In this Article
B2B intent data promises to tell you which companies are researching what you sell before they contact you. Whether that is useful depends almost entirely on how the signal was collected, and that is the part vendor decks skip.
This guide walks the supply side: the three collection models, how a browsing signal gets attached to a company name, why bidstream is the controversial one, and the single test that separates a useful provider from an expensive one.
Key Facts
- Intent data is a probability, not an event. It says an account showed unusual research activity on a topic, which is weaker than any vendor’s marketing suggests.
- Three collection models produce nearly all of it: first-party signals, publisher co-ops, and bidstream. They differ enormously in accuracy and in privacy exposure.
- Account resolution is the weak link. Mapping an IP or a device to a company is reliable for large offices and unreliable for remote workers, which is most of the market now.
- Bidstream data raises real questions. It is derived from advertising auctions, and regulators in several jurisdictions have examined whether that use was ever consented to.
- The honest benchmark is lift, not accuracy. Compare conversion for intent-flagged accounts against a random control from the same list, or you are measuring your own targeting.
Where does intent data come from?
Three models, and every vendor is a blend. We call it the 3-part intent supply model.
| Model | Signal | Accuracy and exposure |
|---|---|---|
| 1. First-party | Activity on your own site, docs, pricing page, emails | Most accurate and least risky; also the smallest volume |
| 2. Publisher co-op | Content consumption across a network of B2B publishers that disclose participation | Reasonable accuracy, disclosed collection, topic-level only |
| 3. Bidstream | Data observed in programmatic advertising auctions | Largest volume, weakest provenance, the model regulators have questioned |
Vendor-by-vendor coverage sits in our roundup of B2B data providers. The ranking is uncomfortable for buyers because it runs opposite to volume. The signal you can most defend is the one you already have and mostly ignore: your own documentation, pricing and comparison pages, which tell you more about a real evaluation than any third-party topic score.
How does a browsing signal become a company name?
Through account resolution, which is the step that decides whether the whole dataset is worth anything.
The classic method maps an IP address to a company via registry ownership and office netblocks. That worked when people sat in offices. It degrades badly for remote and hybrid work, where the address belongs to a consumer ISP, and it fails differently for large enterprises behind shared egress, where thousands of employees share a handful of addresses and the topic signal becomes noise.
Modern vendors supplement with device graphs and with self-identified signals such as form fills and email clicks. Each addition improves coverage and adds assumptions. The question to ask a vendor is not how accurate resolution is overall, but what share of their signal for your segment comes from office netblocks versus inference, because a mid-market SaaS target list is mostly remote.
What are the privacy questions?
Bidstream is the contested one. Data generated for the purpose of serving an advertisement, then retained and resold as a behavioural signal, sits awkwardly against purpose limitation under the GDPR and has drawn regulatory attention in Europe. A vendor’s assurance that data is aggregated to account level does not by itself resolve the question of how it was collected.
Account-level is not automatically anonymous. A signal narrowed to a small company with three employees is close to a personal one, and pairing intent with a named contact combines them outright.
Do your own diligence. Ask for the collection model per data source, the legal basis claimed, and whether individual-level identifiers exist anywhere in the pipeline. If a vendor cannot answer in writing, that is the answer. General information, not legal advice.
How do you test whether it works?
Measure lift against a control, not accuracy against intuition.
- Split the same target list. Half worked with intent flags, half worked in the normal order, same reps, same messaging, same period.
- Compare meeting and opportunity rates, not open rates. Intent should change who converts, not who clicks.
- Run it long enough to clear one sales cycle. Four weeks on a ninety-day cycle measures nothing.
Most pilots skip the control, prioritise the intent-flagged accounts, see better numbers and conclude the data works. What they measured is that prioritising any subset beats working a list alphabetically.
| Situation | Buy intent data when | Skip it when |
|---|---|---|
| Target market size | Thousands of accounts, too many to work evenly | Under a few hundred, where research beats scoring |
| Sales motion | Outbound with a real SDR capacity | Product-led, where first-party signals are richer |
| First-party data | Already instrumented and being used | Not yet instrumented, which is the cheaper fix first |
Where does collection infrastructure fit?
Intent vendors run large-scale public web collection alongside their signal pipelines: tracking which companies publish job postings, funding news, technology changes and product launches. Those are public, defensible signals, and they are collected with ordinary crawling infrastructure.
That is the layer we sell. DataImpulse residential proxies give geo-accurate exits at $1 per GB across 195 countries, which matters because job boards, news sites and company pages serve different content by region. A proxy changes the exit address of a request; it has nothing to do with the lawful basis for holding a signal, and no vendor should imply otherwise.
Related reading: data enrichment sources, technographic data, proxies for web scraping.
Frequently Asked Questions
What is B2B intent data?
A probabilistic signal that a company is researching a topic, derived from content consumption rather than from a stated interest. It comes from three sources: activity on your own properties, publisher co-ops where participating sites disclose data sharing, and bidstream data observed in advertising auctions.
Is bidstream intent data legal?
It is contested rather than settled. Data generated to serve an advertisement and then retained and resold as a behavioural signal sits awkwardly against purpose limitation under the GDPR, and European regulators have examined the practice. Ask vendors for the collection model and legal basis per source in writing. General information, not legal advice.
How accurate is account resolution?
It depends heavily on your segment. IP-to-company mapping works for large offices with owned netblocks and degrades for remote workers on consumer ISPs, which is now most of the mid-market. Ask what share of a vendor’s signal for your specific segment comes from netblock matching versus inference.
How should I test an intent data provider?
Split one target list: half worked with intent flags, half in normal order, same reps and messaging, run for at least one full sales cycle. Compare opportunity rates rather than open rates. Without a control group you are measuring the benefit of prioritising any subset, not the data.
Do I need first-party intent data first?
Usually yes. Activity on your pricing, documentation and comparison pages is the strongest and cheapest signal you will ever have, and most companies buy third-party intent before instrumenting it. Fixing that order costs less and produces better lift.
The public half of the signal, collected cleanly
Job postings, funding news, technology changes: the defensible signals are public, and they render differently by region. DataImpulse residential proxies give country, city and ZIP-level exits at $1 per GB across 195 countries. Create an account and test one segment.
Related: data enrichment · technographic data · is web scraping legal.
Last updated: September 17, 2026.
State/City/Zip/ASN Targeting 
