How to build a B2B lead list.

Step 1: Define the ICP before opening any tool

The most common list-building failure happens before the first search: targeting “companies that could conceivably use this” instead of companies that demonstrably do. Start from evidence. Pull your last twenty closed-won deals and write down what they share — industry, headcount band, geography, the title that signed, the title that championed. If you’re pre-revenue, use your closest competitor’s public customers as a proxy. The output should be one or two sentences with no adjectives: “Logistics and freight companies, 50–500 employees, in DACH and Benelux, where the buyer is a Head of Operations or COO.”

Be equally explicit about exclusions — agencies, nonprofits, companies below a headcount floor, countries you can’t legally sell into. Exclusions are what keep a list tight when the filters get broad.

Step 2: Translate the ICP into firmographic filters

Now convert each clause of the ICP sentence into a database filter: industry, employee range, country or region, and a title or seniority pattern for the contact layer. Industry is where translations go wrong, because most databases are built on classification codes that don’t match how you think about a market. “Logistics” might be one label or a dozen adjacent ones — freight forwarding, warehousing, last-mile delivery, customs brokerage. Search the industry taxonomy by what you sell to, in plain words, and inspect a sample of matching companies before trusting any filter. Ten minutes of spot-checking company names saves a thousand mistargeted sends.

For titles, write patterns rather than exact strings — the same buyer is a “Head of Operations” at one company, a “VP Ops” at another, a “COO” at a third. Decide consciously whether you want decision-makers, champions, or both, and keep them in separate segments so the messaging can differ.

Step 3: Size the market — counts before credits

Before exporting anything, look at the count. Every decent platform will tell you how many companies and contacts match your filters without charging you for the answer, and that number is a sanity check on your entire strategy. If your “tight ICP” returns 400,000 contacts, your filters are vaguer than your ICP — tighten until the count looks like a market you could actually work. If it returns 800, you’ve discovered something more important than a list-building problem: your addressable market may not support your pipeline math, and it’s far better to learn that from a count than from a quarter of missed quota.

Counts are also how you compare data vendors honestly: run the same real-world filter set on each and see who actually covers your market, not who claims the biggest global number. (More on that in our provider comparison guide.)

Step 4: Prioritize segments, don’t export the universe

Resist the temptation to pull the whole matching set at once. Slice the market into segments — by industry vertical, by country, by headcount band — and rank them by expected fit. Then export the best segment first, at a size your sending infrastructure can actually consume in two or three weeks (for most teams, 500–2,000 contacts). This does three things: it keeps your cost matched to your sending capacity, it gives you a feedback loop (reply rates by segment tell you where to spend next), and it means a messaging mistake burns one segment instead of the whole market.

Step 5: Verification — the step that protects everything downstream

An unverified list converts bounce risk into sender-reputation damage, and reputation damage outlives any single campaign. The rule: no address enters your sequencer without a recent verification verdict — ideally a live SMTP probe, not a stamp the vendor applied months ago at collection. On Argorant this step is built into the export itself: every address is probed at the moment you download, invalid rows are filtered out and cost zero credits, and catch-all domains are a separate opt-in rather than silently mixed into “valid.” Because the check happens at export, the verdict always reflects the mailbox’s current state. The full methodology is documented in how we verify.

If you build lists from other sources, the principle stands: budget for verification as a mandatory line item, and treat any vendor verdict more than a couple of months old as expired.

Step 6: Enrich for personalization, not for hoarding

Enrichment means attaching context to a contact: company size, location, industry, the fields your copy will actually reference. The discipline is to enrich only what you’ll use. A first-touch email realistically uses four or five fields — name, title, company, industry, maybe city or a size band. Forty columns of unused firmographics don’t improve reply rates; they bloat your CRM and complicate your compliance posture, since under GDPR you should be able to explain why you hold each field. Export the fields your sequences reference, and let the database remain the system of record for the rest.

Step 7: Export to your sequencer or CRM

Map fields explicitly when you import — especially custom variables your sequences use, because a broken merge field (“Hi {{first_name}}”) is the fastest way to torch a segment. Dedupe against your CRM and your suppression list before anything goes live: existing customers, open opportunities, people who previously opted out. And keep provenance — tag every imported contact with its source segment and export date, so that when replies come in you know which hypothesis is working and when a contact’s verification verdict ages out.

Step 8: Refresh cadence — a list is perishable

B2B contact data decays at roughly 2–3% per month as people change roles, so a list is a perishable good with about a one-quarter shelf life. Build the rhythm in: re-verify anything you’re still actively sending to once its verdict is a couple of months old, re-run your saved searches quarterly to pick up companies that newly match your filters, and retire segments whose reply rates have decayed rather than squeezing them. The teams that treat list-building as a weekly habit — one segment in, one segment reviewed — consistently beat the teams that do a giant annual pull.

The modern variant: let an agent run the loop

Everything above is also expressible as a conversation. Argorant ships an MCP server and API on every paid plan, plus a CLI on npm (npx argorant), which means Claude, GPT, or your own agent can execute this exact workflow: describe the ICP in plain language, have the agent translate it to filters, check counts, compare segment sizes, preview samples, and only then trigger a verified export. The count-before-credits discipline becomes natural — the agent can iterate on filters for free and spend credits only on the final, sized, verified pull. If your team works in an AI-first stack, see how agents connect; if you’re pricing the workflow, plans start under $100/month on pricing.