How Do You Test a B2B Data Vendor's Accuracy Before Buying?

TL;DR: To test a B2B data vendor, pull a random sample of 200-400 contacts from the exact segment you plan to buy, then check each one against a live source: call the phone, send to the email through a verifier, confirm the title on LinkedIn or the company site. Judge the vendor on the share of records that are right-person and reachable, not on the "accuracy score" in their pitch deck. A 100-record sample can only tell you accuracy within roughly ±10 points, so sample size matters as much as the method.

How big a sample do you need to test a data vendor's accuracy?

A random sample of about 400 contacts gives you a margin of error of roughly ±5 percentage points at 95% confidence, and 100 contacts gives roughly ±10 points. That is standard binomial math, and it is the reason a "we tested 50 records and 90% were good" claim tells you very little. If a vendor says 90% accuracy and your 100-record sample finds 82%, the difference could easily be noise. At 400 records, the same gap is meaningful.

If budget is tight, 200 records (about ±7 points) is a workable floor. Below that, treat results as anecdotes.

How do you pull a sample that isn't cherry-picked?

Insist on a random pull from your real target segment, not a vendor-curated "sample file." Define your ICP filters first (titles, industries, company size, geography), have the vendor export a larger set than you need, then randomly select your 200-400 yourself. Vendors who hand-pick samples will show their best records. A sample that skews to large enterprises in the US will also overstate accuracy if you actually sell to SMBs in multiple regions.

Also ask for the record's "last verified" date, if the vendor has one. Older stamps usually correlate with lower accuracy, and a missing stamp is itself information.

What should you check on each record?

Check each field separately, because accuracy varies a lot by field. Typical benchmarks that practitioners report:

• Email: fails deliverability checks anywhere from low single digits to 20%+ on aged data. Run it through a verifier and track hard bounces.
• Phone: often the weakest field. Published decay estimates put phone numbers at roughly 15-25% a year, mobiles somewhat lower, so a vendor database with stale records can easily have 20-30% of numbers that are disconnected, wrong, or not tied to that person.
• Job title and employer: published estimates of annual change run roughly 25-35%. Confirm against the company site or LinkedIn.

These are ranges from vendor and industry sources, not guarantees. Your own sample is the only number that applies to your segment.

What's the difference between "valid" and "right person" in a data test?

A valid number is one that rings; a right-person number is one that reaches the named contact. A number can pass a format check and a carrier lookup and still belong to someone who left two years ago or to a front desk. Many vendors report validity (does it exist?) and call it accuracy. The metric that predicts pipeline is right-party contact: of the numbers that connect, how many reach the intended person?

The only way to measure that is a live check, meaning a human or an automated call that asks for the person. For a 300-record sample, a rep can do this in a few hours, which is cheap compared with a five-figure annual contract. Phone-verification services, including Ringfire, can run the same check at list scale if you want to test the whole file rather than a sample.

How do you score the results?

Report three numbers per field: percent verified correct, percent definitively wrong, and percent unverifiable. Don't quietly drop the unverifiable records; counting them as correct inflates the vendor's score and counting them all as wrong penalizes it. Show both bounds.

A simple scorecard:

1. Phone right-party rate: right person reached / records attempted.
2. Email hard-bounce rate: hard bounces / emails verified.
3. Title/employer match: records still current / records checked.
4. Coverage: share of sampled contacts that had each field at all.

Then compare vendors on the same segment with the same sample method. Run each vendor on the same company list when possible so differences come from the data, not the targets.

What accuracy should make you walk away?

There is no universal threshold, but if fewer than roughly 60-70% of sampled phone numbers reach the right person, or email hard bounces exceed 5% after verification, the data will cost you in wasted dials and sender-reputation damage. Email bounce rates above about 2% are commonly cited as the point where inbox providers start to penalize senders. Weigh price against the cost of bad records: a cheaper list with 30% dead numbers costs more per reached contact than a pricier list with 10%.

Re-run the test annually, or after any big change in your target market. Vendor quality drifts, and so does your segment.