A B2B leads database is a maintained store of companies and the people at them, each record carrying a contact channel, firmographics and a history of touches, built from inbound signups, exports and enrichment and kept usable by regular verification. The usable share is the measure that matters: across 6.5 million non-freemail (business and ISP) addresses verified in 2026, 14.5% were invalid and a further 27% could not be confirmed, so a database that is never re-checked quietly loses a fifth of its reach every year.
A leads database is a set of records, not a file. Each field decays at its own pace, and the contact channel decays fastest. The table lists the fields a working B2B record carries, where each usually comes from and how quickly it goes wrong.
| Field | Usual source | Goes stale | What stale looks like |
|---|---|---|---|
| Work email | Signup, finder, export | Fastest: only 19.2% still valid after 90 days (decay study) | Hard bounce, or a silent catch-all accept |
| Job title and seniority | LinkedIn, enrichment | Fast: promotions and moves | Right person, wrong level, wrong pitch |
| Company | Domain, enrichment | Medium: moves and acquisitions | Email still works at the old domain for a while, then dies |
| Company size and revenue | Enrichment | Slow: annual | Segment drifts; a 40-person firm is now 200 |
| Phone | Enrichment, signup | Medium | Reassigned numbers, switchboards |
| Technology stack | Enrichment | Medium | Trigger fired on a tool they replaced |
| Consent and source | Your own records | Never, if you log it | Missing: you cannot show why you may email them |
| Last verified date | Your verifier | By definition | Older than 90 days means re-check before sending |
The last two fields are the ones most databases lack. Without a source and a verified-at date, every other field is a guess with a confident label.
Source sets the starting quality and the decay curve. A database is usually a blend of all four; knowing the blend tells you how often to re-verify each slice.
Typed by the owner with intent. Starts at 5 to 10% invalid from typos and throwaways, decays with job changes only. Validate at the form through the real-time API and the slice stays the cleanest you have.
Every lead anyone ever touched. Starts clean, but a three-year-old CRM holds contacts verified by nobody since the day they were entered. Expect 25 to 35% invalid on records older than two years.
Apollo, ZoomInfo, Lusha, Cognism and friends. 10 to 25% invalid on the day of export; the vendor "verified" badge is historical. Our 100,000-contact Apollo study measured 5.8% invalid and 18.4% catch-all at export time.
Highest invalid share, most role addresses, most catch-all domains. Benchmark one invalid address in seven on a list not verified in 90 days; .au lists reach 36.5%. Quarantine these in their own segment until verified.
These are the rates from the BounceZero corpus, April to October 2026, 7.93 million addresses verified for customers. A business database that is maintained should beat the business row; one that is not will drift toward the bought-list rates.
Unknown means the mailbox could not be confirmed either way, usually because the receiving gateway refuses verification probes. A vendor that reports zero unknowns on a business list is guessing on a quarter of it.
| Segment | Addresses | Invalid | Catch-all | Unknown | Confirmed valid |
|---|---|---|---|---|---|
| Non-freemail (business and ISP) domains | 6,514,884 | 14.5% | 2.3% | 27.0% | 32.6% |
| Freemail on B2B lists | 1,415,154 | 17.6% | 1.1% | 2.1% | 74.0% |
| .com | 2,158,955 | 16.4% | 4.4% | 14.7% | 59.7% |
| .uk | 21,194 | 17.0% | 3.7% | 10.6% | 65.0% |
| .de | 2,095,440 | 13.9% | 0.0% | 41.2% | 16.5% |
| .org | 56,401 | 24.6% | 12.4% | 25.0% | 36.2% |
| .edu | 17,961 | 32.8% | 8.3% | 18.5% | 38.2% |
Full cuts by provider and country: Lead List Quality Benchmarks 2026. The .de unknown rate is driven by one ISP gateway and is not a data quality problem; route those records to phone or LinkedIn rather than deleting them.
Every vendor sells a snapshot. The moment the export lands, it starts aging at the rate in the first table, and nothing the vendor did before the export slows that down. A database is the process that keeps the snapshot usable: verification on entry, a verified-at date on every record, a re-check cadence, and a rule for what happens to records that fail.
The cost comparison makes the point. A 50,000-record database re-verified monthly costs twelve runs of 50,000 checks a year, which is 600,000 credits, or an Enterprise pack and a Scale pack at $1,480 together. The same database bought fresh from a vendor each quarter costs the vendor's per-record price times 50,000, four times, and still arrives 10 to 25% invalid.
The second reason is the sending domain. sending platforms suspend accounts whose bounce rate passes 2%, and Google and Microsoft throttle senders whose complaint rate passes their published 0.3%. A database with 14.5% invalid addresses cannot be mailed from a domain you care about, whatever it cost.
The loop below keeps a database above the business benchmark without a data team. It runs on a calendar, not on a bounce report.
Forms through the real-time API, imports and exports through a bulk job the same day. Store the result and the date on the record. Invalid never enters; catch-all and unknown enter flagged.
Four segments: confirmed valid, catch-all, unknown, and invalid-held. Sequences read the segment, so a record moving from valid to catch-all leaves the cold sequence automatically.
A saved filter on verified-at date, exported monthly, run as one bulk job, written back. Roughly a third of the database each month; the whole base is never older than a quarter.
Valid to invalid on a business address usually means a job change. That is a trigger to find the person at the new company, not a row to delete.
Compare your invalid, catch-all and unknown rates against the table above. Drift toward the bought-list rates means a source is polluting the base; find the segment and fix the entry point.
Subscription databases and a maintained in-house base are not alternatives; most teams run both. The table sets out what each is for and where the verification step sits in each.
| Option | What it is good for | Email accuracy at export | Where verification sits |
|---|---|---|---|
| ZoomInfo, Cognism, Apollo, Lusha | Discovering new accounts and people by filter | 10 to 22% invalid at export (vendor claims 81 to 95%) | After every export, before import |
| Clay and waterfall enrichment | Filling gaps on known names | 70 to 80% accuracy across sources | After the waterfall, before the CRM |
| Your CRM as the system of record | Every touched lead, consent and history | Whatever your maintenance loop makes it | On entry and before every send |
| Bought files | Volume in a niche the platforms miss | About one in seven invalid unverified | Sample first, then full list, then quarantine |
Platform choice: B2B contact database providers compared. Evaluating a one-off file: B2B database for sale.
Reachable share: confirmed-valid records divided by all records, measured monthly. The corpus business benchmark is 32.6% confirmed, with another 27% unconfirmed but not dead. A maintained database with verification on entry runs at 70% and above. If yours is below 50%, the entry points are the problem, not the re-check cadence.
Sources ranked by quality, build versus buy
Platforms compared on coverage, accuracy and refresh
How to evaluate a one-off file before paying
The corpus cuts behind the numbers on this page
Monthly loss rates by list type
How databases are built and where they decay
A maintained store of companies and the people at them, with a contact channel, firmographics and a touch history per record, fed by signups, exports and enrichment and kept usable by verification on entry and a re-check before every send and at least monthly. The share of records you can still reach is its real size; the row count is not.
Non-freemail (business and ISP) addresses verified in 2026 ran 14.5% invalid, 2.3% catch-all and 27% unconfirmed, with 32.6% confirmed valid. Vendor exports run 10 to 25% invalid on the day they land. A database with verification on entry and a re-check before every send holds 70% or more confirmed valid.
Before every send, and at least monthly, for records in active use, and before any record older than that enters a sequence. B2B addresses decay fast, with only 19.2% of addresses still valid 90 days after verification in the BounceZero decay study, so a re-check before every send catches most job changes before they turn into bounces that damage the sending domain.
Both, for different jobs. Subscription databases find new accounts by filter; your own base holds consent, history and the verified-at date that no vendor supplies. Buy for discovery, verify every export, and keep the system of record in-house.
The mailbox could not be confirmed either way, usually because the receiving gateway refuses verification probes. Non-freemail (business and ISP) domains return unknown 27% of the time. Keep those records out of cold volume, route them to phone or LinkedIn, and re-check them later; never count them as invalid.
Upload the database export, get valid, invalid, catch-all and unknown per record, and set the verified-at date on all of them. 100 free checks a month, bulk jobs to a million rows, unknowns refunded.
Ayoub built BounceZero's 5-stage validation pipeline, its dedicated BGP-announced IP infrastructure, and the Patroni HA PostgreSQL cluster behind every verification. Previously built high-volume email delivery infrastructure. Trained at 1337 Benguerir (École 42 network, 2019). Open-source: bgp_analyzer.
Where lists come from, what bounces, and what to check before you pay
Real estate email address lists in 2026
Buying an email list for lead generation in 2026
Seven B2B contact database providers compared on coverage by region, email accuracy at export, refresh...
Email list websites for lead generation ranked for 2026
Sellers compared, with sample-check rules
How databases are built and where they decay
Continue through related topics