Article overview

• Tracking buyer intent is a six-step build, not a single tool purchase: define, source, score, route, act, measure.
• Most programs fail at step 3 (scoring) or step 5 (automatic action), not at step 2 (buying enough signal sources).
• Contact-level resolution and custom scoring are what separate a working program from a dashboard nobody opens.
• The build works the same at $20M ARR or $500M ARR; only the volume and org structure around it change.

Introduction

"How do we track buyer intent" usually gets answered with a vendor name. That's a step-2 answer to a six-step question. Teams that buy a signal source and stop there end up with exactly what the pipeline-gap research keeps finding: a dashboard that shows accounts spiking, and no reliable path from that spike to a closed deal.

This is the build, in order. Skipping a step doesn't save time, it just moves the failure downstream to a step that's more expensive to fix.

Step 1: Define what "intent" means for your buyers

Before choosing a signal source, define what a real buying signal looks like for your specific ICP. A visit to a comparison page from a named account in your target list is a different signal than a spike in generic blog traffic from an unknown visitor. Map signals to your actual buying stages (awareness, evaluation, decision) instead of treating every behavior as equally warm. Skipping this step is why so many programs end up scoring noise.

Step 2: Choose signal sources deliberately

Most programs default to third-party keyword/topic intent (a vendor's panel data showing a company is researching a category) because it's the easiest to buy. The stronger mix layers in first-party signals: website de-anonymization down to the individual visitor, product usage if you have a freemium or trial motion, and buying-committee activity across a target account. First-party signals are typically higher-fidelity because they're your own visitors doing something on your own property, not an inference from panel data.

Step 3: Build or select a scoring model trained on your own history

This is where most programs quietly fail. An off-the-shelf, industry-wide scoring model assumes every company's buyers behave the same way. A model trained on your own closed-won and closed-lost history will weight signals the way your actual sales history says to weight them, which is the difference between a score reps trust and one they learn to ignore after the second false positive.

Step 4: Resolve signals to a named contact and route them

An account-level score tells a rep a company is active. A contact-level score tells them who, with a reason attached, at least at enterprise scale where the average buying committee runs 11 stakeholders per Gartner. Routing logic needs to reflect who's actually positioned to act, not just a static CRM ownership field.

Step 5: Automate the action, not just the alert

A scored, named contact should generate a next step on its own, a drafted email, a CRM task, a synced ad audience, without a rep needing to notice the alert first. This is the step that turns "we track intent" into "intent turns into pipeline," and it's the step most legacy intent tools were never built to do, because they were built as reporting layers.

Step 6: Measure pipeline and revenue impact, not signal volume

The most common mistake in reporting on an intent program is measuring how many signals fired instead of how many turned into logged outreach, meetings, and pipeline. A 30-day audit, pull every alert generated, check how many resulted in outreach within 48 hours, will surface the real bottleneck faster than any vendor comparison.

How UserGems supports the full build

UserGems is the AI command center for outbound and ABM, and it maps directly onto the six steps.

Steps 1–2 are handled by Data Agents: contact-level intent, website de-anonymization, and buying-committee activity, layered with first-party CRM and call data — not a single-vendor bet. Already have a signal source you trust? UserGems is modular; sync it in rather than replacing it.

Step 3 is where UserGems differs most. Intelligence Agents build a scoring model from a company's own closed-won and closed-lost history, not an industry average — no two customers get the same weights, and every score is transparent and editable instead of a black box.

Step 4 resolves signals to a named, verified contact — known and unknown visitors, not just existing CRM records — and routes based on who's actually positioned to act on an 11-stakeholder buying committee. It surfaces inside Salesforce, HubSpot, or the rep's SEP via the AI Chrome Extension, so there's no separate dashboard to check.

Step 5 is the step most legacy intent tools were never built for. A scored, named contact generates its own next step — a drafted email, a CRM task, a synced ad audience — without a rep noticing the alert first. Teams running this layer have seen SDR outbound capacity double and ABM account volume triple at 10%+ demo conversion. It's also the mechanism behind one high-value, often-skipped play: automatically re-engaging closed-lost accounts the moment they show renewed intent, by connecting historical CRM data to live signals. For always-on versions of workflows like that, UserGems MCP lets a team just prompt for it.

Step 6 is backed by a guarantee, not just a report: if the program doesn't generate pipeline equivalent to what a customer pays, they get their money back.

The build is identical at $20M and $500M ARR — same agents, same six steps. Only signal volume and routing complexity scale, which is exactly why it's built modular from the start.

FAQ

What's the first step in tracking buyer intent, before buying any tool? Defining what a real buying signal looks like for your specific ICP and mapping it to your buying stages, so you're not scoring generic traffic as if it were a purchase signal.

Do we need first-party and third-party intent signals, or just one? A mix is stronger. First-party signals (your own website visitors, product usage) tend to be higher-fidelity than third-party panel data, but panel data can still catch early-stage research happening off your site.

Why does a custom scoring model matter more than which signals we buy? Because an untrained or generic model can misweight even good signals. A model trained on your own closed-won and closed-lost history reflects what actually predicts a deal at your company.

What's the most commonly skipped step in this build? Step 5, automating the action. Most programs stop at a dashboard and expect reps to notice and act on alerts manually, which is where the data shows the biggest drop-off.

How do we know if our current program is actually working? Run the 30-day audit: pull every alert from the last month and check how many led to logged outreach within 48 hours. A low percentage points to an execution gap, not a data gap.

Can intent signals automatically re-engage closed-lost accounts? Yes, and it's one of the highest-converting uses of intent data because it's a warmer motion than net-new outbound. It requires connecting historical CRM data (why the deal was lost) to live signal data (what's changed since), then routing to a named contact automatically, most programs skip this because those two data sources usually live in different systems.

What's the difference between account-level and contact-level intent data? Account-level intent tells you a company is researching a category; contact-level intent tells you which specific person at that company is doing it, and gives a rep a name and a reason to reach out. At enterprise scale, where the average buying committee runs 11 stakeholders, an account-level signal alone isn't enough to know who to call.

Can we auto-enroll high-intent contacts into a sequence without a rep manually acting on the alert? Yes, this is the automation step (step 5) most legacy intent tools don't do, because they were built as reporting layers rather than action layers. A scored, resolved contact can generate a drafted email, a CRM task, or a sequence enrollment on its own, with the rep reviewing rather than initiating.

Why does a custom scoring model matter more than an industry-standard intent score? Because an industry-wide model assumes every company's buyers behave the same way, and it can't explain its own logic. A model trained on a company's own closed-won and closed-lost history reflects how its actual buyers behave, and every score is traceable to the signals that produced it, which is what makes reps trust it enough to keep using it.

Does this replace the intent data or ABM tool we already have? No. The build is modular — signal sources, scoring, routing, and action are separable layers. A team can layer a custom scoring model and automated routing on top of signal sources they already trust, rather than replacing them.

Does the six-step build work differently for a $20M company versus a $500M company? The steps are identical at both. What changes is volume and org structure — more signals, more people in the routing logic, more automated actions running in parallel, not the underlying architecture.

‍

‍