Quick answer
Intent data is behavioral evidence used to infer that people or accounts may be researching a topic or purchase. First-party intent comes from a company's own consented interactions; second-party intent comes directly from another publisher or platform; third-party intent aggregates signals across external sources. Account scoring combines intent with ICP fit, relationship, timing and other evidence to prioritize action. A sound model documents source, identity confidence, topic, baseline, surge, recency, decay, outcome and uncertainty; validates scores prospectively; and routes accounts to proportionate experiences. Intent does not prove a funded project, identify every committee member or grant permission for intrusive outreach.
What is B2B intent data?
Intent data is behavior interpreted as evidence of interest in a topic, problem or purchase. Examples include product or pricing activity, content consumption, event participation, publisher research and aggregated topic activity. The observation is real; the intended purchase is an inference.
Account scoring ranks organizations for a decision such as advertising, research, nurture or sales review. It can combine stable account fit with changing intent and relationship evidence. One score should not silently control every workflow with different costs and risks.
First-, second- and third-party intent
First-party intent comes from owned interactions such as product use, website events, CRM conversations and event attendance under the relevant context. Second-party data is another organization's direct audience evidence shared by agreement. Third-party providers aggregate behavior across external sites or cooperatives.
Closer provenance can improve interpretability but does not guarantee validity. A pricing visit may be a competitor or student; an IP address may represent many employees; a content download may serve learning rather than purchase. Every source needs a documented meaning and confidence.
Design the signal and scoring model
Separate ICP fit, intent, relationship, timing and risk. Within intent, consider topic relevance, recency, frequency, depth, breadth across participants and change from an account baseline. Use decay because old research should not create permanent in-market status.
Define identity confidence and missingness. Account-level aggregation can protect some individual detail but shared networks create false matches. Keep component values visible so users can understand why a score changed and challenge faulty evidence.
Collect
Define lawful signal sources, purposes, topics and retention.
- What behavior occurred?
- What did the person reasonably expect?
Resolve
Match signals to people and accounts with explicit confidence.
- How was identity inferred?
- Could shared networks or devices mislead?
Score
Combine fit, intent, recency, relationship, timing and risk transparently.
- What does each component mean?
- How fast should it decay?
Route
Choose a proportionate next experience with human review where needed.
- What value follows from the signal?
- Is direct contact appropriate?
Validate
Test calibration, incrementality, bias and drift on later outcomes.
- Do higher bands perform better?
- Does acting on the score improve outcomes?
Resolve people, accounts and buying groups
Authenticated first-party events can link reliably when accounts and users are governed. External matching may use domain, IP, cookies, device graphs or publisher identity, each with different error and privacy characteristics. Remote work, agencies and shared infrastructure complicate attribution.
Use confidence thresholds and do not invent named buyers from anonymous account activity. Distinguish one prolific researcher from broad committee activity. Exclude staff, bots and service providers where possible, and audit samples against confirmed account evidence.
How to implement account scoring
Start with the decision and cost of error. Define eligible accounts, outcomes and observation windows. Inventory sources, contracts, consent, provenance and quality. Create an interpretable baseline model before buying more feeds or introducing machine learning.
Backtest only for exploration, then freeze thresholds and validate on future cohorts. Design action bands and service levels, train users, log overrides and monitor drift. Run controlled tests to estimate whether score-based action adds value beyond the score's prediction.
- Use case and error cost defined
- Fit separate from intent
- Source provenance documented
- Topic taxonomy relevant
- Identity confidence visible
- Recency and decay set
- Missing data handled
- Action bands proportionate
- Human review assigned
- Prospective validation used
- Incremental action tested
- Privacy and vendor rights audited
Intent data and scoring example
CedarStack's hypothetical model prevents a general research surge from becoming an automatic sales alert. Topic relevance, account fit, signal breadth and freshness shape interpretation, while the route reflects evidence strength and customer expectation.
Prospective validation distinguishes prediction from intervention. High-scoring accounts may buy because they were already active. Only an experiment or suitable causal design can estimate whether a particular outreach or content treatment improved the outcome.
CedarStack is a hypothetical cloud-governance platform. It receives website behavior, conference participation and an external topic-surge feed, but sales currently treats every surge as evidence that an account is buying.
The team documents each event, topic, collection context, account-match method, latency and expiry. A general cloud-security topic receives less relevance than research tied to the supported governance use case.
ICP fit remains a stable dimension. Intent includes recency, frequency and breadth across a buying group; relationship covers known customer or opportunity evidence. Missing data is not scored as negative certainty.
A moderate score may lead to useful ungated content or account research. Direct outreach requires sufficient fit, fresh relevant evidence, contact permission or a legitimate sales context, and seller review.
Score bands are frozen and followed into verified opportunities, adoption and disqualification. CedarStack tests whether using the model improves outcomes versus existing routing, not merely whether scores correlate with sales activity.
People can opt out where required, access is restricted and the team avoids messages that reveal third-party browsing inference. Vendors are audited for provenance, rights, accuracy and deletion.
CedarStack and all signals are hypothetical. Intent-data use must follow applicable privacy, communications and contractual requirements.
Route signals to useful action
Low or ambiguous signals can improve aggregate content and media planning without direct identification. Stronger account evidence can prompt seller research, tailored education or a customer-success check. Direct contact needs a legitimate context and a message that offers relevant value.
Coordinate routing across marketing, sales and success so an account does not receive duplicated messages. Set caps, suppression, owner and expiry. When a signal cannot support a respectful action, recording uncertainty is better than forcing activity.
Validate accuracy, calibration and lift
Check whether higher score bands show monotonically stronger verified outcomes, with confidence intervals and mature windows. Measure precision, recall, calibration, coverage and segment performance according to the decision. Investigate false positives and false negatives qualitatively.
Then test the action policy. Randomize eligible accounts to score-informed treatment or business-as-usual where feasible. Monitor opportunity quality, cycle time, adoption, complaints and sales effort. A predictive model that triggers harmful action has failed operationally.
Govern privacy, vendors and inference
Document legal basis, purpose, notice, sharing, access, retention, deletion and vendor obligations. Audit how providers obtained data, whether people can exercise rights and whether sensitive topics are excluded. Contract language does not replace customer expectation or ethical review.
Avoid revealing inferred browsing or labeling individuals secretly as buyers. Provide internal explanations and correction routes. Monitor proxies and subgroup errors, and require approval before adding topics related to health, employment vulnerability or other sensitive conditions.
Limitations and common intent-data mistakes
Intent sources cover only part of the web and often overrepresent research on participating properties. Account matching can be wrong, competitors research categories and a buying group may work offline. Vendor scores can hide definitions and changing data supply.
Common mistakes include treating surge as purchase, ignoring fit, using stale scores, contacting every researcher and validating on sales follow-up created by the score. Use intent as time-sensitive probabilistic evidence within a wider account strategy, not an oracle.
Intent data can improve timing when its provenance, identity and uncertainty are visible. It cannot turn inference into permission or certainty.
Frequently asked questions
What is B2B intent data?
Behavioral evidence interpreted as possible interest in a business topic, problem or purchase, collected from first-, second- or third-party sources.
What is an intent surge?
A measured increase in topic activity relative to an account or provider baseline. Its meaning depends on coverage, identity, topic relevance and freshness.
Is website activity enough to contact an account?
Not automatically. Consider fit, identity confidence, context, communication law, preference and whether direct contact offers proportionate value.
How is account scoring different from lead scoring?
Account scoring aggregates fit and behavior at the organization or buying-group level. Lead scoring ranks individual records, though both need governed identity and outcomes.
How do you validate an intent model?
Freeze it and test future score bands against verified outcomes, then separately test whether score-based actions incrementally improve customer and business results.
Sources and further reading
- IAB: B2B Account-Based Marketing Playbook ↗Industry framework for account audiences, data and ABM execution
- IAB: Audience Taxonomy ↗Industry guidance on inconsistent interest and purchase-intent audience labels
- TechTarget: Account-Based Marketing Guide ↗Current practitioner overview of intent data, account mapping and scoring
- EDPB: Guidelines on Automated Decision-Making and Profiling ↗Regulatory guidance for profiling transparency, rights and safeguards