Where Intent Data Comes From and How Much of It Is Inferred

What is intent data?

In B2B sales and marketing, intent data is behavioral evidence that an account or a person is researching a purchase right now, drawn from what they read, search, ask and post. The unit of work is the signal: one observed action, attached to an identity of some strength, carrying a timestamp. Decide which single action would make you contact someone before you buy any of it, because that decision sets which signals you need.

First-party intent data is observed on your own properties. Third-party intent data is inferred from someone else's traffic.

That line settles most of the confusion in this category, including vendors who sell the second as if it were the first.

What is the difference between first-party and third-party intent data?

First-party intent data records actions taken on assets you own, and third-party intent data models actions taken somewhere you do not control. What separates them is the strength of the link between action and identity: your analytics ties a session to a known contact or a logged-in account, while a third-party feed ties a session to a company through an inferred match. Keep the two in separate columns with separate follow-up rules, because a pricing page visit and a modeled topic surge do not deserve the same call.

Second-party intent data sits between them, meaning another company's first-party data shared with you, such as a review site reporting which accounts browsed your category page.

How is third-party intent data actually built?

Vendors build third-party intent data from two mechanisms: IP-to-company matching and co-op content consumption. Reverse IP lookup maps a visitor's IP address to an organization using registry records, ASN ownership and the vendor's own address graph. The co-op is a pool of B2B publishers that contribute pageview and form-fill logs into a shared data set, where the vendor tags each page with topics and flags an account when its consumption of a topic climbs above that account's rolling baseline.

Every step of that chain is probabilistic, and this is the part vendor pages leave out. An IP resolves to a company only while the visitor sits behind a corporate network, so remote work, mobile carriers, VPNs and residential ISPs either break the match or land it on the wrong company. A surge is a statistical comparison against a baseline, not a statement that a human at that account decided to buy. The output names a company, never the person, and never the problem in that person's own words.

Ask any vendor which mechanism produced each signal, whether the record is account level or person level, and how the match rate was measured. Get the answers in writing before you price the contract.

Which intent signals are worth acting on?

Rank intent signals on three properties: how directly the signal states a problem, how fresh it is, and how tightly it attaches to a person you can reach.

The strongest and rarest signal is a public post where a person describes the problem in their own language, because it carries the problem statement, an identity, a timestamp and a place to reply in one record. No model reproduces that, since the person did the work of writing it down. Collecting it is a social listening job, and it is the signal RedReplier acts on: it monitors Reddit, X, Bluesky, Facebook and Hacker News in one place, ranks mentions by buying intent, and explains with AI why each one was flagged.

Why does third-party intent data decay so fast?

Third-party intent data decays because two clocks run against it: the buying window closes, and the account-to-IP link ages. An evaluation moves from research to a shortlist while the vendor's pipeline is still batching last week's consumption, and a signal that lands after the shortlist is written buys you nothing. The baseline adds a third effect, since a surge is measured against the account's own recent history, so sustained research raises the baseline until the same behavior stops registering as a spike.

Date-stamp every record on arrival, route anything older than the refresh cycle to advertising and nurture, and reserve human follow-up for what is fresher.

How do you score buying intent without fooling yourself?

Score each signal on what the person said, not on the topic they touched, because three different people trip the same keyword. Someone asking "what does everyone use instead of X" is selecting a product and welcomes a reply. Someone posting "X went down again this week" is unhappy with an incumbent and has stated no intention to switch, which makes it a brand monitoring event before it is a sales event. Someone writing "how to configure X for staging" is producing a tutorial and will never buy from the reply. Keyword-only feeds treat all three as intent, and the tutorial writers are the largest single source of false positives.

Score with three factors instead of one:

Intent score = signal strength (1 to 10) x recency multiplier x fit multiplier

Set the recency multiplier to 1.0 within 24 hours, 0.5 within seven days and 0.2 beyond. Set the fit multiplier to 1.0 when the account matches your ICP and 0.3 when it does not. A comment asking the category for an alternative, posted two hours ago by someone inside your ICP, scores 9 x 1.0 x 1.0 = 9.0. A modeled topic surge on that same account, refreshed five days ago, scores 3 x 0.5 x 1.0 = 1.5. One account, one number worth a reply and one worth a retargeting audience.

Keep this score apart from your lead scoring model, which answers who the person is while the intent score answers what they are doing this week. Our reference on buying intent works through the language patterns that separate selection from complaint.

What do you do with intent data once you have it?

Route each signal to the channel that matches its strength, and never send a weak signal to a human. Public posts stating a problem go to a person who replies in the thread while it is open, first-party demo requests go straight to sales, and modeled account surges go to advertising, where a wrong guess costs impressions instead of goodwill. Our B2B lead generation page covers the keyword and website setup behind the public-conversation feed, and RedReplier sends email alerts on a per-plan interval so a fresh thread reaches you before the answers pile up.

Frequently asked questions

What is the difference between first-party and third-party intent data?

First-party intent data comes from assets you own, such as your site analytics, your product and your CRM, and it ties an action to a contact you can identify. Third-party intent data comes from someone else's traffic and reaches you as a model output, matched to a company instead of a person. One blended score hides which is which.

Is third-party intent data accurate?

Accuracy depends on the vendor's identity graph and on how many of your accounts sit behind corporate networks, and no single figure holds across providers. The mechanism sets the ceiling: an IP-to-company match fails or misfires whenever a visitor is remote, on a VPN or on a mobile carrier. Run a hold-out test on your closed-won accounts and count how many the feed flagged before the deal opened.

Can intent data identify a specific person?

Third-party topic surges identify an account, not a person, because the match runs on network address instead of identity. First-party data identifies a person whenever they logged in, filled a form or clicked a tracked email. A public post identifies the handle that wrote it along with the problem it describes, which is why that signal supports a direct reply.

How fresh does intent data need to be?

Fresh enough to reach the buyer before the shortlist closes: hours for a public conversation, and inside the refresh cycle for a modeled surge. Date-stamp every record and let recency cut the score instead of deleting old signals. Aged intent still works for advertising and nurture, and it stops working for a phone call.

How is intent data different from lead scoring?

Lead scoring rates who a lead is, using firmographics, role and history, and it moves slowly. Intent data rates what an account or person is doing right now, and it expires. Keep them as two numbers on one record, because a perfect ICP fit doing nothing and an average fit actively shortlisting need different treatment.

Why do tutorial writers show up as high-intent leads?

Keyword feeds match strings, and a tutorial names the product more times than a buyer does. The writer has no purchase to make, so the reply lands wrong. Filter on the shape of the sentence: a question about what to buy, a complaint about a current tool and a set of setup instructions read differently, and only the first two belong in a sales queue.

What is IP-to-company matching?

IP-to-company matching maps a visitor's IP address to an organization using registry records, ASN ownership and the vendor's own address graph. It resolves to the right company only while the visitor sits behind a corporate network. Remote work, mobile carriers, VPNs and residential ISPs break the match or route it to the wrong company.

What is second-party intent data?

Second-party intent data is another company's first-party data shared with you, such as a review site reporting which accounts browsed your category page. It sits between first-party and third-party because the link between action and identity was built by someone else, then handed to you. Treat it as account level, resolved by the site that collected it, and refresh it on that site's schedule, not yours.

How do you calculate an intent score?

Multiply signal strength (1 to 10) by a recency multiplier by a fit multiplier. Set the recency multiplier to 1.0 within 24 hours, 0.5 within seven days and 0.2 beyond that, and set the fit multiplier to 1.0 when the account matches your ICP and 0.3 when it does not. A comment asking for an alternative, posted two hours ago by someone inside your ICP, scores 9 x 1.0 x 1.0 = 9.0, while a modeled surge on that same account from five days ago scores only 1.5.

Should a modeled account surge go to sales or to advertising?

Send it to advertising. A modeled surge names a company, not a person, and it is a statistical comparison against that account's baseline rather than a decision to buy. Sales works from signals a person can reply to, such as a demo request or a public post stating the problem, where a wrong guess costs goodwill instead of just an impression.