Intent Data Scoring: Building a Model That Doesn't Just Reward Website Visits

A vendor's pricing page got visited twice last month by a 40-person marketing team in your ICP. A competitor's product marketer visited the same page fourteen times. If your scoring model counts raw visits, the competitor wins. That's the entire problem with pageview-based intent scoring in one sentence.

This isn't the article about which signals to collect. purple path already covered that ground in firmographic vs. behavioral ICP signal sequencing, which asks which signals exist and in what order you should layer them. This article answers a narrower, more mechanical question: once you have signals, how do you actually weight and score them so the output ranks real buyers above noise?

TL;DR: Raw visit counts fail because they can't tell a buyer from a competitor, a student, or a vendor doing due diligence. A working model scores three signal categories (third-party intent surges, first-party high-intent page visits, content depth and recency) on a weighted basis, then multiplies the result by a firmographic fit score instead of adding it in. In the worked example below, that approach flips the ranking entirely: the account with 60 raw pageviews drops to last place, and the account with 12 pageviews but two pricing-page visits and a confirmed ICP fit comes out on top. Decay the inputs (pricing-page visits lose relevance within about a week, topic surges within about a month) and recalibrate the weights on a quarterly cadence, because models rot the moment your product, pricing, or ICP changes.

Why Raw Pageview Counts Reward the Wrong Accounts

A raw visit-count model treats every pageview as equally meaningful. It isn't. Gartner's research on B2B buying behavior found that buyers spend only 17% of their total purchase-journey time meeting with potential suppliers at all, with the rest spent on independent research, comparison, and internal alignment (Gartner, "Future of Sales," widely cited via Gartner's B2B sales research). Most of that independent research happens anonymously, across dozens of domains, long before anyone fills out a form. A model built only on your own site's pageviews is already missing most of the journey, and the part it does see gets contaminated by three kinds of visitors who look identical to a buyer in a raw count:

Competitor staff. Product marketers, competitive intelligence analysts, and sales reps at rival vendors visit pricing and feature pages repeatedly. They generate more pageviews than most real buyers, because checking a competitor's site is their job.

Students and researchers. University domains show up constantly in B2B web analytics, reading blog content for coursework, with zero purchase authority and zero budget.

Vendors and partners. Agencies, consultants, and potential integration partners browse a site to scope a partnership or a client recommendation, not a purchase.

Forrester has published its own running catalog of these failure modes in its analysis of common intent data mistakes, and the underlying issue in nearly every case traces back to the same root cause: treating volume as a proxy for intent instead of treating it as one input among several. A model that can't separate these three visitor types from a genuine in-market account isn't an intent model. It's a web traffic counter with a sales-sounding name.

The Signal Categories a Scoring Model Actually Needs

A working model separates signals into categories that behave differently and deserves different weight. Three of them should be summed into a base score. The fourth should multiply that base score, not add to it, because fit and intent answer different questions: intent asks "are they active," fit asks "could they ever buy."

Third-party intent data from providers like 6sense, Bombora, or Demandbase exists precisely to catch the 83% of the buying journey that happens before a prospect ever lands on your domain. First-party page visits stay the single strongest within-category signal, because nobody accidentally visits a pricing page fourteen times. Content depth and recency catch the difference between someone skimming one blog post and someone working through four case studies and a webinar in the same week. None of these three should ever run without the fourth: a firmographic fit score, built the same way you'd build it for your TAM list, that gates the final number.

Building the Weighted Formula

The formula itself is simple on purpose: Base Score = (Third-Party Intent x 0.30) + (First-Party High-Intent Pages x 0.40) + (Content Depth/Recency x 0.30), then Final Score = Base Score x Firmographic Fit Multiplier. The weighting percentages above aren't arbitrary; they reflect that a pricing-page visit on your own domain is a harder signal than a topic surge you can't directly verify, so first-party intent gets the heaviest single weight.

Here's the math on three hypothetical accounts, scored 0-100 on each component, to show what the formula does that a raw pageview count can't.

Ranked by raw pageviews, the order is C (60), A (40), B (12), dead last being the only account that was ever going to buy anything. Ranked by final score, the order flips completely: B at 100, A at 15.6, C at 2.85. Nothing about Account B's traffic volume was impressive. Two pricing-page visits and a confirmed topic surge, multiplied by a strong firmographic fit, outscored sixty pageviews of blog reading on an educational domain with no fit at all. That's the entire argument for weighted scoring over raw counting, expressed in seven numbers instead of a paragraph of adjectives.

Time Decay: An Old Signal Isn't Worth What a Fresh One Is

A model that never decays its inputs will keep crediting an account for a pricing-page visit that happened two quarters ago, long after that buying window closed or moved to a different vendor. Decay rates should differ by signal type, because a pricing-page visit and a month of sustained content consumption don't age the same way.

The point of decay isn't to punish old engagement; it's to stop a six-month-old demo request from still outranking this week's pricing-page visit. Without it, your highest-scored accounts skew toward whoever engaged earliest, not whoever is active now, which is the opposite of what a sales team chasing a live deal needs.

Firmographic Fit Belongs in the Multiplier, Not the Sum

Additive models let enough intent activity compensate for terrible fit, which is exactly backward. A 15-person agency that isn't in your ICP at all can still rack up a high intent score through sheer page volume if fit is just one more line item in the sum. Make it a multiplier instead, and a fit score near zero drags the final number toward zero no matter how much raw activity preceded it, the way Account C's 90-point content score still landed at 2.85 once multiplied by a 0.1x fit. This is also where the two purple path pieces connect without repeating each other: firmographic fit itself, and how to sequence it against behavioral data, is the subject of the ICP signal sequencing article; this article assumes you already have that fit score and shows how to fold it into the math.

Model Governance: Recalibrate Before It Rots

A scoring model built once and never touched again degrades for the same reason any sales process degrades: the market underneath it moves. Pricing changes, new competitors enter the category, a new feature launch shifts what "high-intent page" even means, and your ICP itself gets redefined after a round of funding or a new product line. A sensible recalibration cadence has three checkpoints: review weight percentages quarterly against win-rate data from closed deals, re-verify which specific pages count as "high-intent" every time pricing or packaging changes, and re-run the firmographic fit criteria whenever the TAM list gets rebuilt. Once the model is scoring accounts, route the output into the systems your sales team actually works from rather than leaving it in a dashboard nobody opens; purple path's guide to integrating intent data into CRM and marketing automation covers the handoff mechanics. And if any part of the model ingests third-party behavioral data on EU-based contacts, check it against what's legally trackable under GDPR before it goes live, not after legal asks why a German contact's browsing history is sitting in your CRM.

Frequently Asked Questions

Is intent data scoring the same thing as lead scoring?

No. Traditional lead scoring rates individual contacts based on form fills, email opens, and job title. Intent data scoring rates accounts, often anonymous ones, based on research activity across the open web and your own site, well before any individual fills out a form.

Can a model run on first-party signals alone, without a third-party intent provider?

It can, but it will only ever see the part of the buying journey that touches your own domain. Given that most B2B research happens off-site before a prospect self-identifies, a first-party-only model is working from a smaller, later-stage slice of the actual buying journey. It's a reasonable starting point, not a finished model.

How often should the weights actually change?

Quarterly is a defensible default for the weight percentages themselves. The underlying inputs, which pages count as high-intent and which firmographic criteria define fit, should get re-checked any time pricing, packaging, or ICP definitions change, which can happen faster than quarterly.

Should sales and marketing use the same score?

They should use the same underlying model, but the threshold for action can differ. Marketing might act on a lower final-score threshold to trigger nurture content, while sales development waits for a higher threshold before an outreach sequence starts, since a false positive costs sales a wasted call and costs marketing nothing more than one extra email.

How is this different from the firmographic vs. behavioral signals article already on the blog?

That article answers which signals to collect and in what sequence to apply them when building an ICP. This one assumes you've already answered that question and covers the weighting formula and math that turns those signals into one usable account score.

Get the Model Built, Not Just the Framework

A formula on a page is one thing. A scoring model wired into your CRM, decaying correctly, and recalibrated on a schedule your RevOps team actually keeps is another. purple path, the Vienna-based B2B GTM and ABM agency founded by Andy Culligan (formerly CMO at Leadfeeder, Exponea, and Emarsys), builds exactly this kind of scoring infrastructure for B2B SaaS teams who are tired of a dashboard that ranks a competitor's product marketer above their actual buyer. Get in touch with purple path to talk through what a weighted model would look like against your own ICP and pipeline data.