How to Evaluate a B2B Contact Database Before You Sign

B2B contact-database purchases fail predictably: excellent sample file, twelve-month contract, six months later deliverability collapsed and titles stale. The sample was curated; production was not. Experienced operators run a five-question diligence process before any signature. GSDSI maintains 486M+ global contacts via B2B prospecting: here is the framework buyers use. Pair with CRM enrichment QA, seed match testing, and RFP scoring matrix.

Coverage Against the Actual Buyer. Not the Total Universe

CFOs at venture-backed SaaS 100-1,000 employees in North America might be ~40,000 people universe. A 400M global file may cover 25,000; a focused file may cover 38,000. Ask ("How many match this ICP) prove with hashed CRM match without seeing our data?" Clean-room match separates serious vendors from storytellers. Standard in mature B2B prospecting programs.

Decay Rate and Refresh Cadence

Ask re-verification cadence for email and source for title changes. Best providers combine bounce re-validation, job-change APIs, and cooperative signals with per-field refresh schedules. Vendors vague on decay outsource rot to your marketing team. Align Core Email File refresh with base file cadence.

Phone-Number Provenance and TCPA Posture

Mobile direct dials need sourcing and consent documentation under TCPA and TSR. If the consumer disputes a call, what evidence exists? Vendors without per-record artifacts pass liability to buyers at campaign launch.

Match-Rate Testing on the Buyer's Actual Use Case

Send hashed 10K CRM rows; measure fill on fields you activate:

• Verified business email
• Direct dial (landline vs mobile; consent on mobile)
• LinkedIn URL (title freshness proxy)
• Job-change date
• Technographic/firmographic depth for segmentation

Skipping this step buys 60% solutions at 100% price. Feed results into enterprise pilot checklist.

Pricing Governance at Signing and Renewal

Cap per-seat overages, renewal price stability, termination for documented systemic quality failure, not only non-payment. Vendors refusing all three signal renewal dynamics. Teams feeding identity graphs need stable enrichment economics mid-contract.

Weekly QA sample: 200 records stratified by tier; disposition trends in vendor QBRs beat match-rate slides alone.

Coverage Is Never Flat Across Seniority or Company Size

Aggregate coverage figures hide the shape that determines whether a database works for you. Contact data is systematically denser for mid-level roles at large, well-documented companies, and thinner in the two places most sales teams need it: senior executives who keep a low public profile, and small companies with little published presence. A database quoted as strong overall can be weak in exactly the cells your territory depends on.

Ask for coverage as a grid rather than a number, seniority against company size, restricted to your target industries. Then check the cells you actually sell into. A vendor unwilling or unable to produce that grid is asking you to accept an average as a description of a distribution, and the shape of that distribution decides how many of your named accounts are reachable at the level where budget sits.

How a Record Was Obtained Predicts How Long It Lasts

Two records that look identical in a sample can behave very differently, because a contact assembled from public sources and one confirmed through a direct interaction decay at different rates and fail in different ways. Provenance also determines what you are permitted to do with the record, which is a separate question from whether the values are accurate and is easier to answer before signing than after.

Ask how each field is sourced, how it is refreshed, and what evidence exists that the person still holds the role. Job changes are the dominant decay mechanism in B2B, and a database that refreshes company attributes frequently while leaving person-level roles stale will look fresh in an audit and route a meaningful share of your outreach to people who moved on. Test that specifically by checking a sample of records against current public profiles rather than against the vendor’s own timestamp.

The Clauses That Matter More Than the Price

B2B contact contracts are usually negotiated on record volume and unit price, while the terms that determine the value are elsewhere. Whether records already exported remain usable after the term ends, whether credits expire, whether suppression lists you supply are honored across the whole platform, and what happens to your own enriched data on termination all matter more over two years than a modest difference in rate.

Settle the exit terms while you still have leverage, which is before signing rather than at renewal. Ask specifically what you retain if you leave, what must be deleted, and how a mid-term coverage shortfall is remedied, since a coverage commitment with no remedy attached is a description rather than an obligation. See the RFP scorecard for scoring these alongside the technical criteria.

Match, Fill, and Accuracy Are Three Different Results

The hashed CRM test returns three numbers, and vendors tend to report them as one. Match asks whether the vendor found a record for the person you sent. Fill asks whether the field you plan to activate is populated on that record. Accuracy asks whether the populated value is currently true. A file can match nearly everything you send, fill a mobile number on most of it, and still be wrong on the value that decides whether a dial connects. Report those three columns for every field group and every ICP tier, or you will approve a purchase on the friendliest of the three.

Accuracy is the column you have to grade yourself, because a vendor cannot mark its own answer. Hold back a slice of records where you already know the truth from a recent outcome, such as a call that reached the right person, an email that drew a human reply, or a title confirmed on a live call. Run those through the same test and compare field by field. Where the vendor's value disagrees with yours, ask which source produced it and when that source was last re-verified. Two vendors with similar match rates often separate sharply once the disagreements are graded rather than counted.

Insist that the test file be produced by the production pipeline instead of by an analyst. Specify the query yourself, ask for a random draw rather than a curated selection, and require the null profile for every field in the schema, not just the fields the demo highlighted. Then request a second draw against the same query after a refresh cycle and diff it, since that shows records actually moving instead of a static extract relabeled as current. See seed match testing for the draw mechanics.

Write the pass criteria before results land, per field group, so a strong firmographic fill cannot carry a weak mobile-consent result across the line. Then put those agreed thresholds into the contract as the working definition of the systemic quality failure your termination clause refers to.