Intent data explained: First, second and third party, and what a surge score can and cannot tell you

Intent data is research behavior tied to a company, not a person, and it works as a prioritization signal rather than proof of a purchase. First party comes from your own site, product and email, second party from a partner's site, and third party from external providers. A Bombora surge compares an account's past 3 weeks on a topic with its prior 12 weeks. In 6sense, 70 or higher counts as surging.

What is intent data?

Intent data is research behavior tied to a company rather than to a person. Demandbase defines it as "behavioral information that indicates an account may be researching topics related to a business problem, product, or solution." The unit is the account. That is why it fits account-based marketing more naturally than lead-level programs.

Treat it as a ranking input. Demandbase calls intent data most useful as a prioritization signal, and says it cannot tell you with certainty that an individual buyer is ready to purchase. It answers one question: which accounts are researching your topics more than usual right now. It does not answer who will sign.

How a surge is measured

A surge score is a comparison against the account's own baseline. Bombora, as described in 6sense's documentation, looks for sustained intent on a given topic in the past 3 weeks, compared to typical levels of intent over the past 12 weeks. Inside 6sense, an account counts as surging when it has a Company Surge Score greater than or equal to 70, and the data is updated each Sunday on a weekly cadence.

It is our own example, not the provider's formula.

Account B has changed more. A surge model ranks B higher because the signal is the change, not the volume. Keep that in mind when a large enterprise with steady research activity never appears on your surge list.

The weekly refresh also matters for routing. If your sales team works a surge list on Monday, the signal behind it reflects activity from before Sunday's update. Build your service level agreement for follow-up around that cadence.

How practitioners use it

On outcomes, over 85% of companies using intent data have achieved business benefits. On spend, 30% of companies investing in intent data plan to increase their spend, while only 10% have plans to reduce their investment.

Each provider sees a different slice of research activity, so each produces a different baseline and a different surge list. For definitions used throughout this piece, see our intent data glossary entry.

First, second and third party intent data compared

The three labels describe who collected the data and who holds the relationship with the buyer. They say nothing about quality on their own. Clearbit defines first-party data as what "you collect from your audience directly, via your own channels." Everything else is someone else's collection that you rent or buy.

First party is your own web, product and email activity. Foundry describes it as information collected from your owned digital properties. It warns that if you rely on it alone, you only see buyers who land on your website.

Second party is someone else's first-party data that you access through a partnership. Foundry names software review sites like G2 or TrustRadius as the main second-party offerings for B2B marketers. Clearbit files the same G2 data under third party, as data a company collects on its own sites. Ask each vendor which label it uses and why.

Third party and co-op data is gathered by external companies tracking online activity across multiple platforms. Co-op data is a subtype: many organizations pool behavior, and the provider scores it by topic. Bombora describes its pool as publishers, brand websites and data providers contributing research behavior.

Foundry is specific about the trade-offs. Second-party data will not reveal top-of-funnel activity. Third-party data does not necessarily indicate buying intent.

A worked example

Take one target account across a month. The co-op flags it as researching a topic in your category. The account then compares your product with a competitor on a review site. Finally, someone from the account visits your pricing page twice.

Each type sees a different stage of the same research. In our reading, a stack that holds all three lets you time retargeting to the co-op signal and send sales in on the first-party one. You also need a lead scoring model that weights the types separately.

How B2B intent data is collected from a publisher co-op

Bombora runs one of the documented publisher co-ops, and its own descriptions lay out the method in three stages: gather content and visits from publishers, map that content to topics, then work out which company each visit came from. What follows is the company's method as published. The checks at the end are our reading.

Stage 1: Publisher relationships and visit volume

It said visit data was consented and written to first-party cookies.

On its current product page, Bombora states that 86% of the data in our Data Co‑op is shared exclusively with Bombora for deriving intent.

Stage 2: Content and metadata mapped to topics

The co-op gives access to the full text of articles and their metadata. Bombora uses that material to derive the Intent topics. Topic assignment is separate from identity: the company states that topics have no reliance on cookies or IP addresses.

The published scale figures vary by source and date:

Stage 3: Resolving the visit to an account

Identity comes from visitor data, specifically IP address and first-party cookie data. Bombora described three methods used together:

1. Consented IP addresses, tied to publisher consent strings.
2. A composite built with outside identity vendors, which it called a ‘graph-of-graphs’ approach, naming LiveRamp as one vendor.
3. In-house behavioral modeling of activity on IP addresses.

Bombora said that weaker identity would mostly affect topics with very little interest, or companies with very low activity in the Co-op. Influ2 makes a similar point about coverage: some industries or international segments simply don’t generate enough publisher data.

What to check before you load scores

These checks are our reading, not part of the published method:

• Topic fit. Pull the topic list your category maps to and confirm it matches the words your buyers use.
• Coverage on your list. Upload your target accounts and count how many show any activity at all. A thin match rate points to the coverage gap described above.
• Resolution path. Ask which of the three identity methods produced each account match, so you can see how the account was assigned.

For how these scores feed prioritization, see account-based marketing and lead scoring.

What an intent surge score measures

A Company Surge score compares an account with itself. In the definition 6sense publishes, Bombora "looks for sustained intent on a given topic in the past 3 weeks, compared to typical levels of intent over the past 12 weeks" (6sense). The reference point is the account's recent history on that topic. It is not a comparison with peers in the same industry.

The score range in one integration

The 6sense Kickstart Package provides scores "ranging from 70-100" (6sense). An account shows as surging when the most recent week's score is 70 or higher for that account and topic, and that threshold cannot be customized (6sense). The data refreshes each Sunday on a weekly cadence (6sense).

Other tools present the same signal differently. HubSpot shows a research level of High, Medium, Mild or Low on each event instead of a number (HubSpot).

Topic entitlements and selection

A score exists only for topics you have chosen.

Source: 6sense. Topic changes there can take up to 24 hours to appear across the platform. We suggest choosing the 12 default topics from the problems your product solves and the category terms buyers search for, not your brand name, and then leaving them in place long enough to read a trend.

1. It clears the 70 line, so the account appears in a surging segment for that topic.
2. Consumption on that topic over the past three weeks ran above the account's typical level for the prior 12 weeks.
3. It does not tell you who read the content, how many people did, or what stage they are in. Influ2 notes that a high Surge score "doesn't necessarily mean someone is ready to buy" (Influ2).

To turn that into points, see lead scoring and the lead scoring model template.

What intent data cannot tell you

A surge is an account-level observation. It tells you that research activity moved. It does not say who moved it, whether money is attached, when a decision lands, or who else is in the running. Demandbase's own guidance lists what intent should not be treated as proof of, including that a specific individual has decided to buy and that an account has selected your company or product.

Four gaps follow from that.

1. Which person is researching. Account-level co-op data does not name contacts. One reviewer summary of Bombora put it plainly: you won’t know anything about the contacts who may be in the market for your product.
2. Whether budget exists. Topic consumption carries no signal about funding. A team can read about a category for months with no line item behind it.
3. When they will buy. A rising score is not a date. Demandbase cautions against reading intent as proof that an account is ready for immediate sales outreach.
4. Which vendors are on the shortlist. Research on a topic does not show whether you, a competitor or nobody is being evaluated.

There is also a volume problem. Even after filtering, the lists can still be hundreds or even thousands of accounts long. In our reading, that is why each gap needs a second signal before a seller gets the account.

Gap-to-signal template

Use it as a gate. A surging account that clears none of the checks stays in nurture or advertising audiences. One that clears the person and fit rows moves to sales development research. Only budget and shortlist confirmation should move it toward a sales qualified lead.

This mirrors how Demandbase frames the job: combine intent with account fit, first-party engagement, CRM data, buying-group information, and other signals before deciding how to act.

Consent and privacy limits on intent data

The co-op model described earlier rests on consent captured by publishers. That consent is the part of intent data most likely to come up in a procurement or security review, so it pays to know what a provider can show you before legal asks.

What consent-based collection looks like

Bombora describes its co-op data as collected where consent strings can be traced and data can be deleted under regulations such as CCPA or GDPR. For identity, the same source says it uses IP address that is consented through the use of the consent strings from the publisher’s Co-op.

In practice that gives you two things to check. One is a record of consent tied to the signal. The other is a working path to remove data when a person asks.

How bidstream-derived data differs

Bidstream data comes from the information passed around during programmatic ad auctions. The person browsing did not agree to share it with an intent vendor. Bombora classes bidstream data as unconsented and says it is in questionable territory and does not meet privacy standards at most enterprise organizations. It also claims that Most Intent providers use third-party cookies and/or fingerprinting to determine identity. Treat both as one vendor's view of its competitors, and ask each provider directly.

In our reading, bidstream is not a fringe source. Bombora lists it among the narrow Intent data sources such as bidstream data, review sites, or publisher-specific consumption some providers rely on.

Privacy questions to put to any provider

Bombora's own evaluation guidance lists privacy and compliance considerations alongside data quality and scale. The table turns that into questions. The right column is our reading of what a usable answer contains.

Send these in writing and keep the replies with your contract file. Your own counsel decides what the answers mean for your obligations. We make no legal judgment here.

One operational step: map where intent data lands in your stack, such as your customer relationship management system and lead scoring models. A deletion request has to reach every copy.

How accurate is intent data, and how to check a sample

Accuracy is not one number a provider can hand you. Forrester lays out an evaluation order. First, teams should start by understanding the scale of relevant signals within their core-fit accounts and offerings. Next, they determine the relative level of accuracy. Last, they look at overlap with sources they already have, to predict the incremental business impact of each new one.

Set up the sample

Intent providers rarely send a static file. The same Forrester post says buyers more often get a limited data feed or short proof-of-concept trial, so the pace of new signals is visible. Keep the scope tight: a narrow keyword focus, clear geographic or industry definitions, and a tight timeframe (weeks, not months). Generic category keywords may obscure differentiation between providers.

Give every provider the same account list and the same topic map. Then pick the segment your team knows best. Forrester's advice is to focus on areas you know best, because that is where you can check a signal against what sales already knows.

Run the three checks

A worked example with our own numbers. You send 400 core-fit accounts and a narrow topic map for a four-week trial. The provider flags 60. Your website and form data already showed 18 of the 60, so 42 are incremental. Run the same math on each provider and compare the incremental count, not the raw total. Then size the follow-up with the pipeline coverage calculator.

Expect more than one provider

That makes the overlap check the one that decides budget. Ask each provider where its signals come from. Bombora, for example, states that 86% of the data in our Co-op is shared exclusively with Bombora for deriving intent. In our reading, a claim like that is a reason to run the overlap test, not a substitute for it. Your own incremental count is the number to bring to finance.

How to test whether intent data works with an account holdout

A vendor dashboard shows which accounts surged and which of them later became opportunities. It cannot show what those accounts would have done if you had left them alone. Bombora says its Company Surge product lets marketing and sales teams pinpoint the accounts most likely to be in-market. An account holdout tests whether acting on that signal adds pipeline, which is the number your renewal depends on.

In the example, contacted surging accounts converted to opportunities at 7.0%, held-out surging accounts at 4.0% and non-surging accounts at 2.5% within 90 days.

Worked example: opportunity rate by account group in an intent holdout test

Item	Value
Surging, contacted	7 % of accounts
Surging, held out	4 % of accounts
Not surging	2.5 % of accounts

Naive comparison (contacted surging versus not surging): 2.8x. True lift of acting on the signal (contacted versus held out): 1.75x. Source: Take the Market worked example (illustrative inputs)

The method is a standard incrementality test applied to accounts, the same logic we use in incrementality testing for paid media. Every surging account gets randomly assigned to one of two groups. The treated group gets sales outreach and ads. The holdout gets nothing extra from you for the test window.

The common mistake is comparing contacted surging accounts with non-surging accounts. That gives 2.8x, and it mixes two effects: surging accounts were already more active, and you also worked them. The holdout isolates the second effect. True lift is 7.0% divided by 4.0%, or 1.75x.

The calculation

from math import sqrt, erf

def two_prop_test(x1, n1, x2, n2, z_crit=1.96):
    p1, p2 = x1 / n1, x2 / n2
    pooled = (x1 + x2) / (n1 + n2)
    se_pooled = sqrt(pooled * (1 - pooled) * (1/n1 + 1/n2))
    z = (p1 - p2) / se_pooled
    p_value = 2 * (1 - 0.5 * (1 + erf(abs(z) / sqrt(2))))
    se_diff = sqrt(p1*(1-p1)/n1 + p2*(1-p2)/n2)
    ci = (p1 - p2 - z_crit*se_diff, p1 - p2 + z_crit*se_diff)
    return p1 / p2, z, p_value, ci

print(two_prop_test(14, 200, 8, 200))

The difference is 3 points, but the 95% confidence interval runs from about minus 1.5 points to plus 7.5 points. At 200 accounts per group, this result could still be noise.

Steps to run it

1. Freeze the surge list on day one and assign accounts by hashing the account ID, so nobody hand-picks the treated group.
2. Suppress the holdout in sales sequences and ad audiences, and log any accidental touches.
3. Count opportunities created in 90 days for both groups, using the same sales qualified lead definition.
4. Run the code above and record lift, p-value and the interval.

Go or no-go rule for renewal

• Go: the lower bound of the interval sits above zero, and the extra opportunities, valued at your win rate and deal size, exceed the contract cost.
• Extend: the interval crosses zero, as in this example. Rerun with a larger split before deciding.
• No-go: the upper bound sits near zero after an adequately sized test.

What this page does not cover

This page does not compare, rank or recommend intent data providers, and it does not cover pricing, contract terms or contact-level identification. The privacy section lists questions to put to a provider and is not legal advice: your own counsel decides what the answers mean for your obligations. Nothing here is financial advice.

Related on Take the Market: ABM playbook for one-to-few programs: the account list, the plays, the metrics and a 90-day plan; Demand creation vs demand capture: definitions, the 95-5 rule, budget split and measures; Email deliverability requirements in 2026: the Google, Yahoo and Microsoft sender rules, warm-up and the numbers to wa.