First-Party Intent Data: Building Buyer Signals You Own

Most B2B teams' first experience of intent data is a subscription: a vendor sells a feed of companies "researching your category", the feed arrives weekly, the reps ignore it within a month. The signals are plausible but unverifiable, the accounts rarely convert above baseline, and nobody can explain why a given company appeared. The feed is cancelled at renewal and "intent data" is filed under things that do not work.

The category is fine. The source was wrong. Intent data you buy describes what someone might be doing somewhere on the internet. Intent data you own describes what a specific account did on your site, in your product, in your emails and at your events, this week. One is a rumour. The other is a record.

What first-party intent actually is

Any observable behaviour by a known or resolvable account that indicates where it is in a buying process. Concretely, in rough order of strength:

• A pricing or comparison page viewed, especially more than once, especially by more than one person from the same company.
• A high-intent document downloaded: a security overview, an integration guide, a pricing sheet.
• A reply to outbound, of any kind, including a polite no with a timeframe.
• Repeat visits from one account within a short window.
• Product usage crossing a threshold, for anyone with a trial or free tier.
• A webinar or event attended, weighted by how much of it they stayed for and whether they asked a question.
• A job posting for the role you sell to, or a hire into it.
• A tool appearing or disappearing from their stack.
• Engagement with content, weighted low, because reading a blog post is not intent.

Two properties separate this from bought data. Every signal is checkable: you can point to the event and the timestamp. And every signal is yours: it costs nothing per record, it does not expire with a subscription, and no competitor is receiving the same feed.

The infrastructure it needs

Intent that lives in five tools is not intent, it is five reports. The signals above arrive through analytics, the CRM, the email tool, the product database and a data provider. Unless they land in one place, resolved to one account, nobody can see the pattern that matters: three people from one company, on pricing, twice this week, two days after their VP of Marketing was hired.

That one place is a first-party signal ledger: server-side event capture with account resolution, consent recorded on every event, and a schema every downstream system reads. Building it is the actual work of "doing intent data". The scoring and routing are comparatively easy once the events exist in one table.

Scoring without fooling yourself

Three rules, all of them regularly broken.

Keep fit and intent separate. A score that mixes "is this a good customer" with "are they active right now" produces a number that cannot be acted on. A perfect-fit account with no activity and a poor-fit account browsing constantly can score identically and mean opposite things. Two scores, and a routing matrix: high fit plus high intent goes to a rep now; high fit plus low intent goes to nurture with a trigger watch; low fit plus high intent gets investigated, because your profile may be wrong. The Lead Scoring Ruleset has the full model.

Decay everything. A pricing visit four months ago is not intent. Every behavioural point has a half-life. Without decay, scores only go up and the queue fills with accounts that were interested last spring.

Cap repeated behaviour. Fifty blog visits from one enthusiastic reader should not outrank one pricing visit from a buying committee. Each event type has a maximum contribution.

Then validate: run the model against the last two quarters of closed-won and closed-lost. What share of the wins would it have surfaced, and how much noise would the reps have received? A model that surfaces everything surfaces nothing. If it does not separate your own history, it will not separate the market.

From score to queue

The failure after scoring is the dashboard. Scores go on a screen, the screen is reviewed in a Monday meeting, and by Wednesday the signal is stale. Intent has a half-life measured in days.

The output of an intent system is a queue, not a chart. Each item is an account, the signals that put it there, the person to contact, and a suggested first line that references the actual signal. It arrives where the rep already works, with a response time attached. If a high-fit account hits pricing three times and nobody has touched it within two working days, that is a process failure with a name on it.

Agents do the assembling here well: resolving the account, pulling the research, drafting the touch into a review queue. A person decides whether to send. That combination is the allbound motion working as designed.

Where bought data still helps

Third-party intent is useful for one thing: widening the top of the funnel to accounts that have never touched you. Treated as a weak signal, weighted accordingly, and never routed to a rep on its own, it can suggest which accounts to add to an awareness audience or a research list. Treated as intent, it produces the disappointing feed most teams remember.

The order is the point. First-party first, because it is yours, it is checkable and it is where the buying committee actually shows up. Third-party last, and only as a hint.

Frequently asked questions

What is first-party intent data?

First-party intent data is the record of observable buying behaviour by identifiable accounts on channels you own: pricing and comparison page visits, high-intent downloads, outbound replies, repeat visits, product usage thresholds, event attendance, and observable changes such as relevant hires. It differs from third-party intent data in that every signal is checkable, costs nothing per record, and is not sold to your competitors.

Why does third-party intent data usually disappoint?

Because it describes possible research behaviour somewhere on the internet, cannot be verified, and is delivered to every vendor in the category at once. Routed to reps as if it were intent, it produces low conversion and gets ignored. It is useful only as a weak signal for widening awareness audiences.

How should intent be scored?

Separately from fit, with behavioural points that decay over time and are capped per event type, routed through a matrix rather than a single threshold. The model should be validated against the last two quarters of won and lost deals before it goes live, and reviewed against outcomes quarterly.

What infrastructure does first-party intent require?

A first-party signal ledger: server-side event capture with account resolution, consent recorded on each event, and one schema every downstream tool reads. Without it, signals live in separate tools and the patterns that matter, such as several people from one company on pricing in one week, are invisible.