How to test a B2B data provider before you sign
Updated
Never buy on a vendor's stated accuracy rate. Run a test on your own sample, score every vendor on the same grid, and decide on cost per usable contact rather than price per number.
The protocol below takes two weeks and five steps. It is published in full, spreadsheet included, and it applies to Listar exactly as it applies to anyone else.
Why a vendor-supplied sample proves nothing
Because you are not being shown a random slice of the database.
A demo file is selected, and the selection is the product. The accounts in it are the ones where coverage is strongest, usually large well-documented companies in the vendor's home market. Your actual target list is not that, and the gap between the two is precisely what you are trying to measure.
The second problem is the denominator. A vendor reporting 95% accuracy on numbers delivered, while delivering numbers for 40% of your list, is reporting something very different from a vendor at 80% on 80% of the list. Both sentences contain a high percentage. Only one of them fills your calendar.
So the test has to run on a list you choose, and every figure has to carry its denominator.
A note on the thresholds you will be offered
Most guidance you will find on this question comes from vendors, and the numbers in it are usually unsourced.
The most frequently cited guide in this category recommends a 20 to 30 call spot check and a 500 to 1,000 contact pilot, which is sound. It then gives a connect rate benchmark of 15 to 25% attributed to "many outbound teams", and an annual decay estimate of 25 to 35% attributed to "vendor and analyst commentary". No study, no sample, no method, for either.
Those ranges may well be about right. The point is that you cannot calibrate your purchase against them, because nobody can tell you what they were measured on. Set your thresholds against your own current baseline instead. If you do not have one, the first run of this protocol becomes it.
The protocol, in five steps
Step 1. Build the sample, do not accept one. Take 300 to 500 contacts from your real target list, stratified the way your pipeline actually is: by country, company size and seniority. Include the segments you expect to be hard. A sample that excludes them measures the wrong thing.
Step 2. Include contacts you already know. Twenty to fifty records where you already hold a confirmed mobile. These are your control group, and they are the only way to catch a vendor that returns plausible-looking numbers for people it has no data on.
Step 3. Send the identical list to every vendor. Same file, same fields, same day. Ask each one for its own method, and record what you were charged, including charges for records where nothing was returned.
Step 4. Call, and score with one grid. Same reps, same script, same time windows across vendors. Score every attempt against the grid below, with no free-text outcomes.
Step 5. Compute four numbers, then decide. Not before. Vendors are good at reframing a result mid-test.
The scoring grid
Every call outcome maps to exactly one of these. Allowing an eighth category is how a test becomes unusable.
| Outcome | Counts as |
|---|---|
| Number does not exist | Data failure |
| Number not assigned | Data failure |
| Wrong person | Data failure |
| Former employee | Data failure |
| Company switchboard | Not a direct line |
| Personal voicemail with the right name | Right line, no conversation |
| Right person reached | Success |
The distinction between the last two matters more than it looks. A named personal voicemail confirms attachment without producing a conversation, so it belongs in your reachability numerator and outside your conversation numerator. Vendors that report "connect rate" rarely say which of the two they mean.
The four numbers
coverage = numbers delivered / contacts submitted
right-person rate = right person reached / numbers delivered
reachability = (right person + named voicemail) / numbers delivered
cost per usable contact = total spend / right person reached
The fourth one decides. The first three explain it.
Run them per segment as well as overall. A vendor can win on aggregate and be unusable on the one country that carries your quarter, and that is the single most common outcome of running this test honestly.
Set your thresholds before you see any results
Write them down before the first number comes back, and have whoever owns the budget sign off on them. A threshold set after the results is not a threshold, it is a justification.
Decide in advance:
- The minimum coverage below which a vendor is out regardless of quality.
- The minimum right-person rate below which cheap stops being cheap.
- The maximum cost per usable contact your average deal size supports.
- The per-segment floor for the geography or seniority you cannot afford to miss.
The clause to negotiate
Once you have a measured baseline, it belongs in the contract.
Ask for the right-person rate you measured to be stated as a service level, with a remedy if it is missed on a future delivery: credits returned, or the delivery re-run. Vendors confident at the attachment level will discuss it. Vendors that verify only the line will decline, and the decline is information you did not have before.
Two other terms worth settling at the same time, because they move your real unit cost more than the rate does: whether you are charged when nothing is found or when a number reaches the wrong person, and what the commitment term, notice period and renewal uplift are.
The spreadsheet
The grid, the formulas and the thresholds are in a workbook you can download and use as it stands. It takes a 300-contact sample, up to 600 call records, computes the four numbers automatically and holds your thresholds next to the results.
No form, no email required. The page argues the protocol should be public, and putting it behind a capture form would contradict that.
Our own results, under this protocol
Listar measured 95% right-person rate on delivered mobile numbers: 10,000 real calls across 100 separate files of 100 numbers each, between May and September 2026.
What that figure is and is not: internal, not audited by a third party. The denominator is numbers delivered, not contacts submitted. It was measured by real calls, not by querying a verification database. And it is a French sample, which is where our coverage is strongest.
At our published rate of €0.50 excluding VAT per mobile, that puts cost per usable contact at €0.53.
Run the protocol against us. If the numbers come out differently on your sample, those are the numbers that should decide your purchase, not ours.
FAQ
How large does the sample need to be? Three hundred to five hundred contacts gives a reliable overall reading. Below a hundred, the differences you observe are as much noise as signal. Per-segment conclusions need their own volume, which is why stratifying the sample matters more than enlarging it.
How long does this take in practice? Two weeks. A few days for the vendors to return the enrichment, one week of calling, a day to compute and compare.
Can I run it on a single vendor? Yes, and it is worth doing on your current one before you shop. Most teams discover their incumbent's real right-person rate for the first time and find the comparison much easier afterwards.
What if a vendor refuses to enrich my sample? That is a result. A vendor confident in its data will enrich a list you chose, usually at its own expense for a few hundred records. A refusal, or an insistence on supplying the sample itself, tells you what the test would have told you.
Should I tell vendors they are being compared? Yes. It changes nothing about the data they hold, and it usually improves the commercial terms you are offered during the test.