How do you ask a supplier for a sample?
Most B2B data suppliers will provide a sample if you ask directly, but the quality of the sample you receive depends heavily on how you frame the request. A vague ask ("can you send us some records?") often results in cherry-picked contacts from the supplier's cleanest verticals. A precise ask tied to your actual criteria is harder to game.
When you make the request, specify the exact filters you plan to use on the full order: target UK SIC 2007 codes, job titles or job functions, company size band, and geography (region or county, or specific Royal Mail postcode areas). Ask explicitly that the sample be drawn randomly from records matching those criteria, and confirm in writing that it comes from the same compiled file that will generate your full delivery.
A legitimate supplier will agree to this without hesitation. Pushback ("we only send pre-selected samples") is itself a data point worth noting.
What sample size do you actually need?
Fifty records is the absolute floor. Below that, a single cluster of stale records from one industry segment can distort every metric you calculate. At 200 records you start to see meaningful variation across sectors and geographies. For most purchases under 10,000 contacts, 100 to 150 records gives you a statistically reasonable picture without consuming hours of verification time.
One rule of thumb: if you are buying more than 5,000 contacts, push for at least 200 in the sample. The additional verification time is worth it because the cost difference between a 70%-accurate file and a 90%-accurate file, measured in wasted calling time or bounced emails, is material at that volume.
The six evaluation dimensions
1. Completeness: are all expected fields populated?
Open the sample file and calculate a fill rate for each field. Count the number of populated cells in each column and divide by the total record count. Do this for: company name, registered address, postcode, main telephone, direct-dial or mobile, email address, job title, SIC code, and company size (employees or turnover band).
Expect 90% or higher for company name, postcode, and job title in a well-maintained file. Direct-dial numbers typically appear on 60-75% of records in most UK commercial sectors; mobile numbers are sparser, often 30-50%. Email addresses at corporate domains should reach 70-80% for senior roles in well-documented industries such as financial services or professional services. If any core field falls below 60%, ask the supplier to explain why before proceeding.
2. Accuracy: does the data match independent sources?
This is the only dimension that requires manual effort, and it is the most revealing. Pick 20 records at random, then verify each one against the company's own website, the relevant LinkedIn company page, or the free Companies House public search at find-and-update.company-information.service.gov.uk.
For each record, check: Does the company still trade at this postcode? Does the job title correspond to a real role at this company? Is the email domain correct? If a direct-dial number is given, does it connect to a real individual rather than a switchboard or dead line?
Three or fewer failed verifications across 20 records is an acceptable accuracy rate (85%+). Four to six failures (70-80%) warrants a conversation with the supplier about when the file was last updated. Seven or more failures means the file is likely to cause significant campaign waste, and you should either decline or negotiate a substantial price reduction that reflects the expected clean-up cost.
What "accurate" means in practice
An accurate record is one where the person named still holds the role described at the company shown, reachable on the channel supplied. A record where the company has moved premises but everything else is correct is borderline. A record where the named contact left two years ago is a failure regardless of whether the phone number still rings.
3. Freshness: when were these records last updated?
Ask the supplier for the "last verified" or "last updated" date on records in the sample. Most well-maintained B2B files carry a field-level or record-level timestamp. UK B2B contact data decays at roughly 30-35% per year for decision-maker roles: people change jobs, companies restructure, and contact details shift. A file last updated 18 months ago may be 45-50% stale by the time you call it.
If no timestamp exists, the manual spot-check from the accuracy step is your proxy for freshness. Multiple records where the named person has left or the company has changed are a signal that the file has not been refreshed recently.
4. Targeting precision: does the sample reflect what you asked for?
Check that the SIC codes, job titles, and geographies in the sample actually match your brief. This sounds obvious, but misclassification is common. A file classified as SIC 62020 (IT consultancy) may include records from SIC 62090 (other IT and computer service activities), which might include managed print suppliers, IT resellers, and desktop support companies alongside the software consultancies you actually want.
Similarly, "Director" in a job title field could mean a Managing Director of a 200-person firm or a Director-level individual contributor at a company with 40 directors. Look at actual job title strings, not just the category label the supplier applied. If the targeting precision looks loose, ask to see the filter logic the supplier applied, and consider whether a more tightly defined industry filter would improve quality.
5. Suppression cleanliness: are dead records and opted-out contacts removed?
Within the sample, look for obvious dead records: company names you recognise as having ceased trading, postal addresses marked as vacant, or phone numbers that already appear on your existing do-not-contact list. The presence of companies in administration or liquidation in a "currently trading" file is a clear quality failure.
For telemarketing use, ask the supplier to confirm in writing that phone numbers in the sample have been washed against the Telephone Preference Service (TPS) within the last 28 days. Under the Privacy and Electronic Communications Regulations (PECR), calling a TPS-registered number without a prior business relationship is a regulatory breach, so written confirmation is essential before you use the data for outbound calls. Read more about the cost implications of suppression quality in our guide to B2B data pricing in the UK.
6. Uniqueness: how much overlaps with your existing CRM?
Export your CRM contacts into a spreadsheet and cross-match the sample on email domain plus job title, or on company name plus postcode. Even a rough match gives you a suppression rate to apply to the full file. If 20 records out of a 100-record sample already exist in your CRM, budget for 20% suppression on the full purchase and factor that into your cost-per-new-contact calculation. A supplier who offers a formal pre-purchase suppression match (where they run your CRM against their file before you pay) is removing a meaningful financial risk from the transaction.
The 20-record manual verification protocol
This process takes about 45 minutes for 20 records. It is worth doing before any purchase above a few hundred pounds.
- Export the sample and select 20 records at random (use a random number generator if the file is sorted by alphabet or SIC code, to avoid cluster bias).
- For each record, open the company website using the registered domain from the email address or a web search on the company name and postcode.
- Look for a "team" or "about us" page. Confirm the named individual or an equivalent role exists at the company.
- Cross-check the postcode against the address shown on Companies House. If they do not match, note whether the discrepancy is a minor formatting difference or a substantively different address.
- If a direct-dial number is given, call it. A connected call to a real person or a personalised voicemail is a pass. A generic switchboard message is neutral. A disconnected line or "number not recognised" is a fail.
- Log each record as Pass, Borderline, or Fail. Total the fails at the end and calculate your accuracy rate as (Passes / 20) x 100.
Borderline records, such as someone who has changed role within the same company, count as half a failure for this purpose. In our experience, any supplier whose random 20-record check returns fewer than 3 failures is providing data worth serious consideration.
What does a good sample look like vs. a poor one?
| Dimension | Good sample | Poor sample |
|---|---|---|
| Completeness (core fields) | 90%+ fill rate on company name, postcode, job title | Significant blanks in key fields; or 100% across every field (suspiciously clean) |
| Manual accuracy check (20 records) | 17+ records verified independently (85%+) | 7+ failures (below 65%); multiple companies ceased or relocated |
| Freshness | Record-level timestamps within 12 months; few stale job titles | No timestamps available; multiple "left company" contacts on spot-check |
| Targeting precision | SIC codes and job titles match your brief; no obvious off-target records | Mixed SIC categories, generic job title strings ("Manager", "Director"), mismatched regions |
| Suppression cleanliness | No ceased-trading companies; written TPS confirmation available | Companies in administration present; no TPS confirmation offered |
| Uniqueness vs. your CRM | 10% or lower overlap with your existing contacts | 20%+ overlap without an offered suppression match before purchase |
What is sample bait-and-switch, and how do you avoid it?
Bait-and-switch in data sales means a supplier sends a small, hand-picked sample of their cleanest records, then delivers a full file compiled to a different or lower standard. The sample passes every test; the full file is significantly worse.
Three warning signs are worth knowing. First, a sample with 100% field completion across all contact fields when your sector typically sees 70-80% on email addresses and 60-70% on direct dials. Second, a sample with zero obviously stale records, including no company that has changed address or no individual who has changed role in the past two years. Third, a supplier who returns the sample within an hour of a Monday morning request for a niche industry filter.
The most practical defence is to insist in writing that the sample was drawn randomly from the same compiled query that will generate the full delivery. Follow that with a clause in the purchase agreement that gives you a right of dispute if the full file accuracy rate (as measured by your same spot-check protocol) falls more than 15 percentage points below the sample accuracy rate. Most reputable suppliers will accept this. Those who resist it are telling you something important.
For context on what a direct-dial number specifically should look like in a quality file, see our guide to B2B direct-dial numbers in the UK.
When should you ask for a larger sample?
A standard 100-record sample is sufficient for most purchases. Ask for a larger sample, 200 or more records, in the following situations:
- The purchase order is above 10,000 contacts, where even a 5% accuracy shortfall represents 500 wasted records.
- The targeting is narrow, such as a single SIC code or a specific county, where a small sample may not show natural variation within the file.
- You are evaluating a supplier for the first time with no prior purchase history to rely on.
- The initial 100-record sample shows borderline results: say, 80% accuracy rather than clearly passing or failing.
- You need the file for telemarketing specifically, where call-connection rate is directly tied to direct-dial accuracy, and accuracy failures cost real time.
A supplier who refuses to provide more than 50 records regardless of the order size, or who charges for a sample, is an immediate caution flag. Samples are a cost of sale for the supplier. Treating them as a premium add-on usually signals either a genuinely limited file or a commercial model built on low-scrutiny buyers.