A useful quality report should tell you what was checked, how it was checked and what remains unknown. A single score hides too much when coverage, identity and contact fields measure different things.
At Leadpipe, we recommend defining the evidence before comparing providers. This guide describes a review process. It does not report a new vendor benchmark. The accuracy-method article explains the correction to earlier score claims.
Define the measures before sampling
| Measure | Calculation | Evidence required |
|---|---|---|
| Coverage | Unique returned entities divided by eligible inputs | A consistent population and identity unit |
| Identity correctness | Correct identities divided by reviewed matches with ground truth | Independent evidence of who visited |
| Field completeness | Records with every required field divided by records reviewed | A written required-field list |
| Attribute currency | Reviewed attributes confirmed current divided by attributes reviewed | Appropriate current sources and timestamps |
| Sales acceptance | Records meeting fit and usability rules divided by records reviewed | Documented acceptance criteria |
| Delivery | Accepted records reaching the correct owner divided by records intended for delivery | Destination and ownership logs |
State the unit in every metric. Companies, suggested employees and matched individual visitors belong in separate counts. Deduplicate repeat activity before calculating person-level coverage.
Establish ground truth where possible
A controlled evaluation can use participating visitors with independently logged identities and visits, under appropriate permission. Preserve the visit timestamp, relevant browser conditions and geography. Blind the reviewer to the provider where practical.
A real professional profile can help confirm a person’s employer or job title. It cannot by itself establish that the person visited the website. A provider’s own label and agreement between two providers are also insufficient as independent ground truth.
Classify each reviewed identity as correct, incorrect or unknown based on the available evidence. Report the unknown count alongside any calculated correctness rate. Do not silently exclude difficult cases and present the remaining percentage as representative of all output.
Sample for the decision you need to make
Use a representative selection from the eligible population. If paid traffic, geography or visitor type matters, preserve those segments in the review. Record the sample-selection method and which groups it does not represent.
Choose the sample size around the uncertainty the business can tolerate. A handful of successful matches can demonstrate a workflow but cannot establish a narrow error-rate estimate. Report counts with rates so the reader can assess the evidence.
Inspect fields separately from identity
After reviewing identity, check current employer, required email or phone fields, duplication and suppression. Record a reason for each failure. A correctly identified person may still lack a field required by the sales process; a complete record may still have an unresolved visitor identity.
Complete the handoff
Send accepted records to the intended destination with source labels. Verify company association, existing owner, excluded customers and repeat updates. Count manual correction and administration as operating effort.
Use the evaluation worksheet to preserve the review evidence. The parallel-comparison guide explains why observational vendor cohorts differ from randomized outreach groups.
Decide using the full workflow
Compare accepted output, reliable delivery and operating cost before following the cohort into conversations and opportunities. Keep pipeline separate from revenue and profit. Use a defined observation window and avoid assigning a universal vendor rank from a narrow traffic sample.
For a person-level website workflow, review Leadpipe identification and current pricing. Evaluate the product on the same requirements as every other candidate. The trial is 7 days or 500 profiles, whichever comes first, no card. Start a trial.
