a service level per band. A top-band lead gets a human within an hour; a bottom-band lead gets a sequence. Write the numbers down, and measure them — speed-to-lead is the one routing metric that reliably moves conversion.
• Re-route on re-score. When an event changes the score, it changes the owner. A lead who moved from a 50-person company to a 5,000-person one is a different account, and probably a different rep's.

Territory is the routing dimension most exposed to bad data: it depends on the location field, which is declared, often incomplete, and less predictive since remote work. Route on company location for enterprise and on person location for SMB, and accept that some percentage will need a human reassignment.

Building the model, in four steps

1. Start from converted customers, not from intuition. Pull the fit fields for the last two hundred deals that closed and the last two hundred that did not. The attributes that separate the two groups are your scoring attributes. Anything that does not separate them is decoration.
2. Define each attribute precisely. "Mid-market" is not a definition; "201 to 1,000 employees, observed headcount, refreshed within 90 days" is. Write down the source, the normalisation and the freshness window for every field.
3. Assign points from the conversion gap. If leads at companies of 201–1,000 convert at three times the base rate, that band earns three times the points of the base band. Let the data set the weights; adjust by hand only where you have a reason you can write down.
4. Set the threshold from capacity, not ambition. The line between "sales-ready" and "nurture" should produce roughly the number of leads your team can actually contact within the service level. A threshold that generates twice that number is a threshold that will be ignored.

What a score cannot tell you

Two limits, stated plainly. A lead score is a ranking, not a prediction: it says this lead is more likely to convert than that one, not that either will. And B2B fit data describes a person's professional situation — role, employer, seniority — not their individual intent. Nothing in a profile tells you whether this particular VP is in the market this quarter. Engagement and intent signals exist to answer that question; fit data exists to decide whether the question is worth asking.

There is a data limit too. Firmographic and people data comes from publicly available professional sources. Companies with a small web footprint, very new companies, and individuals with no public profile will be under-described, and a model that penalises missing data will systematically under-score them. Treat null as unknown, not as zero.

Frequently asked questions

Which fields should a B2B lead scoring model use?

Six fit fields do most of the work: company headcount, industry, seniority, job title, location and technology stack. Add engagement (what the lead did with you) and intent (what they are doing in the market) to refine within the fit bands. Data for the fit fields comes from a company data and people data API rather than from your own systems.

How often should lead scores be recalculated?

On events rather than on a schedule. In technology industries, around 4.8% of professionals change job or company each quarter — about 18% a year — so any fixed schedule leaves a growing share of scores pointing at jobs people no longer hold. Subscribe to changes and re-score the records that moved.

Is job title or company size more important?

Company size decides whether you can sell at all; title decides whether you are talking to the right person. Size is the stronger gate, title the faster-decaying field. Score both, refresh title more often.

What is the difference between lead scoring and lead routing?

Scoring ranks leads; routing assigns them. They should run on the same fields, so that a lead's segment and its owner never disagree, and a change in score should trigger a change in routing.