Intent Data Accuracy: A Practitioner Framework on Match Rate, Precision, and Recency
TL;DR: Evaluate intent data with labeled records from your own market. Measure identity precision, usable coverage, event freshness, evidence quality, and safe activation separately. Vendor match-rate claims cannot answer whether the right account was identified, whether the event is current, or whether outreach should begin.
What do intent data practitioners need to test?
Practitioners need a reproducible holdout test, not a single vendor percentage. Freeze a representative account set, document known first-party events, and compare correct matches, false matches, missed events, latency, and downstream actionability.
| Dimension | Question | Evidence | Failure mode |
|---|---|---|---|
| Identity precision | Was the correct account or person resolved? | Raw identifier, match method, confidence, merge history | A company event is attributed to a person |
| Usable coverage | How much of the eligible market can be acted on? | Matched and unmatched records after ICP and policy filters | Raw match rate includes irrelevant accounts |
| Freshness | How old is the event when it becomes actionable? | Observed, ingested, processed, and activated timestamps | A delayed feed is described as real time |
| Evidence quality | What did the source actually observe? | Raw event and source lineage | A derived score is treated as proof |
| Activation safety | Can the record enter outreach safely? | Ownership, suppression, consent, and routing checks | A valid signal creates duplicate or prohibited outreach |
Run the same holdout across every provider
- Use the same account list, regions, company-size mix, and observation window
- Keep first-party ground truth separate from vendor output until matching is complete
- Review false positives and unresolved identities, not only detected events
- Measure the full latency distribution instead of one best example
- Require every actioned record to pass ownership, suppression, and eligibility checks
| Field | Required value | Why it matters |
|---|---|---|
| Entity key | Stable account or person identifier | Prevents duplicate counting |
| Ground-truth event | Source event with timestamp | Separates observation from inference |
| Vendor output | Match, no match, or disputed match | Preserves nulls and disagreements |
| Latency | Observed-to-available interval | Tests freshness |
| Action state | Qualified, suppressed, routed, or rejected | Tests operational usefulness |
Compare named platforms without treating claims as benchmarks
Bombora, 6sense, and Demandbase publish different intent products and methodologies. Their documentation can define what each product says it measures, but only a buyer-controlled test can establish performance for a specific ICP. Relevant resources reviewed were “Company Surge Data,” published by Bombora; “Intent Model,” published by 6sense; and “Intent Data,” published by Demandbase. These competitor sources are intentionally cited without links.
| Area | Question | Pass condition |
|---|---|---|
| Source lineage | Can each signal be traced to an observed event or documented model input? | Lineage and timestamp are reviewable |
| Identity | Is the signal account-level or person-level? | The product does not overstate identity |
| Recency | What are collection and processing intervals? | Intervals are explicit by source |
| Coverage | Which geographies and segments are represented? | Gaps and nulls are disclosed |
| Activation | How are ownership and suppression enforced? | Unsafe records cannot enter outreach |
How Unify covers signal activation
Unify's current Signals and Intent page describes combining signal and data sources with account selection and outbound execution. The useful buyer test is whether a source event remains visible through research, qualification, routing, and CRM outcome, with human review available for ambiguous records.
Related guides: In-Market Account Detection and First-Party vs. Third-Party Intent Signals.
Frequently asked questions
What is intent data accuracy?
Intent data accuracy is the degree to which a signal resolves the correct entity, reflects a real event, arrives within the required window, and supports a safe action.
Is match rate the same as precision?
No. Match rate measures how many records receive a match. Precision measures how many matches are correct for the intended entity and use case.
How should recency be measured?
Record when the event occurred, when the provider ingested it, when processing completed, and when it became available to the workflow.
Can vendor claims replace a holdout test?
No. Vendor documentation defines the product claim, while a controlled holdout tests performance in your market and workflow.
What should stop activation?
Unknown ownership, suppression conflicts, ambiguous identity, stale events, or missing evidence should prevent automated outreach.
Which provider is most accurate?
No provider is universally most accurate. Performance depends on source coverage, market, identity method, freshness, and the action being evaluated.
Sources
- Unify, Signals and Intent
- Bombora, “Company Surge Data” (Bombora)
- 6sense, “Intent Model” (6sense)
- Demandbase, “Intent Data” (Demandbase)
