Blog / Healthcare Lead Generation / B2B data

How to Verify a B2B Data Provider's Accuracy Claims

Every provider claims high accuracy. Here is how to test the claim yourself, in about an hour, before any money changes hands.

Harshith Kumar

Healthcare data writer

Updated 6 Nov 2026

10 min read

TABLE OF CONTENTS

Reading progress

QUICK ANSWER

A B2B accuracy claim is testable in about an hour. Request a random sample from your exact segment, validate the emails, manually check 20 records against public sources, count fill rates on the fields you filter by, deduplicate on a stable identifier, and check verification dates. A headline accuracy percentage almost always describes email deliverability only, which is the easiest attribute to achieve and the weakest predictor of campaign performance.

Accuracy claims in B2B data are close to meaningless as published, because they compress several different questions into one number and almost always answer the easiest one.

Ninety-five per cent accurate usually means ninety-five per cent of emails will not hard bounce. It does not mean the job titles are current, the companies are right, or the people still work there. Those are separate claims requiring separate evidence, and they are the ones that decide whether a campaign works.

The useful response is not to argue about the percentage but to test it. Everything below can be run on a sample before purchase, and together it will tell you more than any vendor statement.

Why Accuracy Percentages Mislead

A single accuracy figure hides four separate questions, and a provider can answer the easiest one honestly while the file remains unusable for you.

Accurate on which field? Email, job title, company and phone have very different error rates.
Accurate for which segment? Providers refresh high-demand segments often and leave smaller ones for much longer.
Accurate as of when? A figure measured at verification says nothing about a record checked two years ago.
Measured how? Deliverability can be tested mechanically. Title accuracy requires comparison against independent sources, which is slow and rarely done at scale.

There is also a structural reason to be sceptical of very high figures. Every verification process produces records it cannot resolve. A provider reporting near-perfect accuracy with no unresolved category has either not run those checks or has absorbed the uncertain records into the clean count.

Key Takeaways

Test, do not negotiate the number. Six checks on a sample take about an hour.
Insist on a random sample from your segment, not a curated one.
Deliverability is table stakes, not a differentiator.
Manual verification of 20 records is the single most informative test.
Fill rate matters only on fields you filter by, not file-wide.
A zero unresolved rate is a warning sign, not a selling point.

The Six Dimensions of B2B Data Accuracy

Before testing, separate what you are testing. Each dimension fails differently and needs different evidence.

Dimension	The question	How it fails
Deliverability	Will the email arrive?	Hard bounce; immediately visible
Identity	Is this a real, distinct person?	Duplicates and fabricated records
Employment	Do they still work there?	Delivers to a dead mailbox or wrong person
Role	Is the job title current?	Delivers, irrelevant, silent failure
Completeness	Are my filter fields populated?	Segment collapses when filtered
Recency	When was this last confirmed?	Unknowable without per-record dates

Role and employment accuracy are where the money is lost, because both deliver perfectly. The campaign report shows good delivery and poor conversion, which teams routinely misdiagnose as a messaging problem and respond to by rewriting subject lines.

Test 1: Insist on a Random Sample

Everything else depends on this. A sample selected by the provider measures their best records; a random draw from your target segment measures the product you will actually receive.

Ask explicitly for a random sample from the segment you intend to buy, not from the database. Two hundred records is enough for mechanical checks and twenty for manual ones. If a provider will only supply a curated sample, treat that as information about the file.

Also request an awkward edge segment alongside the main one. Coverage gaps and source bias surface there first, and performance on a common, high-demand segment tells you little about the file you will be working with.

Test 2: Validate Deliverability

Run the full sample through an email validation service. This is mechanical, cheap and fast, and it establishes the floor rather than the ceiling of quality.

Compare the result against what the provider claimed. A meaningful gap is itself the finding: it means either the data is older than stated or the claim was measured differently from how you are measuring it. Either is worth raising before purchase.

Expect a proportion of indeterminate results, since some mail servers accept all addresses regardless of whether the mailbox exists. Honest providers report those separately rather than counting them as valid, and asking how they are classified is a good test of candour.

24 TO 48 HOUR DELIVERY

Want a Random Sample You Can Test?

Send your target criteria and we will confirm counts and share sample records with verification dates, drawn from your actual segment.

Test 3: Manually Verify 20 Records

This is the most informative hour you will spend, and the test almost nobody runs. Take twenty records and check them against public sources.

Company website — does the organisation exist under that name, and is it still operating?
Team or leadership pages — is the person listed, and is the title the same?
Professional profiles — does the current employer match the record?
Email pattern — if every contact at one company follows an identical pattern, addresses may be derived rather than verified.

Score it simply. Zero or one mismatch in twenty is good. Two or three indicates normal drift and a file that needs refreshing. Four or more means role and employment data is not being independently verified, whatever the deliverability figure says.

Test 4: Count Fill Rates on Your Filter Fields

A file can be large, accurate and still unusable if the attributes you segment on are sparsely populated. This is the most common post-purchase surprise in B2B data.

Count populated cells on the two or three fields you will actually filter by, and ignore the rest. File-wide completeness averages are not informative, because they are lifted by fields nobody uses.

Then apply your real filter combination to the sample and see what proportion survives. A sample of 200 that collapses to 30 when you apply your actual criteria tells you the addressable segment is far smaller than the headline count implied, which is exactly what a confirmed count before purchase is meant to prevent.

Test 5: Deduplicate and Check Dates

Two quick checks that frequently expose how a file was assembled.

Deduplicate on a stable identifier where one exists, then on name plus company domain. If more than roughly 5% collapses out, the count you were quoted was inflated. Avoid deduplicating on email alone, since one person can legitimately hold several addresses.

Check verification dates per record. Not a file-level claim that the database is refreshed quarterly, but a date attached to each row. Calculate the median rather than reading the best few. If the field does not exist, you cannot prioritise re-verification or distinguish a record that was wrong when sold from one that has since changed.

Reading the Results

Use the thresholds below to turn the tests into a decision rather than a feeling.

Test	Acceptable	Walk away
Sample type	Random from your segment	Curated only
Email validation	Within a few points of the claim	Large unexplained gap
Manual check (20 records)	0 to 3 mismatches	4 or more mismatches
Fill rate on filter fields	Above 80%	Below 60%
Filtered survival	Close to the quoted count	Collapses by more than half
Duplicate rate	Under 5%	Above 10%
Verification dates	Present, median under 12 months	Field absent entirely
Unresolved share	Reported and non-zero	Claimed as zero

A provider who passes these is not necessarily the cheapest, but they are selling a measurable product. One who refuses the sample, or cannot answer about unresolved records, is asking you to buy on trust in a category where trust is unusually hard to justify.

Want Counts Confirmed Against Your Exact Filters?

Share your targeting criteria and we will confirm what is available in your segment before you commit.

How MedicoLeads Reports Accuracy

MedicoLeads reports accuracy per segment rather than as a single database-wide figure, because a headline percentage cannot describe the field and segment you are actually buying.

Counts are confirmed against your exact filter combination before purchase, each record carries its own verification date and method, and records that could not be resolved are removed rather than absorbed into the delivered count. Samples are drawn from your target segment so the tests above can be run on the real product.

The layered verification approach, and why deliverability and record accuracy require different evidence, is set out in our explanation of how layered verification works.

Conclusion

Do not negotiate over an accuracy percentage. Test it. A random sample, a validation run, twenty manual checks, a fill-rate count and a deduplication pass will tell you more in an hour than any amount of discussion.

The providers worth buying from are the ones who make this easy, because a measurable product is the only kind you can plan a campaign around. Willingness to be tested is itself the strongest accuracy signal available.

Frequently Asked Questions About B2B Data Accuracy

How can I verify a B2B data provider's accuracy claim?

Request a random sample from your exact segment and run five checks: validate the emails, manually verify 20 records against public sources, count fill rates on the fields you filter by, deduplicate on a stable identifier, and check per-record verification dates. Together these take about an hour.

What does a 95% accuracy claim usually mean?

Almost always that about 95% of emails will not hard bounce. It rarely means job titles are current, companies are correct or the people still work there. Those are separate claims requiring separate evidence, and they are the dimensions that actually determine whether a campaign converts.

Why is a curated sample a problem?

Because it measures the provider’s best records rather than the product you will receive. Always ask for a random draw from the segment you intend to buy. If a provider will only supply a hand-picked sample, treat that refusal as information about the file rather than a procedural detail.

How many manual checks are enough?

Twenty records is sufficient to be informative and small enough to complete in under an hour. Zero to one mismatch is good, two to three indicates normal drift, and four or more means role and employment data is not being independently verified regardless of the stated deliverability figure.

Why is a zero unresolved rate a warning sign?

Because every genuine verification process produces records it cannot resolve. Some mail servers accept all addresses, and independent sources sometimes disagree about employer. A provider reporting none has either not run those checks or has counted uncertain records as clean, which inflates the headline accuracy figure.

What fill rate should I expect?

It depends entirely on which fields you filter by, so measure only those. Above 80% on your actual filter fields is workable; below 60% makes reliable segmentation impossible. File-wide completeness averages are not informative because they are lifted by fields nobody uses.

Should I deduplicate on email address?

No. One person can legitimately hold several valid addresses, so email-based deduplication both misses true duplicates and creates false ones. Deduplicate on a stable identifier where one exists, then on name combined with company domain. More than roughly 5% collapsing out means the quoted count was inflated.

What if a provider refuses to supply a sample?

Treat it as a decision rather than an obstacle. In a category where accuracy claims are unverifiable by inspection, willingness to be tested is the strongest quality signal available. A provider who will not let you measure the product is asking you to buy on trust alone.

Ready to test a sample against your own criteria?

Tell us the roles, industries and regions you need. We will confirm counts and share sample records drawn from your segment.

Sources and References

ABOUT THE AUTHOR

William Coogan

William Coogan is a healthcare marketing professional with 15 years of experience in communication strategy. At MedicoLeads, he writes about data-driven outreach that helps healthcare providers reach and engage their audiences.

FREE SAMPLE

Get a sample you can actually test

Random sample records from your target segment, with verification dates attached.

No commitment. Reply within one business day.

Popular lists

TALK TO US

Ready to Buy Data You Can Measure?

Tell us who you want to reach. Our data experts will confirm available counts and share a sample within one business day.