Why B2B Intent Data Is Wrong More Often Than You Think

Intent data is the most persuasive thing a sales data vendor can put in a demo. A list of companies, ranked, with a label saying they are researching what you sell right now. It reframes outbound from cold guessing into something that feels like knowledge.

It is also, in its most widely sold form, considerably less accurate than buyers assume, and the reason is structural rather than a matter of one vendor doing it badly.

A marketing leader we spoke with recently described intent signals as the reason her campaigns at a previous company had been “way more intelligent.” She is not wrong that the idea is powerful. But when we walked through how the signal is actually produced, the conversation changed. This article is that walkthrough: how intent data is built, the specific reason it misfires, and what to use instead.

How third-party intent data is actually produced

Start with the mechanism, because almost every limitation follows directly from it.

Third-party intent providers operate a co-operative of publisher websites: trade publications, review sites, technical blogs, whitepaper libraries. Across that network, they observe content consumption. When someone reads an article about warehouse automation, that visit is recorded.

The visitor is anonymous. What the provider can see is their IP address. So the next step is IP-to-company resolution: matching that IP against a database of corporate IP ranges to conclude “this visit came from Northwind Logistics.”

Aggregate that across a network, compare against a baseline of normal consumption for each company, and when a company’s reading on a topic spikes above its own baseline, it is flagged as showing intent. That is a “surge.”

The logic is reasonable. A company reading unusually heavily about a category is more likely than average to be in market for it. As a statistical statement across thousands of accounts, that holds up.

The problem is what happens when you use it to decide whether to email one specific person.

The core flaw: the signal is company-level, the action is person-level

Here is the failure in a sentence: intent is measured at the company, but you act on it at the individual.

An IP address identifies an organisation, not a human. When a signal fires for a 2,000-person company, you know somebody there read something. You do not know who, what their role was, or whether they have any connection to buying decisions.

The example we use, because it is so ordinary, is this: a developer at a company googles “how to connect to Salesforce” because they are debugging an integration for an existing implementation. They read three articles. The IP resolves to their employer. The company’s consumption on the CRM topic spikes. The account surfaces as showing intent for CRM software.

Sales sees a hot account. In reality, one engineer was troubleshooting software the company already owns. There is no buying process, no budget, no evaluation. The signal was technically accurate (that content genuinely was consumed from that company) and completely misleading about what it implied.

This is not an edge case. It is the standard case, for reasons worth spelling out:

• Existing customers research the products they already use. Support docs, feature guides, integration questions. Their consumption looks identical to a prospect’s.
• Job seekers research employers and their tools. Someone preparing for an interview reads heavily about a category. Same footprint.
• Competitors research you and the category constantly. Often more intensively than any real buyer.
• Students, analysts and journalists consume category content for reasons unconnected to purchasing.
• Anyone who is simply curious. Professional reading is not buying behaviour.

None of these are distinguishable from genuine buying research when all you observe is an IP address and a topic.

Why the noise got worse, not better

The IP-to-company link was always the weak joint, and several trends have degraded it further.

Distributed work. A large share of knowledge workers now work from home at least part of the week. Home IPs are residential and resolve to an ISP, not an employer. The employee doing the most serious evaluation research, often at night and on their own machine, is frequently invisible. Meanwhile whoever is in the office generates the signal.

VPNs and zero-trust networking. Corporate traffic increasingly egresses through cloud security providers, so many employees’ traffic resolves to a security vendor’s IP range rather than the employer’s.

Shared and cloud infrastructure. Traffic from cloud environments, co-working spaces and shared offices resolves ambiguously or to the wrong organisation entirely.

Privacy infrastructure. Apple’s iCloud Private Relay, browser tracking protection and similar measures obscure or randomise the signals these networks depend on.

The consequence is a signal that is simultaneously incomplete, missing your most engaged researchers, and polluted by traffic attributed to the wrong organisation. Both errors point the same way: toward confident-looking output you cannot audit.

The unfalsifiability problem

The deepest issue with third-party intent is not the error rate. It is that you cannot check it.

When a provider tells you an account is surging, you receive a company name and a topic. You do not get the person, the page, the timestamp, or the reasoning. So when you email the account and hear nothing back, there is no way to distinguish between these possibilities:

• The signal was wrong.
• The signal was right but you reached the wrong person.
• The signal was right, you reached the right person, and your message was weak.
• The signal was right but late, and the deal was already committed.

Every one produces silence. Without provenance you cannot attribute the failure, which means you cannot improve. And an input you cannot evaluate tends to survive on faith rather than evidence, because the wins are memorable and the misses are invisible.

This is the same provenance question that determines whether AI research output is trustworthy, and it is not a coincidence but the same underlying discipline. We covered that in why AI prospect research gets it wrong.

What actually predicts a buying window

The useful alternative is not “no signals.” It is signals that are observable, attributable and specific, where you can see the underlying evidence and judge it yourself.

Most of these are public. That is the point: their value is not exclusive access, it is that almost nobody reads them carefully.

1. Hiring signals, read properly

Job postings are the most underused source in B2B, because most teams read them as recruitment ads rather than as disclosures about a company’s internal state.

A job description tells you what a company is building, what it runs today, what it is struggling with, and who now owns the problem. Specifically:

• The tools named in the requirements are the current stack. If the JD asks for experience with a competitor’s product, they run it. That is a displacement conversation with a known incumbent.
• A “first” hire signals a function being built from zero. First RevOps hire, first security engineer, first data engineer. Tooling decisions are open and no vendor has incumbency.
• Volume in one function signals investment. Three data engineering roles at once is a platform project with allocated budget.
• The responsibilities section describes the pain. “Improve reporting reliability” or “reduce manual reconciliation” is the problem statement, written by the person who feels it.
• Seniority tells you the timeline. A director-level hire precedes purchasing. Junior hires suggest a decision already made.

The critical difference from third-party intent: you can read the posting yourself. It is dated, attributable and specific. You can quote it.

2. Leadership changes in the function you sell to

The most reliable timing signal in B2B, and it is fully public. New leaders re-evaluate inherited tooling, typically within their first ninety days, and they arrive with both a mandate to change things and preferences from previous roles.

A new VP of Sales is a sales tooling opportunity. A new Head of Security reviews the security stack. The window is real, dated and observable on LinkedIn.

3. Technology changes

Adopting or dropping a platform triggers adjacent needs. A company migrating to a new CRM needs everything that integrates with it. Detectable through job postings, engineering blogs, conference talks and public tech-stack data.

4. Funding and structural events

Funding creates budget and a mandate to grow, usually with a stated purpose. A round earmarked for international expansion tells you what they will buy in the next two quarters. Acquisitions force consolidation decisions across two overlapping stacks.

5. First-party signals, which outrank all of the above

Someone visiting your pricing page is a stronger signal than any third-party surge, because it is your own data, attributable to a session, and directly about your product rather than a category.

The hierarchy is worth stating plainly:

1. First-party behaviour. Your site, your content, your product. Highest confidence.
2. Public company signals. Hiring, leadership, funding, technology. Verifiable, dated, attributable.
3. Third-party anonymous intent. Directional at the account level, weak at the individual level.

Most teams invert this, buying tier three while under-instrumenting tier one. Website visitor tracking that can trigger a sequence is usually a better investment than an intent subscription, and it is a capability you may already have. Ours is part of the sales engagement platform.

Where third-party intent still earns its place

This is not an argument that intent data is worthless. It is an argument about the level at which it is valid.

Used at the account level, for prioritisation, in aggregate, it is genuinely useful. If you have 5,000 target accounts and need an order to work them in, intent is a reasonable input. Statistical noise matters less when you are ranking a large list, and being directionally right about which 500 to work first has real value.

Used at the individual level, as a trigger, as fact, it misleads. “Sarah is researching this, email her today” is a claim the underlying data cannot support, because the data never knew about Sarah.

A defensible way to use it:

• Treat it as a prioritisation input, not a trigger.
• Require a second, verifiable signal before acting. Intent surge plus a relevant job posting is meaningfully stronger than either alone.
• Never reference it in your message. “I saw you were researching X” is both slightly unsettling and frequently wrong.
• Measure it honestly: hold out a control group and compare conversion on intent-flagged accounts versus matched non-flagged ones. Very few teams run this test. The ones that do are often surprised.

That last point deserves emphasis. If intent data works for you, a holdout test will prove it and justify the spend. If it does not, you have been paying for a feeling. Either outcome is worth knowing, and the test costs nothing but discipline.

The scale problem nobody has fully solved

It is worth being honest about why the noisy version of intent persists: the accurate version is genuinely hard.

Deep signal analysis, reading job descriptions properly, tracking leadership changes, correlating technology adoption, works extremely well one account at a time. Doing it continuously across millions of companies is expensive, and signals decay. A hiring signal is valuable this month and stale in six.

So there is a real tension. Anonymous IP-based intent scales cheaply and is noisy. Deep signal analysis is accurate and expensive. Most vendors chose scale, then marketed it as precision.

The direction that actually resolves this is narrowing the surface: rather than monitoring every company continuously, monitor your target accounts continuously and deeply. A focused list of a few hundred accounts, watched for hiring, leadership and technology changes, produces far better signal than a broad anonymous feed, and it is computationally tractable.

That is also how the best teams already worked before intent data existed. They kept a target account list and watched it closely. The technology should make that easier, not replace it with something less accurate and harder to check.

A practical framework

If you are rebuilding how your team uses signals, this is a workable structure.

1. Instrument first-party thoroughly. Visitor identification, content engagement, product usage if you have it. Highest signal, lowest cost, fully attributable.
2. Define a target account list. A few hundred accounts you genuinely want, not an entire TAM.
3. Monitor those accounts on public signals. Hiring, leadership changes, funding, technology. Weekly is sufficient.
4. Use third-party intent for prioritisation only, if at all. A tiebreaker for working order, never a trigger.
5. Require two signals before outreach. One signal is a coincidence. Two is a pattern.
6. Reference the verifiable signal, never the anonymous one. “I saw you are hiring three data engineers” is specific and checkable. “I saw you were researching data platforms” is neither.
7. Run a holdout test after one quarter. Compare intent-flagged accounts against matched controls. Let the result decide the renewal.

The seventh step is the one that changes behaviour, because it converts an article of faith into a measurement.

The bottom line

Third-party intent data is a genuine statistical signal wrapped in a promise it cannot keep. The underlying observation, this company consumed unusual amounts of content on this topic is real. The inference sold on top of it, this specific person is in market right now is not supported, because the data never identified a person, cannot distinguish a buyer from a developer or a job seeker, and increasingly cannot even reliably identify the company.

The alternative is not to abandon signals. It is to prefer ones you can see, date and verify: hiring patterns, leadership changes, technology moves, and above all your own first-party data. They are less impressive in a demo and considerably more useful in a quarter.

The question to ask any vendor, and it is the same question that separates good AI research from confident guessing: show me the evidence behind this signal. If the answer is a score with no underlying record, you are buying a feeling.

If you want signals attached to records you can actually inspect, hiring activity, role changes and company data on your target accounts, plus first-party visitor tracking that can trigger a sequence, that is how Salesgear approaches prospecting. For how this connects to research quality, see why AI prospect research gets it wrong, and for the vendor landscape, ZoomInfo competitors and best B2B data enrichment tools.

Frequently asked questions

How accurate is B2B intent data?

At the account level and in aggregate it is directionally useful for prioritising a large target list. At the individual level it is unreliable, because the signal is derived from IP addresses that identify an organisation rather than a person. A company flagged as showing intent may have generated that signal from an existing customer reading support docs, a job seeker, or a competitor.

Why does intent data flag companies that are not in market?

Because it cannot distinguish who inside the company consumed the content or why. Existing customers researching features, engineers troubleshooting integrations, job candidates preparing for interviews and competitors monitoring the category all produce the same footprint as a genuine buyer. The classic case is a developer researching an integration for software the company already owns, which surfaces the account as in-market for that software.

Does remote work affect intent data quality?

Significantly. IP-to-company matching assumes people work on corporate networks. Home connections resolve to residential ISPs, VPN and zero-trust setups route traffic through security vendors’ ranges, and privacy tools obscure the signal further. The result is both missed research and traffic attributed to the wrong organisation.

What are better alternatives to third-party intent data?

First-party behaviour is strongest: your site, content and product usage, attributable to a real session. Next are public company signals you can verify yourself, job postings that reveal the current stack and active projects, leadership changes in the function you sell to, funding events and technology migrations. All are dated, specific and quotable in outreach.

Should I cancel my intent data subscription?

Test before deciding. Run a holdout: compare conversion on intent-flagged accounts against matched accounts without the flag, over a quarter. Very few teams run this test, and it settles the question with evidence rather than intuition. If you keep it, use it for prioritisation only, require a second verifiable signal before outreach, and never reference the intent signal in your message.