B2B data providers, compared honestly.

Why headline numbers mislead

The category sorts itself by total record count: ZoomInfo and Apollo each advertise databases in the hundreds of millions, Cognism leads with European depth, Lusha and Hunter with simplicity, and Argorant’s own number is 614M contacts across 184 countries. None of these figures, including ours, answers the question you’re actually asking — which is whether the database covers the eight thousand companies you sell to, with mailboxes that still exist. A vendor can lead a global count and be thin in mid-market German manufacturing; a smaller database can be dense exactly where you operate. Total size tells you about the vendor’s collection ambitions, not your hit rate.

So the first rule of evaluation: ignore the brochure and test coverage against your real market. The rest of this guide is the checklist for doing that, in the order the questions matter.

1. Coverage against your actual ICP

Take your genuine ICP filters — industry, headcount, geography, title — and run the identical search in every platform you’re evaluating, on a trial or free tier, before any contract discussion. Record three numbers: matching companies, matching contacts with emails, and (by sampling) how many of those companies are really in your segment versus misclassified. Then check ten accounts you know well — current customers, dream logos — and see whether the right people show up at each. An afternoon of this produces a coverage table no analyst report can give you, because no analyst has your ICP. Geography deserves special attention: providers differ far more outside the US than inside it, and EU/DACH/APAC depth varies enormously between vendors that look identical on a US-only test.

2. Verification methodology — and when it runs

“95%+ accuracy” appears on nearly every site in the category and means nothing without two qualifiers: how was the verdict produced, and when. Methodologies range from pattern inference (guessing first.last@) through MX-level checks to live per-mailbox SMTP probes — each layer categorically stronger than the last. Timing matters just as much: a verdict applied when the record entered the database decays at 2–3% per month in inventory, while a verdict produced at the moment of export is as fresh as it can physically be. Ask every vendor: do you SMTP-probe per mailbox? At collection or at export? How are catch-all domains labeled — as their own category, or folded into “valid” to flatter the accuracy stat? A vendor that can’t answer per-row “when was this verified?” is selling you the past. Our own answers are documented publicly in how we verify, and the underlying mechanics are in our verification guide.

3. Pricing motion: self-serve versus quote-led

The buying experience predicts the contract. Self-serve vendors (Apollo, Hunter, Lusha, Argorant) publish prices, let you start without talking to anyone, and bill monthly. Quote-led vendors (ZoomInfo, Cognism, and most enterprise platforms) route you through sales, price by negotiation, and typically require annual commitments — frequently five figures before the platform has proven anything against your ICP. Neither motion is wrong; they optimize for different buyers. But be honest about which one fits your stage: a quote-led annual contract is a bet placed before the coverage test, while self-serve lets the product fail fast on its own merits. If a vendor won’t show pricing without a call, price that opacity into your evaluation.

4. The credit fine print: seats, exports, expiry

Two plans with the same sticker can differ several-fold in effective cost once you read the metering. The questions that matter: Are credits pooled across the team or trapped per seat? Do unused credits roll over or expire monthly? Is a credit charged for viewing a contact, or only for exporting it? Are bulk CSV exports on your plan, or gated to a higher tier (a common ceiling on entry-level plans)? And — decisive if you read section 2 — do you pay for records that turn out to be invalid? Most per-record pricing charges the same for a live mailbox and a dead one, which means the vendor monetizes its own decay. The alternative model, which Argorant uses, is pay-per-valid: invalid rows are filtered at export and cost zero credits, so the freshness risk sits with the seller.

5. API and agent access

Increasingly, the entity running searches isn’t a person with a browser — it’s a workflow, a CRM sync, or an AI agent. That makes programmatic access a first-class criterion. Check whether the API is included or an enterprise upsell, what the rate limits are at your tier, and whether the full workflow (search, count, preview, export) is exposed or just lookup endpoints. If your team uses Claude, GPT, or custom agents, ask specifically about MCP support — it’s the difference between an agent that operates the database natively and one that scrapes a UI. Argorant ships MCP and API on every paid plan, plus a CLI on npm (npx argorant); for most legacy vendors, equivalent access starts at enterprise tiers. See AI agent access for what that looks like in practice.

6. Contract terms and exit cost

Read for the exit before you sign the entrance. Annual commitment or monthly? Auto-renewal with a narrow cancellation window? Seat-count minimums? What happens to exported data if you leave — most vendors license rather than sell records, and the license terms on departure vary widely. Also check compliance posture (GDPR handling, a signable DPA, a documented opt-out process for data subjects), because your outbound program inherits your data vendor’s legal standing. None of this shows up in a demo, all of it shows up in year two.

The metric that decides it: cost per usable contact

Collapse everything above into one number. Take a real segment from your ICP, export it (or its equivalent) from each finalist, verify the output independently if the vendor didn’t verify at export, and divide the true monthly cost by the count of contacts that are in-segment and deliverable. That’s cost per usable contact, and it routinely reorders the field: a platform with a low sticker and 30% dead rows is more expensive than one that charges more per credit but only for valid records. It also converts every vendor claim into something testable on a few hundred rows.

Where does Argorant sit in this framework? Deliberately narrow. We don’t ship intent data, a built-in dialer, or org-chart visualizations — vendors like ZoomInfo, Apollo, and Cognism bundle far more workflow. What we do is the contact layer itself, done strictly: 614M contacts in 184 countries, every export SMTP-probed live at download, catch-alls opt-in, invalid rows free, plans from under $100/month with API, MCP, and CLI access on every paid plan. If you need the full platform bundle, buy the bundle. If you need verified contacts at a testable cost per usable row, run us through this exact framework — the side-by-side breakdowns live in compare and alternatives, and the numbers are on pricing.