Intent Data, From Where the Signal Is Collected to the Number You Get Quoted
Before a vendor quotes you a topic count, learn where the readings come from, what a surge is measured against, and which published numbers hold up.
Intent data is a record of what people at a company read online, matched to the company rather than to the person, and sold as a sign that the company may be shopping. What separates one feed from another is where those readings come from, and there are four different places.
A surge score sits on top of that collection: it is arithmetic against a baseline rather than a judgment, and one provider publishes the window it measures against.
The numbers published about all of this are worth checking before somebody quotes you one. Four vendor pages give four different topic counts for the same underlying feed.
What Intent Data Is, and the Line Between Interest and Intent
Intent data is behavioral information collected online and used to predict buying, and it goes by two other names on vendor pages: buyer intent data and purchase intent data. Cognism and Dreamdata both give all three names for the same thing.
The word doing the work is behavioral. Demandbase describes third-party intent as activity collected across external websites, content networks, research environments and other sources, used to identify patterns of interest around particular topics or categories.
Almost every page selling this is selling it for business-to-business use, and the reason is structural. A single consumer's browsing is one person's browsing. A company's is many people at once, and their reading can be pooled into one account-level pattern.
Interest is not intent, and the gap between them is where most of the disappointment lives.
Dreamdata separates the two by intensity. It defines strong intent as patterns of clicks, scrolls, downloads and keyboard activity around a call to action, measured through engagement, and weak intent as the absence of that engagement.
Demandbase is blunter about it. Its page lists five things intent data can indicate.
- Which accounts are researching topics relevant to your business.
- Whether interest around a topic appears to be increasing.
- Which subjects or problems appear most relevant to an account.
- Which accounts may deserve more marketing or sales attention.
- How an account's research activity compares with normal behavior.
Read that list closely and none of the five is a purchase. Four are statements about reading, and one is a suggestion about where to put attention.
Where the Signal Actually Comes From
The signal comes from four places, and which one your feed uses decides what it can and cannot resolve. Three of the four are third-party, meaning the reading happened somewhere you do not own.
Bombora, which sells the third-party kind, sets out those three on its own site.
- A data cooperative. Traffic, engagement and content-consumption signals unified from a group of publishers, brands, websites and apps, to build one view of a research journey. Bombora names Fortune, Fast Company and Law.com among its co-op members.
- Publisher direct. Traffic and engagement data collected from one publisher or set of sites focused on a niche audience. Deep on its subject, and narrow everywhere else.
- Bidstream. In Bombora's own description, fragmented and surface-level data passed from a publisher's site during the real-time bidding process, giving only a momentary glimpse of visitor activity. Cognism explains the mechanism: publishers collect information about their users and their available ad inventory, then share it with advertisers who can bid on it.
The fourth route is your own website, and it is first-party. When an anonymous visitor arrives, a provider matches the internet protocol (IP) address behind the visit to a database of business records and returns the company name.
Untitled, a provider-comparison site, describes Lead Forensics doing exactly that, and Cognism lists reverse IP tracking among the collection methods.
Consent is a property of the mechanism rather than a separate subject. Bombora says its collection runs on consent-driven protocols, through a proprietary tag on every co-op site.
Untitled's own guidance is that compliance with the General Data Protection Regulation, or GDPR, is not automatic, and that it requires the provider to have a lawful basis for collecting personal data.
There is a second question underneath all four, and it decides how useful the feed is on a Monday morning.
Does the signal resolve to an account or to a person?
IntentData.io, which sells the contact-level version, puts the complaint bluntly. Other third-party sources deliver intent only at the account level, it says, and then guess at contact identities.
Vector says the same from the buyer's side. Account-only data makes it nearly impossible to know who inside the company is showing intent, which pushes teams into targeting whole companies.
That is not a reason to avoid account-level data. It is a reason to know which one you are buying, because an account-level feed and a contact-level feed answer different questions and cost different money.
What a Surge Measures, and What It Is Measured Against
A surge is a calculation, not a mood, and one provider publishes the arithmetic.
Bombora states that it monitors company research activity against 21,600+ topics, and finds elevated intent by comparing an account's most recent three weeks of activity against a 12-week historical baseline, for every account and topic pair.
Two things follow from that shape.
The comparison is relative to the account itself. A large company that always reads about a topic does not surge on it, and a small one that starts reading does, so a surge list is not a list of the busiest accounts.
And the window has a length. Three weeks of activity against twelve weeks of history gives a signal a shelf life measured in weeks rather than quarters.
ZoomInfo publishes the other half, which is what a spike algorithm weighs. It lists five inputs: the amount of content consumed, the number of consumers, the types of content consumed, time on page and scroll speed.
Bombora describes a similar stack in different words, saying it uses natural language processing and named entity recognition to read the context of research, layered with scroll velocity and time-on-page data from its own tag.
Three weeks against twelve, on your own accounts
Put a few accounts in and the comparison above runs on them: each account's most recent three weeks against that same account's preceding twelve. Count whatever you can count, pages, sessions, documents, as long as both boxes hold the same thing.
| The account | Previous 12 weeks | Most recent 3 weeks | Baseline, per week | Recent, per week | Against its own baseline |
|---|---|---|---|---|---|
| 80 | 82 | +2.5% | |||
| 30 | 28 | -6.7% | |||
| 4 | 18 | +350.0% |
The two windows are different lengths, so each total is divided by its own number of weeks before anything is compared. The calculation runs for every account and topic pair, so everything here is one topic. Empty both boxes on a row to leave it out.
Ordered by how much they read
- 1Always reads this topic246
- 2Steady all quarter84
- 3Just started reading54
The three-week total, largest first. This is the list a volume count hands you.
Ordered against each account's own baseline
- 1Just started reading (number 3 by volume)+350.0%
- 2Always reads this topic (number 1 by volume)+2.5%
- 3Steady all quarter (number 2 by volume)-6.7%
The same accounts, each one read against its own preceding twelve weeks.
Always reads this topic reads the most, and Just started reading is the one that has moved. Ranked by how much they read, Always reads this topic is first. Ranked against its own preceding twelve weeks it is number 2, above its own baseline at +2.5%. The same numbers, two different lists.
The comparison is relative to the account itself, which is why a surge list is not a list of the busiest accounts.
Seeded with an illustration rather than measured data, and no row here is a real company. Every figure is arithmetic on the boxes: each window total divided by the weeks in that window, then the recent rate against the baseline rate. The two per-week columns are rounded for display and the comparison uses the totals you typed, so on very small numbers the two will not multiply back exactly. Nothing is scored and no cutoff is applied, because what the section above publishes is the comparison.
Where ZoomInfo's Intent Data Comes From
ZoomInfo's own sourcing is one of the questions people ask about this category, and the honest answer is that the vendor does not publish it.
Its guide to intent data says third-party signals are gathered from publisher networks, review sites such as G2 and TrustRadius, and content syndication platforms. Beyond that it names no supplier and no collection route, and the word bidstream appears nowhere on the page.
ZoomInfo also describes a zero-party route, saying it surveys millions of business professionals who are incentivized to share their priorities, projects and pain points, which is intent that was declared rather than inferred.
For the third-party half, the most specific public statement comes from a competitor. Cognism's provider comparison writes that ZoomInfo is believed to rely on bidstream data and uses machine learning to determine buying signals.
Believed to is a hedge, and a hedge from a rival is not a specification.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/intent-data/"><img src="https://neerajjivnani.com/infographics/intent-data/collection-routes.png" alt="The four places an intent signal is collected, compared side by side: a data cooperative pooling publisher and brand sites, a single publisher's own network, the real-time bidding stream behind online advertising, and reverse-IP matching on your own website, with the resolution each one returns and what each one cannot see." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/intent-data/">Neeraj Jivnani</a></p>Neeraj Jivnani, "Intent Data, From Where the Signal Is Collected to the Number You Get Quoted", neerajjivnani.com, https://neerajjivnani.com/blog/intent-data/Free to republish with a link back to this page.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/intent-data/"><img src="https://neerajjivnani.com/infographics/intent-data/surge-window.png" alt="The published surge calculation drawn as a timeline: an account's most recent three weeks of research activity compared against its own preceding 12-week historical baseline, computed separately for every account and topic pair, with the same absolute reading producing a surge for one account and nothing for another." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/intent-data/">Neeraj Jivnani</a></p>Neeraj Jivnani, "Intent Data, From Where the Signal Is Collected to the Number You Get Quoted", neerajjivnani.com, https://neerajjivnani.com/blog/intent-data/Free to republish with a link back to this page.
First-Party, Second-Party and Third-Party
The three party labels describe who did the collecting, and each one buys something the others cannot. Demandbase names all three.
First-party is what you collect on your own properties: website activity, content downloads, webinar engagement, product interactions and email engagement. Vector calls it the highest-quality intent data because buyers submit it directly, and names its limitation in the same breath, that it only covers people who are already engaging with you.
Second-party is somebody else's first-party data, bought from them directly. TrustRadius, which sells it, defines it as first-party intent data that comes from another company, purchased through a data partnership, and notes that relatively few companies offer it. The main second-party offerings in business software come from review platforms, which is what G2 and TrustRadius sell alongside their listings.
Third-party is collected across the open web by an aggregator and sold as a product. It reaches accounts that have never visited you, which is its whole appeal, and it carries trade-offs the other two do not.
Underneath the party labels are the signal families themselves.
Cognism names search intent, where somebody types a query into a search engine, and engagement data, where they interact with content, and both can appear inside any of the three parties depending on who owns the property the interaction happened on.
Firmographic and Technographic Data Are Not Intent
Two data products are filed alongside intent data on vendor pages, and neither one contains any behavior. Cognism lists firmographic and technographic data among four types of intent data source, and Boomi files both under other sources of buyer intent data.
Read what those two products hold, in the vendors' own definitions.
- Firmographic data is size, location, industry and revenue, in Cognism's wording, gathered from business directories, website forms and data providers. Boomi gives the same four attributes as sector, location, size and revenue.
- Technographic data is the software, hardware and networks a company runs. Cognism says it is often collected by surveys, questionnaires and polls, and Boomi says the same.
Nobody browsed anything to produce either one. They describe what a company is and what it owns, which is a standing condition rather than a change in behavior.
The distinction is worth holding because the two data types answer different questions. Firmographic and technographic data tell you who is worth selling to, and intent data is supposed to tell you when.
The vendors themselves use them exactly that way.
ZoomInfo pairs intent with firmographic and technographic data to narrow a list to accounts that are high-fit, and Demandbase says combining intent with firmographic, technographic, CRM, engagement and buying-group data gives more context for deciding which accounts to prioritize.
Buying a technographic feed and expecting timing out of it is buying the wrong product. It will tell you every company running a competitor's software, and nothing at all about which of them is looking to change.
What the Published Numbers Do and Do Not Say
One provider's feed sits underneath several of the products on sale, and its specification is published in five different versions.
Factors.ai's own comparison table lists Bombora as the third-party source behind Factors.ai, SMARTe, Cognism and 6sense, so the same signal reaches buyers through several different logos.
Here is what each page states about that feed's topic count and its exclusivity, alongside the provider's own figures.
| who publishes it | intent topics | exclusivity claim | co-op size |
|---|---|---|---|
| Bombora, its own Company Surge and Data Co-op pages | 21,600+ | 86% of the data in its co-op, shared exclusively | thousands of B2B media sites, 1,765+ added in two years |
| Untitled, provider comparison | 12,000+ | 70% of its dataset is exclusive | 5,000+ B2B websites |
| Cognism, provider guide | 14,000+ | 70% of websites in the co-op are exclusive | 5,000 B2B sites |
| Lead Onion, 2025 provider fact sheet | 16,000+ | 86% of the dataset is unique | 5,000+ B2B sites |
| Factors.ai, platform comparison | 18,000+ | not stated | 5,000+ premium B2B publisher sites |
The Topic Count, in Five Published Versions
Every reseller figure carries a plus sign, so none of them is false against 21,600.
A page saying 12,000+ topics is telling the truth about a feed with 21,600, in the way a shop saying it stocks 12,000+ items is telling the truth when it stocks 21,600.
A buyer who sizes the feed off the smaller figure still gets it wrong. And the provider's own number appears on none of the four pages.
What a spec sheet records is the day somebody wrote it down. Nothing on these pages says which day that was.
The Exclusivity Claim Carries Three Denominators
The exclusivity figure is the messier one, because the percentage moves and so does what it is a percentage of.
Three of the four resellers state a figure: two say 70% and one says 86%, against the provider's own 86%.
The denominator moves as well, from a dataset to websites in the co-op and back to a dataset, against the provider's wording of the data in its co-op.
Those are not the same measurement. A percentage whose denominator changes between sources is not a specification anyone can act on.
The Figure They Agree On Is Not the Provider's
The number all four resellers agree on is the co-op size, 5,000 sites.
That figure is on none of the three Bombora pages that describe the co-op. Read in September 2026, those pages say hundreds of publishers and brands and thousands of B2B media sites, and that 1,765+ new sites were added over the last two years.
When a specification matters to a decision, take it from the provider's own page and write down the date you read it, because a reseller's spec sheet is a photograph of whenever it was written.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/intent-data/"><img src="https://neerajjivnani.com/infographics/intent-data/published-specs.png" alt="Five published specifications for one intent data feed set side by side, showing topic counts of 12,000 plus, 14,000 plus, 16,000 plus and 18,000 plus from four reseller pages against the provider's own 21,600 plus, and the exclusivity claim splitting between 70 percent and 86 percent across three different denominators." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/intent-data/">Neeraj Jivnani</a></p>Neeraj Jivnani, "Intent Data, From Where the Signal Is Collected to the Number You Get Quoted", neerajjivnani.com, https://neerajjivnani.com/blog/intent-data/Free to republish with a link back to this page.
What Intent Data Cannot Tell You
Intent data is not proof that anybody has decided to buy anything.
Demandbase publishes the boundary as four things it should not be treated as proof of, and that list is the safest thing to hold on to when a surge score arrives.
- That a specific individual has decided to buy. The reading is a company's, pooled across whoever was doing it.
- That an account has selected your company or product. Somebody researching the category is researching the category.
- That every research signal is commercial intent. An analyst, a student and a competitor all read the same pages a buyer does.
- That an account is ready for immediate sales outreach. A signal is a timing hint, and timing hints are wrong often.
Three Properties of the Data Itself
Underneath those four boundaries sit three properties of the data, and each one is stated by a company that sells it.
TrustRadius, on the noise share: only a small percentage of the signals a buyer receives from third-party sellers are actual users looking to buy their product, and the rest is noise.
That comes from a review platform selling the second-party alternative, which is worth knowing when you read it. The share is not quantified anywhere, so plan on a hit rate you cannot see in advance.
TrustRadius again, on exclusivity: third-party intent data is publicly available to anyone who pays for it, so you lose the exclusivity that first-party and second-party data carry, and your competitors likely hold the same account list and are working it the same way.
Vector, on decay: intent data is a snapshot in time, so it goes out of date quickly and leads to missed opportunities or wrong assumptions.
Whether It Works, and What a Performance Number Is Worth
Performance claims in this category come from the vendor whose product they describe, and they arrive without denominators.
Dreamdata answers the effectiveness question in adjectives: a clear lift in metrics, and more leads, engagement, clicks and conversions.
Untitled credits its own product with a 37% reduction in cost per lead and a return on ad spend of four to eight times, neither with a sample behind it.
So the working assumption for any performance number quoted at you is that it has no denominator until somebody shows you one. Ask how many accounts, over what period, measured against what.
Use this chart — embed code and citation
<a href="https://neerajjivnani.com/blog/intent-data/"><img src="https://neerajjivnani.com/infographics/intent-data/what-it-cannot-say.png" alt="The four things a published vendor boundary list says intent data is not proof of, set against the three properties of the data that limit it, showing that a signal is not evidence a person decided to buy, that an account chose you, that every research signal is commercial, or that an account is ready for outreach, alongside the noise share, the non-exclusivity and the decay." width="1200"></a>
<p>Chart: <a href="https://neerajjivnani.com/blog/intent-data/">Neeraj Jivnani</a></p>Neeraj Jivnani, "Intent Data, From Where the Signal Is Collected to the Number You Get Quoted", neerajjivnani.com, https://neerajjivnani.com/blog/intent-data/Free to republish with a link back to this page.
Who Sells It, and How to Judge One
The same names recur across the provider lists in this category: Bombora, ZoomInfo, 6sense, Demandbase, Cognism, G2, TrustRadius, Lead Forensics and Factors.ai. Ranking them is a different job from this one, and who sells a product matters less than one structural fact about the list.
Several of them resell the same underlying feed.
Factors.ai's table names Bombora behind three of the other names on that list, and Cognism's own page says it gets its intent data from Bombora.
So four of those nine names sit on one signal, and choosing between them is choosing an interface, not a feed.
What Actually Separates One Provider From Another
Four criteria separate providers better than a feature list does.
- Data freshness. How often the feed refreshes. If it is not often enough, the moment the account came into market has already passed.
- Breadth and depth of signals. How many kinds of buying signal the provider offers, since combining intent topics with other signals beats relying on one.
- Flexibility of topics. How easily you can change what you are watching, so the feed follows the way your buyers research.
- Display inside the workflow. Whether signals appear in the tools your team already works in, rather than arriving as a file somebody has to upload.
A fifth criterion belongs beside them, because it decides what the other four are describing. Ask which of the four collection routes each topic's signal comes from, and whether the feed resolves to an account or to a person.
That question separates two products that look identical on a pricing page. A bidstream feed and a co-op feed are both third-party intent, and they are not the same purchase.
Putting a Signal to Work
Putting the feed to work splits along the two teams that read it, and they want different things from the same file.
The common thread is that a signal changes the order of a list and the timing of a message, and does not change what is being sold.
What Each Team Takes From the Same File
Marketing uses it to decide where budget goes.
Prioritize the accounts and contacts showing active intent, set channel mix and personalization depth by intent tier, and match the asset to the topic behind the signal.
Sales uses the same file for order and context.
Work the strongest signals first, build a structured follow-up rather than a single touch, and treat a customer's sudden competitor research as an early warning of churn.
Both teams feed the same scoring model. ZoomInfo describes weighting a lead score with intent data to move purchase-ready accounts up, and Vector describes embedding intent in scoring so the hottest opportunities float to the top.
For an account-based program the role is narrow and worth stating narrowly. Intent tells the program which accounts are in market this quarter.
Everything after that, the committee mapping, the tiering and the spend per account, is a separate piece of work the signal does not do.
The Failure Mode, and What to Measure
The temptation is to send the email that says you were seen looking at the website, because the signal feels like proof you were paying attention.
Factors.ai calls that the biggest mistake teams make, and creepy is its word for the email. Sending it turns a timing advantage into a privacy complaint.
Its advice is to use the signal as timing cover instead. Send the case study or the checklist that addresses the problem the account was researching, and do not mention the tracking.
Measuring whether any of it worked gets one short list: conversion rates on leads tied to high-intent signals, deal velocity, pipeline influence, win rates, how much intent shifts the lead score, and which content and channels engage high-intent prospects.
What to Settle Before You Buy Any of It
Four things settle whether a feed will do anything for you, and all four are answerable before money moves.
Get the provenance in writing. A co-op, a single publisher, the bidding stream and a reverse-IP match on your own site are four different products behind one label, and a rival's hedge is not an answer to which one you are buying.
Ask what the score is relative to, and over what window. A surge compared against an account's own baseline behaves nothing like a raw volume count, and a three-week reading against a 12-week history tells you the signal expires.
Ask whether the signal resolves to an account or to a person, and price the difference honestly: an account-level list still needs somebody to work out who inside the company to contact.
Then take the specification from the provider's own page, and note the date you read it.
Which leaves the expectation itself. The feed tells you who has been reading, not who has decided, and every use of it that works treats it as an ordering rule rather than an answer.