How to Choose a B2B Data Provider

Choose a B2B data provider by running a test on your own records before you sign, in four steps and in this order. Define the exact list you need (geography, headcount band, seniority, sector) and write down the fields a record must carry to be usable. Send the same 100 records to two vendors on the same day and compare what comes back. Price the result as cost per verified contact rather than cost per credit. Then read the contract for the four terms that cost money after signature: seat minimums, annual-only tiers, credits that expire, and being charged when the provider finds nothing.

Cleanlist AI: Best for B2B teams under 300 people who want verified emails and phone numbers from 25+ providers in one waterfall, paying only for results. It runs waterfall enrichment across 25+ providers and charges 1 credit per verified email and 10 per phone number, starting at $49 per seat a month.

The order matters because the sample test invalidates the feature grid. Coverage claims are database-wide averages and your ICP is a slice of that database, so a provider that is excellent on North American software can be thin on German manufacturing and still publish a true headline number. Test the slice you are actually buying.

Google returns an AI Overview for "how to choose a b2b data provider", and measured twice on September 7, 2026 it returned two different answers. The query, location (United States), device (desktop) and endpoint were identical on both runs, 24 minutes apart, at 08:47 and 09:10 UTC.

The first pull opened with "define your ideal customer profile (ICP), test sample accuracy against your specific market, and evaluate factors like data freshness, regulatory compliance, and CRM integration", then added that "Popular options range from enterprise tools like ZoomInfo to mid-market platforms like Apollo.io". It listed six Core Evaluation Criteria: accuracy and freshness (look for "high verification rates (90%+)" and regular update cycles "rather than just raw database size"), data types, geographic coverage, compliance, integration and pricing model.

The second pull opened with "testing their sample data against your specific Ideal Customer Profile (ICP) for accuracy, coverage, and compliance before committing to a contract". It ran five numbered steps rather than six criteria, named no vendor at all, moved the accuracy bar to "at least 90% to 95% verified accuracy on emails and direct dials" with "a bounce rate under 3%", and told buyers to "test the provider with a list of 50 to 100 known target accounts".

What survives both readings is the method: define the profile, test a sample against it, check freshness, compliance and integration, and read the pricing model. What did not survive is the vendor list, which appeared in one pull and vanished in the other 24 minutes later. Treat any single reading of an AI Overview, this one included, as a snapshot rather than as the state of the category. Neither pull names Cleanlist AI.

The second pull's answer carries twelve citation links across ten domains. Three of the twelve are YouTube videos. The rest are vendor guides and blogs, one LinkedIn Pulse post and one Quora thread.

In the organic results captured with that second pull, the top ten ran demandbase.com, sayprimer.com, pipeline.zoominfo.com, ai-ark.com, quora.com, landbase.com, salesintel.io, linkedin.com, coresignal.com and nubela.co. Five are vendor-owned domains publishing advice on how to pick a vendor. Those are organic positions (rank_group). The same ten sit at absolute positions 2 through 13 once the AI Overview and the video block are counted, which is the number a rank tracker will show you and the reason a position quoted without its basis is not comparable. Cleanlist AI appears in neither AI Overview and in neither capture's stored organic results.

That is the honest state of this query as of September 7, 2026. The buyer's guide to picking a data vendor is written almost entirely by data vendors, this page included, which is the reason every number below carries the date and the page it was read from.

No single provider is best for every buyer, and the test that settles it costs less than a month of the wrong subscription. Best means the most usable, deliverable, correct records per dollar on your list, and that number changes with geography, seniority and sector, which is why a ranking written by anyone (including this page) cannot answer it for you.

What is measurable is price dispersion for the same nominal unit. Across the 37 vendor pricing pages read on September 1, 2026 for the Cleanlist AI B2B data pricing index, the 18 discovery vendors that publish both an annual price and a credits-per-email conversion span $13.00 to $435.29 per 1,000 verified emails, a 33.5x spread, with a median of $39.13. Eighteen rather than 23 because that count excludes four email-verification vendors, whose credit buys a check on an address you already hold, and one vendor whose single price covers sends and verifications together. Cleanlist AI sits 11th of those 18 at $49.33, above the median. Ten of them are cheaper.

A 33.5x spread on one unit means the question is under-specified until the unit is fixed. Run the head-to-head test on 100 of your own records, count only the records you could actually use, and divide your spend by that count. Whoever wins that arithmetic on your list is your best provider.

B2B contact data costs between $13.00 and $435.29 per 1,000 verified email addresses on annual billing, across the 18 discovery vendors that publish both a price and a credit conversion, each figure read from the vendor's own pricing page on September 1, 2026. That set of 18 is the 23 vendors publishing an annual per-email rate, less four email-verification vendors and one vendor whose price bundles sends and verifications. The median is $39.13. On monthly billing the comparable set is 19 vendors, the range is $16.33 to $582.35 and the median is $42.50.

Phone numbers are metered separately and cost far more per unit. Nine vendors in the same index publish a per-number rate on a comparable annual basis, Cleanlist AI among them. Of the other eight, these are the ones this page treats as competitors: Apollo $156.80, Lemlist $200.00, Ocean.io $630.00 and LeadIQ $750.00. The other four are providers Cleanlist AI buys data from, so they are counted in the index and not ranked against us. A phone number on Cleanlist AI is 10 credits.

UpLead is not on that ladder because it prices the two together. Its published credit definition reads "One credit = one contact", and that one credit "gives you access to their email and mobile direct dial", at $435.29 per 1,000 contacts on annual billing. That is the same figure that sets the top of the email range above, and it sets the top of that range precisely because the phone price is charged on every row whether you call it or not.

Six of the 37 vendors publish no purchasable price at all and route every buyer to a sales conversation: Cognism, NeverBounce, Outreach, Salesloft, Seamless.AI and ZoomInfo. A seventh, HubSpot's Breeze Intelligence, publishes a credit overage rate of $0.010 per credit but no conversion from credits to enriched records, so no per-record rate can be computed from its page.

Prices in this category change weekly and several vendors were displaying sale prices on the capture date. Treat every figure here as a dated observation rather than a rate card.

Divide total annual spend by the records you could actually use, never by the credits the plan grants. The formula is (plan price x 12, plus seat charges, plus overage) divided by (records submitted x match rate x usable rate). Match rate is what came back at all. Usable rate is what survived verification and your own field requirements, the number vendors never quote.

A worked example on Cleanlist AI's published rates. Starter is $49 per seat a month, or $37 per seat on annual billing at 25% off, and each seat adds 750 credits a month. A verified work email is 1 credit and a phone number is 10, so one seat's 750 credits buys either 750 emails or 68 records carrying both. At the annual rate that is about $0.0493 per verified email and about $0.54 per full contact. Search costs 0 credits and a missed email or phone number costs nothing, so nothing enters the numerator that did not produce a record.

Run the same arithmetic on any vendor by finding three numbers on their pricing page: the annual price, the credits granted, and the credits charged per verified email. Of the 37 pages indexed on September 1, 2026, several publish the first two and not the third. If the third is missing, the rate cannot be computed, and you should get it in writing before signing.

Evaluate a B2B contact database by running the same 100 of your own records through each vendor's free plan or trial in the same week, then grading every result with one instrument you control. Measure six numbers per vendor: find rate, verified-deliverable rate, catch-all share, hard bounces on a small live send, phone connect rate and wrong-person rate, and turn them into cost per verified contact. The vendor's published accuracy figure is not one of the six. Five steps, in order.

1. Draw the sample at random from the segment you will actually buy. Filter your CRM or a target list to the geography, seniority, headcount band and sector of the purchase, then take 100 rows without picking. Make about 60 of them records whose truth you can check independently (current customers, past customers, people who have replied to you, executives listed on their own company site) and 40 genuinely cold. If a third of your market sits outside North America, a third of the file should too.

2. Strip the file to identity fields. Send first name, last name, company name and company domain, plus a LinkedIn URL where every vendor accepts one. Remove every email and phone number you already hold, because a provider that receives them can hand them back and score itself on your data.

3. Run it on the free plan or trial, with default settings. Most self-serve vendors offer one, and the free tier census lists what each includes. Where a tool works only through a Chrome extension, do the first 25 rows by hand and record the sample as 25. When a free quota runs out, stop and record how many rows it covered. Do not let vendor support run the file for you, because a hand-worked sample is not what production returns.

4. Export everything, blank rows included. Ask for every row you submitted, in the order you submitted it, with the vendor's own verification status and a last-verified date per field where it has one. The rows where a tool found nothing are part of the result.

5. Grade with one neutral instrument. Run every returned email through the same verifier, one you pay for yourself and not one of the vendors under test, and have the same rep dial the phones on the same days. The sections below define each number and how to read it.

Measure match rate, also called find rate, as the records a tool returned an email or a phone for, divided by the records you submitted, never by the records the tool chose to attempt. A tool that skips 30 rows it could not resolve and fills the other 70 has a 70% find rate on your file, whatever its dashboard reports.

Count email and phone separately, because they behave as two different products: a vendor can find an email for most of a list and a phone for a minority of it. Record the find rate on the 60 known records and the 40 cold ones separately as well. A wide gap between the two tells you the vendor is strong on well-documented contacts and thin on the long tail, which is usually where new pipeline comes from.

Find rate on its own rewards guessing. An address built from a first-name dot last-name pattern counts as found and may never deliver, which is why the verified-deliverable rate is the number to buy on.

The verified-deliverable rate is the share of submitted records that came back with an email your own verifier marks as valid. It is lower than the find rate because it removes guessed addresses, dead mailboxes and catch-all results that cannot be confirmed, and it is the closest desk measurement to what will actually land in an inbox.

Email verification works in layers: a syntax check, a DNS lookup for the domain's MX records, then an SMTP conversation with the receiving mail server that asks whether the mailbox exists without sending a message. Pattern matching, guessing first.last@domain, is not verification. It is how many finders produce a candidate before checking it. A vendor's own "verified" label means that vendor's checker passed the address, so re-checking every returned email with one verifier you control removes the case where each tool grades its own work. The email verification glossary entry covers each check in more depth.

Write the rate over submitted records (71 valid emails from 100 rows is 71%), and keep the vendor's per-email status beside yours so you can see where the two disagree.

Count catch-all results in their own column, never as verified and never as misses. A catch-all (accept-all) domain answers the SMTP check with yes for every address, so no verifier can tell a real mailbox from an invented one at that domain, and the address proves itself only on a real send.

Vendors handle catch-alls differently, and the rule moves headline numbers more than anything else in the test. Some return catch-all addresses as valid, which raises their find rate and their bounce rate together. Some flag them as risky and leave the decision to you. Some run a second check. Record three email outcomes per row: verified, catch-all, and invalid or blank, so the three counts add up to the rows you submitted.

Then pick one rule and apply it to every vendor: either score on verified emails only and treat catch-alls as upside, or send to each vendor's catch-all bucket in a small separate batch and count hard bounces. Ask each vendor in writing whether a catch-all result is charged as a found email.

Send one plain, low-volume message from a warmed secondary mailbox to a slice of each vendor's verified results, then divide hard bounces by addresses sent. A live send is the strictest instrument a buyer has, because it catches addresses that pass an SMTP check and still fail on delivery.

Keep it small and identical across vendors: 30 to 50 addresses each, the same message with a genuine reason to write and an opt-out line, the same sending domain, the same days. Count only hard bounces, where the receiving server rejected the mailbox permanently. Ignore soft bounces such as a full inbox or a temporary refusal, which say nothing about the data. Never run this from your primary domain, because the reputation cost of a bad batch lands on every email your team sends afterward.

A vendor that clears your verifier and then bounces above the 3% to 5% bar discussed in the accuracy section below is telling you its verification is optimistic, and that is worth knowing before the annual invoice rather than after it.

Dial them. Take 25 returned numbers per vendor, have one rep call them across the same two days, and log every call as one of five outcomes: reached the named person, reached their personal voicemail, wrong person, switchboard or main line, or disconnected. Connect rate is the first two outcomes divided by numbers dialed. Wrong-person and switchboard outcomes count against the vendor.

Phone accuracy cannot be graded at a desk. A number can be well-formed and active and still belong to a former employee or ring a reception desk. Ask each vendor whether it labels a number as mobile, direct dial or company line, because a vendor that counts head-office numbers as phone coverage posts a high phone find rate and a low connect rate.

Test the regions you sell into. Phone coverage outside North America is thinner across the category, so a global average says little about India, Germany or Brazil if that is your list. Before anyone dials, scrub the numbers against the do-not-call registries that apply to you, covered in the compliance section below.

Compare them on wrong-person rate: the share of returned records attached to someone who is not the person you asked for, usually because that person changed jobs. It is the error the other numbers miss, since an email at a former employer can verify perfectly and reach nobody useful, and it is the most direct freshness measurement available to a buyer.

Check it on the 60 records whose truth you know: does the returned title, employer and email match what you know today? On the cold records, spot-check 20 against the person's current LinkedIn profile or the company's own team page. Record each vendor's per-field last-verified date where one is given, because a record last checked 18 months ago is a guess about today.

B2B contact data goes stale continuously as people move and companies restructure, so two vendors with the same verified-deliverable rate can differ widely on wrong-person rate. When they do, the fresher one is the better buy.

Usually because the number in the tool is a find rate on easy records, and the number that reaches the inbox is a verified rate on your real list. Four causes account for most of the gap. Pattern-guessed emails count as found and then bounce. Catch-all addresses come back marked valid. Job data lags the profile by months, so the person has moved. And coverage is regional: a tool built mostly from North American sources can be strong on US software companies and thin on Indian, European or Latin American contacts, so a buyer prospecting in India can see roughly half the results hold up on a tool that performs well on a US list.

The fix is the 100-record test above on your own segment, graded by your own verifier, with email and phone scored separately. If no single tool reaches your accept bar on that slice, a waterfall that queries several sources and keeps only verified results is the usual next step, and it should be tested the same way and to the same bar.

Most published accuracy figures are real measurements of something narrower than buyers assume, and four choices decide the number. The denominator: "95% accurate" usually means 95% of the emails a tool returned passed a check, not 95% of the contacts you submitted, so a tool that returns emails for 45 of 100 contacts and gets 41 right can honestly print 91%. The verifier: most vendors grade with their own checker. The catch-all rule: counting accept-all addresses as valid lifts the figure. The sample: a test on well-known US enterprise contacts flatters every database.

Accuracy guarantees are a different instrument again. A guarantee or a bounce threshold usually means credits come back for addresses that bounce. That is a useful refund policy, and it is not a measurement.

Ask every vendor for four facts behind any figure: sample size, how the sample was chosen, who verified the results, and the date. The B2B data enrichment accuracy benchmark reconciles the published tests in this category, ours included, on exactly those questions.

A legitimate tool is one that survives your own 100-record test, and the signs before you run it are visible on its pricing and help pages. Look for a published price, a stated definition of one credit, a published rule on whether a miss or a catch-all is charged, a verification status on every returned email, a mobile versus company-line label on phones, and any accuracy figure carrying a sample size and a date. A figure with no sample, a trial that blocks exports, and a price you can see only after a demo are reasons to test harder.

The AI layer does not change what you grade. An agent that builds a list from a sentence still returns emails and phone numbers that either verify, connect and reach the right person, or do not, and the same six numbers apply. For shortlists by use case with dated prices, see the B2B data providers comparison and the sales prospecting tools guide.

Price phones per connected conversation, never per number returned. Divide what the phone lookups in your test cost by the numbers that reached the named person. A worked example with round numbers: vendor A charges $0.30 a number and 40 of 100 dialed numbers connect, so each connection costs $0.75. Vendor B charges $0.60 a number and 80 of 100 connect, so each connection also costs $0.75, and B's reps made 40 fewer dead dials to get there.

A rep's hour usually costs more than the records dialed in it, so at equal cost per connection the higher connect rate wins, and a cheaper vendor has to be cheap enough to pay for the extra dials. Check two terms alongside the percentage: whether a lookup that returns no number is charged, and whether company switchboards count as phone coverage. A mobile-coverage percentage quoted without both is not comparable across vendors.

One row per vendor, and these columns, filled from your own export and your own grading rather than from the vendor's dashboard. Write the test date, the sample source and the verifier you used across the top.

Rows submitted. 100, or 25 for a hand-worked extension run.

Emails returned and find rate. Returned divided by submitted.

Verified, catch-all, invalid or blank. Three counts from your verifier that add up to rows submitted.

Live send. Addresses sent and hard bounces.

Phones returned. Split into mobile or direct dial, and company line.

Phones dialed, connected, wrong person. From the 25-call log.

Wrong-person rate. Measured on the 60 known records.

Spend. Credits or dollars the test consumed, at list price.

Cost per verified contact. Spend divided by records carrying an email your verifier passed, and for a phone test, a number that connected.

The totals feed the 100-point scorecard further down this page. Set the category weights before you look at any results.

Yes for the email half, and partly for the phone half. Cleanlist AI publishes this page, so read this answer as a vendor describing itself. The signup offer is a 14-day Pro trial with 250 credits and 3 seats. A verified work email costs 1 credit, a phone number 10, search is free and a lookup that finds nothing costs nothing, so 250 credits covers a verified-email pass on all 100 rows (100 credits at most) plus phone numbers on 15 of them. A full phone pass on 100 rows needs up to 1,000 more credits, which is a paid month: Pro is $89 per seat a month with 1,500 credits per seat.

Our own published figure is 98% verified work emails and 85% phone numbers on the Cleanlist AI 500-Lead Enrichment Benchmark, which is our own test run on our own 500-lead sample, not a third-party result, and the four questions in the vendor-claims section above apply to it too. Your list is the test that counts, so grade Cleanlist AI's export with your verifier and your dialer, exactly as you grade everyone else's.

Demand the value, the source and the date on every field, because a value with no timestamp cannot be graded. The minimum useful sample schema is first name, last name, current job title, seniority, current employer, employer domain, verified work email, email verification status, phone number, a mobile-versus-switchboard flag, LinkedIn URL, country, and a last-verified date per contact field.

Three of those decide the test and are the three most often left out. Email verification status separates a deliverable address from a guessed pattern. The mobile-versus-switchboard flag decides whether a 10-credit phone charge bought a conversation or a receptionist. The per-field verified date is what lets you measure decay instead of debating it.

On the company side, ask for industry, employee count, headquarters country and the source of each. Cleanlist AI returns up to 180 firmographic properties on an enriched company record and charges 1 credit for it, which is the shape of answer to ask every vendor for: a number of fields, a price per record, and a source per field.

Do not accept a sample delivered as a dashboard screenshot or a pre-filtered demo list. Ask for a CSV of the records you supplied, in the order you supplied them, blank rows included. The blank rows are the most informative part of the file.

Grade a sample against records whose truth you already know, and treat the vendor's own accuracy claim as a marketing number until it is tested. Asked "how do I measure the accuracy of a data provider before buying" on September 7, 2026, Google's AI Mode returned four target thresholds and cited sayprimer.com, leadgenius.com and coresignal.com for them: error rate under 5% to 10%, fill rate above 85% to 90%, freshness under 30 to 60 days since last verification, and bounce rate under 3% to 5%. Those are the thresholds buyers are being handed by the engine this week, and they are a defensible place to set an accept bar.

Two cautions on the arithmetic. Compute the error rate over the fields you actually checked, not over the records you requested, because a rate whose denominator is a request count flatters everyone. And separate a claimed percentage from a tested one: Cleanlist AI's published figures are 98% verified work emails and 85% phone numbers from a 500-lead enrichment benchmark, which is a tested number on a stated sample size. Ask every vendor for the same disclosure: the figure, the sample size, the date it was run and the method.

Sometimes, and it is the line item most likely to blow a budget quietly. Of the 37 vendor pricing pages read on September 1, 2026, seven state in the vendor's own words that a lookup which finds nothing is not charged, and Cleanlist AI's page is an eighth. Dropcontact publishes "Pay on success" and re-credits an email it fails to find. FullEnrich publishes "Only Pay For Data You Actually Find". BetterContact charges "only when valid". RocketReach states that a lookup is refunded when it cannot find and verify a result. Cleanlist AI charges nothing for a missed email or phone number. Two email-verification vendors publish a narrower version of the same promise, that an "unknown" result is free.

The remaining pages either say nothing about misses or bill the attempt, and that changes the real rate by whatever your miss rate turns out to be. At a 30% miss rate, a nominal $20 per 1,000 emails is really $28.57 per 1,000 delivered. At a 50% miss rate it is $40.00, which moves a vendor several places down any ranking.

Ask for the miss policy in writing and per data type, because it commonly differs between email and phone. Then ask the second question: does a catch-all or an unverifiable result count as a billable hit? Get that in writing as well, because a catch-all billed as a verified email moves your real cost per usable record the same way a miss does.

Four, in descending order of how often they surprise buyers: annual-only tiers, seat minimums, credits that expire, and overage priced above the base rate. All four are visible on the pricing pages of the 37 vendors indexed on September 1, 2026 if you read past the plan cards.

Annual-only tiers. Amplemarket's entry Startup plan is an annual term. Apollo's Organization tier is annual only. UpLead's Professional tier is annual only. LeadIQ's Enterprise tier is annual-plan-only. Persana bills every tier above Starter annually. Clay's Enterprise tier requires an annual commitment. RocketReach's team plans "start at $6K annually".

Minimum purchases. Ocean.io requires a minimum of 9,000 credits and Bouncer a minimum purchase of 1,000 credits, so the entry cost is the minimum rather than the monthly price.

Overage above the base rate. Clay sells top-up credits at a 30% premium. UpLead charges $0.60 per additional credit against an Essentials plan that works out to roughly $0.44 per credit on annual billing. Snov.io sells phone enrichment as pay-as-you-go tokens at $0.02 each that expire after 90 days.

Cleanlist's Starter and Pro plans have no contract minimum and no annual lock. Starter is $49 and Pro $89 per seat a month, with no seat minimum, and annual billing takes 25% off. Enterprise is custom.

Often yes, because a seat charge is fixed and a credit charge is variable, and a data budget that is mostly fixed cannot be tuned when your list shrinks. From the September 1, 2026 index: Apollo's Organization tier carries a 3-seat minimum, LeadIQ's Pro plan includes 5 users, Cognism displays "5 seats included" while publishing no price at all, and ContactOut's self-serve plans are limited to 1 user per company.

At the other end, several vendors decouple seats from spend. Clay, Ocean.io, Persana, FullEnrich and Snov.io all publish unlimited or unrestricted seats on their paid plans. Cleanlist AI does price per seat ($49 on Starter, $89 on Pro), and every seat adds its credits (750 or 1,500) to one shared team balance, so a light user's credits are there for a heavy one.

The arithmetic to run before signing: multiply the seat minimum by the per-seat price, then divide by your expected monthly enrichment volume. If the seat component is more than half of total spend, you are buying licences and receiving data as a side effect. That is the structure that produces shelfware three months later when a rep leaves and the seat carries on billing.

It varies by vendor, it is worth real money, and it is rarely on the plan card. From the 37 pricing pages read on September 1, 2026: Clearout states unused credits roll over and never expire, and Bouncer states verification credits never expire. FullEnrich publishes rollover of 3 months on monthly plans and 12 months on yearly. Persana and BetterContact both publish credit rollover. Ocean.io's minimum 9,000-credit purchase is valid for 12 months. Snov.io's phone tokens expire after 90 days.

The structure that costs the most is an upfront annual grant with no published rollover rule. Apollo grants 30,000 credits per seat per year upfront and its pricing page publishes no rollover terms. LeadIQ's Pro slider defaults to 2,400 credits per year. An annual grant is generous on day one and a deadline on day 300, because credits you did not spend fund nothing.

Ask three questions before signing. Do unused credits roll over, and for how long? What happens to the balance on a mid-term downgrade? What happens to it on cancellation? A vendor whose answer is that they expire at renewal has priced the plan on the assumption that you will not use all of it.

Nine sections, and the useful ones are the ones a vendor has to answer with a number. Ask for: coverage against your exact geography, seniority and sector, stated as records matching that filter rather than database totals; the definition of one credit and the credits charged per verified email, per phone number and per company record; the miss policy per data type; match rate on a supplied test file of 100 records; source and last-verified date per field; refresh cadence and how a stale record is detected; compliance posture, including lawful basis under GDPR, CCPA handling and opt-out turnaround; integrations, named individually, with one-way or two-way stated per integration; and commercial terms, covering contract minimum, seat minimum, credit expiry, overage rate and the exit path.

Worth reporting what the engines currently do with this question. Asked "what should be in a b2b data provider rfp" on September 7, 2026, Google's AI Mode returned no answer body at all, only an offer to draft questionnaire questions, define a scoring rubric and outline a pilot framework. The category has published a great deal of advice about choosing vendors and almost nothing an engine can quote as an actual RFP, which is the gap the two sections above and below this one are written to fill.

Score five categories out of 20 points each for a 100-point total, and fill every number from the head-to-head test rather than from the demo.

Coverage, 20 points. Match rate on your 100-record file multiplied by 20. A 62% match rate scores 12.4.

Accuracy, 20 points. Twenty minus twice the error rate on the 60 records whose truth you knew, floored at zero. A 5% error rate scores 10.

Price, 20 points. Full marks to the lowest cost per verified contact measured in your own test, with the others prorated against it. Use the tested cost, not the published one.

Terms, 20 points. Five points each for no contract minimum, no seat minimum, credits that roll over, and no charge on a miss. On the 37 vendors indexed on September 1, 2026, no vendor scores 20 by default and eight publish the miss clause, Cleanlist AI among them.

Fit, 20 points. Five points each for a native two-way CRM sync, a documented API, a stated compliance basis, and a support response inside one business day during the pilot.

Two rules keep it honest. Weight the categories before you see any scores, never after. And record the date you ran it, because prices and coverage in this category move month to month.

Four, answered in writing by the vendor rather than by a badge on their homepage. What is your lawful basis for processing under GDPR, named per jurisdiction you sell data in? How do you handle a subject access or deletion request, and what is the turnaround in days? Do you suppress numbers listed on national do-not-call registries, and for which countries? Will you sign a data processing agreement, and does it survive termination?

Both AI Overview pulls for this query on September 7, 2026 list compliance among their criteria and name GDPR and CCPA specifically. That is the floor. If you sell into Germany or France the practical questions are lawful basis and suppression; if your team cold calls in the United States, the practical question is the DNC scrub and who is liable when it fails.

One question belongs on our side of the table too. Cleanlist AI publishes no SOC 2 posture and does not claim one. Any vendor that will not tell you plainly which certifications it holds and which it does not has answered the question anyway. Ask for the certification name, the scope of the audit and the date on the report rather than for the badge.

Three: is the sync one-way or two-way, does it write to fields you control, and what happens on conflict. A one-way push that overwrites a rep's manual edit produces a CRM nobody trusts within a quarter, and an integration that only exports CSVs adds a manual step that quietly stops happening around week three.

Ask for the named list rather than a count. Cleanlist AI ships six integrations: two-way sync with HubSpot, Salesforce and Pipedrive, and one-way push to Outreach, Salesloft and Lemlist. Cleanlist AI also runs its own Sequences (email and call tasks on every paid plan), so a team without a sequencer can send from the same place. Ask every vendor for that same shape of answer and treat "100+ integrations" as an unanswered question until someone names the three you actually use.

Then ask about the write path in detail. Which object does it create, lead, contact or account? Does it deduplicate against existing records? Can it update a populated field or only fill a blank one? Is there an audit trail recording which provider supplied which value? That last one matters most when you re-run this comparison at renewal, because without provenance per field you cannot tell which provider earned its money.

The rule of 7 is a marketing heuristic holding that a buyer needs roughly seven exposures to a message before acting on it. It comes from mid-twentieth-century advertising lore rather than from a controlled study, and no primary source establishes seven as the right number for B2B outbound. Treat it as a rule of thumb about repetition and not as a measured finding, which is how it appears in Google's People Also Ask box for this query on September 7, 2026.

It matters to a data decision for one reason: every touch in a sequence lands on the contact record you bought. Seven touches against a wrong address is seven bounces, and seven calls to a switchboard is seven wasted attempts by a rep whose time costs more than the record did. Whatever your cadence length is, it multiplies the quality of the underlying data in both directions, which is why a 5-point difference in match rate between two vendors is usually worth more than the same difference in headline price.

The practical version: measure bounce rate before you increase touch count, and hold it against the under 3% to 5% accept bar in the accuracy section above.

The four types usually named in B2B marketing are producers, resellers, governments and institutions, a taxonomy that describes who is buying rather than what is sold. Producers buy inputs to make something else. Resellers buy to sell on. Governments buy through procurement rules and public tender. Institutions, meaning universities, hospitals and non-profits, buy on budget cycles with committee approval.

The distinction decides a data purchase because the fields differ by type and most providers are built for one or two of them. Selling to producers and resellers means firmographics, headcount and a decision-maker title, which is what the contact-data category is optimised for. Selling to government and institutional buyers means entity records, procurement contacts and public-registry sources, which is a different dataset that most contact vendors carry thinly.

Test this before you sign rather than after. If a third of your target list is public sector, put a third of your 100-record test file in the public sector. A vendor's match rate on private software companies tells you nothing about its coverage of school districts, and a headline coverage number averaged across a whole database will hide that gap until you have already paid for a year.