B2B Database: What to Test Before You Pay

A sales lead compares two contact databases on the number each one puts at the top of its page, picks the larger, and signs an annual contract. Three weeks later the team runs its actual segment against the product, which is mid-sized logistics companies in two countries with a named role, and the query returns a few hundred usable rows out of hundreds of millions.

Nothing in the comparison predicted that, because nothing in the comparison was about that segment. The total record count is the number every vendor in this category leads with and the one that decides least about whether the product works for you.

What follows is the set of questions that does decide it, in the order worth asking them, and the one test that settles the whole purchase in an afternoon. It is deliberately not a ranked list of vendors. Which company sells the best database is a question whose answer changes by segment, by geography and by quarter, and a ranking published today is a claim about somebody else's segment.

The one test that replaces every published number

Take a hundred companies you already know belong in your target segment. Not a sample the vendor picks and not a filter you build inside their interface. A hundred names your team could defend in a room, sourced from your own closed-won list, your own pipeline and a manual scan of the market.

Hand that list to the vendor during the trial and ask for a match. Count how many of the hundred the product finds at all, then count how many come back with the specific field your campaign is going to filter on, which is usually a role rather than a company attribute. Then take twenty of the returned people and check them by hand against the company's own site and their own professional profile.

That is a day of work and it produces the only three numbers that describe your purchase: what share of your segment exists in the product, what share of those carries the field you need, and what share of the returned values is currently true. Every published figure is a claim about a population you cannot inspect, which is the same problem match rate describes from the vendor side.

Total record countWhat the page leads with

• Counts every contact in every market the vendor sells into
• Grows whenever the vendor adds a geography you do not sell into
• Comparable across vendors and predictive of nothing
• Costs the vendor nothing to increase
• Cannot be wrong, because no definition of a usable record is attached

Coverage of your segmentWhat you are actually buying

• Counts only companies that pass your own written definition
• Falls when your segment is narrow, which is when it matters most
• Measurable only on a list you built yourself
• Decides the ceiling on every campaign you run this year
• Frequently an order of magnitude below what the headline implies

The reason to do this during a trial rather than after signing is that the answer is not negotiable afterwards. A database either holds your segment or it does not, and no amount of account management changes that inside a contract term.

Coverage and accuracy are two numbers and only one of them gets quoted

Coverage is cheap to compute and easy to publish. Submit rows, count how many came back with a value, print the percentage. Accuracy needs a person to check whether the returned value is true today, one row at a time, against a source that is itself current.

The same asymmetry decides what a sales lead database is worth, because the count on the page is a claim about every market the vendor sells into and the coverage that matters is of the segment you wrote down.

A provider returning a plausible but stale job title for most of a list scores identically on coverage to a provider returning the correct title for the same share. The number that reaches a comparison spreadsheet is the one that is cheap to produce. CRM enrichment works through the same split on the enrichment side, and the consequence is identical here: stale data is worse than missing data, because a blank field gets checked before somebody writes to that person and a confidently wrong field gets used.

The field that goes wrong most expensively is the employer. A message naming somebody's previous company is not a near miss. It is evidence to the reader that nobody checked, delivered to exactly the person you wanted to impress, and the mechanism behind it is ordinary data decay rather than a bad vendor.

The meter is where the money actually moves

Every vendor in this category sells credits and no two of them mean the same thing by the word. This is the part of the purchase that is genuinely readable in advance, because the vendors publish it, and it is the part buyers skip because it looks like small print.

Two examples from the vendors' own pricing pages, which are quoted here because they are unusually explicit rather than because they are recommendations.

UpLead's pricing page states that "A credit unlocks a contact for download or CRM export and gives you access to their email and mobile direct dial." and that "One credit = one contact." So the meter is the export, one for one, as published on uplead.com on 2 September 2026.

Cognism's pricing page states that "Credits are used to reveal, enrich, or export a contact." and that "1 credit = 1 revealed contact", adds "No charge for viewing previously revealed contacts", and says "Each seat includes an allocation of credits as part of the subscription." On its enrichment side the same page states that "Credits are only used again when key details change, like a job move". So the meter is the reveal, seats carry allocations, re-viewing a record you already paid for is free, and maintenance of a record you already hold is not a new charge, as published on cognism.com on 2 September 2026.

Those are two coherent and quite different pricing models sitting behind one word. The questions that fall out of them apply to any vendor in the category, and the answers belong in writing before the contract rather than in a sales call.

Does a credit spend on submission or on a result. A tool that charges for a lookup that returned nothing is charging you for its own coverage gap.

Do credits reset or roll over, and what happens to unused ones at renewal.

Are seats and credits separate lines, so that adding a person to the team changes the bill twice.

Does re-exporting a record you already paid for cost again, which decides whether keeping the data in your own system is worth the engineering.

Is enrichment of an existing record metered on the same pool as new discovery, and does a refresh of a record whose details did not change cost anything.

The arithmetic that follows is invented for illustration and is not a measurement of any vendor. A team enriching four thousand records a quarter pays four thousand credits under a per-submission meter whatever the match rate, and pays roughly two thousand four hundred under a per-result meter at a sixty percent match. Same product, same work, a difference of about forty percent on the line item, decided entirely by a sentence on the pricing page.

Recency is a property of the record, not of the vendor

A vendor refresh cadence stated as a company-wide policy tells you very little, because refresh is not applied evenly. Records that many customers query get re-verified often, and records in a thin segment can sit untouched for a long time. The refresh a vendor advertises is an average over a population weighted towards the popular rows, and your segment is unlikely to be the popular rows.

Two questions get past the policy statement. Does the product expose a last-verified date on each record, at field level rather than at row level, so that your own pipeline can decide what to trust. And does the vendor distinguish between a record it re-checked and a record it merely re-imported from the same upstream source it took the row from originally.

Where the answer to the first question is no, the practical response is to verify immediately before the send rather than at build time, which is the discipline set out in list hygiene and priced in email list cleaning. A verification result is a claim about one moment, and a file verified at purchase and sent a month later has been unverified for a month.

What the contract says about the rows after you leave

This is the question that separates the three things sold under the word database, which email marketing database works through in full: a database you own and build, subscription access to somebody else's, and a purchased export.

Subscription access is the middle option, and it is the one on the table whenever a vendor is selling seats and credits rather than a file. Read what the terms say about the rows you exported during the term. Some subscriptions permit continued use of exported records and some do not, and the difference decides whether the contract is a purchase or a rental. It also decides how much engineering it is worth doing to hold the data in your own systems, because building a pipeline into a store you have to empty at renewal is work with an expiry date on it.

The second contract question is exclusivity, which does not exist and should not be expected to. Every competitor in your category queries the same product with broadly similar filters, so a segment defined by three common conditions has been pulled repeatedly by companies selling adjacent things. The rows are correct and the audience is tired.

Before you sign

• ✓Yes: A hundred companies from your own segment were matched during the trial and the result was counted
• ✓Yes: Twenty returned people were checked by hand against the company site and their own profile
• ✓Yes: The credit meter is written down: on submission or on result, and what a miss costs
• ✓Yes: Seat count and credit pool are separated, so team growth changes one line rather than two
• ✓Yes: Records carry a last-verified date you can read and filter on
• ✓Yes: The terms state what happens to exported rows when the subscription ends
• ✕No: The decision was made on total record count because it was the only comparable number
• ✕No: Accuracy was accepted from a published percentage rather than measured on your own sample

When a database is the wrong purchase

Two situations, and they are common enough to be worth naming before the trial rather than after it.

The first is a segment defined by something no vendor sells. Companies running a particular piece of software, companies that opened a second location this year, companies whose service mix matches a specific offer. Those definitions are frequently the good ones, because they carry a reason to write rather than a demographic, and they are exactly the definitions a filter menu cannot express. When the ideal customer profile is written in terms the product does not stock, enrichment fills the fields the vendor happens to have and the campaign still cannot be built. Writing the segment query first and listing the fields it needs, as the ideal customer profile guide sets out, is what surfaces that mismatch before the invoice rather than after it.

The second is a programme small enough that the subscription is the wrong shape. A team contacting a few hundred carefully chosen companies a quarter is buying an annual platform to answer a question that a per-campaign build answers better, and the per-campaign build produces a list nobody else is working. The mechanics of that route, and the distinction underneath it between a filter and a signal, are the highest-leverage idea in B2B prospecting.

1. Step 1Write the segment first

   Ranges, lists and exclusions, specific enough that somebody outside the team could build the list from the words alone.

2. Step 2List the fields the query needs

   Usually four rather than forty. Anything the campaign does not filter on is not part of this purchase.

3. Step 3Build the hundred-company control list

   From closed-won, from pipeline and from a manual market scan. Never from the vendor's own interface.

4. Step 4Match the control list in the trial

   Count matches, count the field you need, then hand-check twenty of the returned people.

5. Step 5Read the meter and the exit terms

   How a credit is spent, what a miss costs, and what happens to exported rows at the end of the term.

Where we stand

RevenueFlow builds lists per campaign rather than maintaining a subscription database, and every campaign carries exactly one message with no bumps and no thread replies. Those two positions are connected. A single message only works when the list is tight enough that the message is true for everybody on it, and a filter menu that can express three demographic conditions will not produce a list that tight. The narrowness is the product.

That is a preference rather than a rule for everybody. Teams running high volume across many segments get real value out of a subscription, and the evaluation above is written for them as much as for anybody. What does not survive scrutiny is choosing between two of them on a headline record count, which is a comparison of two marketing departments.

For the fields a database cannot supply, the ordering that keeps the cost down is in waterfall enrichment, and the selection question that sits above all of it, which person at the company is worth writing to at all, is in finding and reaching decision makers.

The short version

Total record count is the number every vendor leads with and the one that predicts least. Coverage of your own segment, measured on a hundred companies you chose yourself, is the number that decides whether the product works, and it can only be measured during a trial.

Coverage and accuracy are two different numbers and only the cheap one gets published, so hand-check twenty returned rows before believing any percentage. Read the credit meter in writing, because vendors publish genuinely different definitions of the same word and the difference between charging on submission and charging on result moves the bill by a wide margin.

Ask whether records carry a readable last-verified date, verify immediately before the send whatever the answer, and read what the terms say about exported rows at the end of the term. Where the segment is defined by something no vendor stocks, or the programme is small enough that a per-campaign build produces a better list, the honest answer is that this is the wrong purchase.

If the real question is whether a list built against your own definition beats a subscription you can sign this afternoon, see what a first campaign produces and compare the two on replies rather than on row count.

Credit and meter wording is quoted from UpLead's and Cognism's own pricing pages, fetched 2 September 2026. Vendors change these terms. Verify current terms with the vendor before relying on them.