The B2B Database Building Guide: from ICP to verified list

A B2B database is built in six stages: define the ideal client profile, source companies that match it, identify the decision-makers inside them, enrich each contact with a role and an email address, verify every address before it is ever mailed, and load the result into your sending tool. The list is the foundation of the entire outbound system — good copy cannot save bad targeting, and no amount of clever sequencing rescues a database of the wrong people. This guide covers each stage in order, plus the questions that sit around them: where UK data comes from, whether to buy or build, how many prospects you need, and how quickly the data rots.

If you want the broader context — how the database fits alongside copy, sending infrastructure and follow-up — the UK B2B outbound playbook covers the full system. This guide goes deep on the data layer alone.

Why is the list the foundation of outbound?

Because every other part of the system multiplies it. Copy, deliverability, follow-up — each one is a percentage applied to the list underneath. When the list is wrong, every percentage is applied to the wrong people, and the output is zero regardless of how well the rest performs.

The failure is easy to spot in the numbers. A well-run campaign to a well-built list should see somewhere around a 4% positive reply rate. A campaign to a bad list produces bounces, spam complaints and silence — and the bounces then damage your sender reputation, which drags down every future campaign too. Bad data does not just waste one send; it taxes the next one. The cold email deliverability guide covers that mechanism in detail.

So the order of operations matters. List first, copy second, volume last. Most firms do it backwards.

How do you define an ICP precisely?

An ideal client profile has three components, and all three must be named specifically:

• Industry — not "professional services" but a sub-vertical: commercial insurance brokers, structural engineering consultancies, managed IT providers.
• Size — a headcount or revenue band. A 10-person firm and a 200-person firm buy differently, even in the same industry.
• Role — the person who owns the number your service moves. Sell to whoever owns the number. If you reduce cost-per-hire, that is whoever is measured on cost-per-hire, not "HR" in general.

If you cannot fill in those three fields without hedging, stop before sourcing a single company. I have written separately about why naming your buyer precisely is the single disqualifying question for outbound — it is the most common point of failure we see, and it happens before any tool is opened.

Where does UK B2B data actually come from?

There is no single clean source, which is why building is a pipeline rather than an export. In practice UK B2B data is assembled from several layers:

• Companies House — the public register. Good for legal entities, filing history, incorporation dates and rough size signals; useless for contact details.
• Commercial data platforms — aggregators that index companies and people at scale. Coverage of UK SMEs is patchier than their marketing suggests, and role data is frequently stale.
• LinkedIn and company websites — the most current source of who actually holds which role, but unstructured and manual to extract at scale.
• Trade bodies, directories and awards lists — narrow but high-precision sources for specific sub-verticals.

No single layer is sufficient. A usable database cross-references them: the register confirms the entity exists and its size, the platforms provide candidate contacts, and the live web confirms the person still holds the role.

What does the build pipeline look like, step by step?

This is the mechanism we run for every database build. When a client gives us an ICP, then the pipeline executes in this order:

1. ICP criteria in. Industry, size band, geography, role — written down as filters, not vibes.
2. Source companies. Pull every company matching the filters from the layered sources above, then deduplicate.
3. Identify decision-makers. For each company, find the named person who owns the relevant number. One or two contacts per company, not ten.
4. Enrich contact data. Attach role, seniority and a work email address to each person.
5. Verify every address. Each email is checked before it is ever sent to: hard bounces removed, catch-all domains flagged, risk scored. Nothing skips this step — why verification comes before sending is a rule, not a preference.
6. Load. The surviving records go into the sending tool, segmented by sub-vertical, ready for sequencing.

Records fall out at every stage. Companies that matched the filter turn out to be dormant; decision-makers cannot be identified; addresses fail verification. That attrition is the point — every record removed is a bounce or a wasted send avoided.

Should you build with AI or by hand?

Both, in different places. AI-assisted building has changed the economics of steps 2–4: an agent can cross-reference sources, scan hundreds of company pages and identify likely decision-makers at a scale no researcher can match, and it does not get bored on record 400. I have documented exactly how our AI list-building agent works as a build log.

What AI does not do well is judgement. Defining the ICP, deciding whether an edge-case company is genuinely a fit, choosing which of two plausible contacts owns the number — those calls still need a human who understands the client's business. The practical division of labour: humans define and spot-check, machines source and enrich, and verification tools gate what gets loaded.

Bought lists vs built lists — what is the difference?

A bought list is a snapshot someone else took, of an audience someone else defined, at a time you cannot verify, sold to you and typically to others as well. A built list is assembled against your ICP, this month, verified before load.

The problems with bought lists are structural, not incidental:

• The selection criteria are generic, so the targeting is loose — and loose targeting reads as spam.
• The data was current when compiled, not when you send.
• The same addresses have commonly been mailed by other buyers, so recipients and spam filters have seen the pattern before.

Bought lists appear cheaper per record. They are more expensive per meeting, which is the only unit that matters.

How many prospects do you need?

Fewer than most firms assume, because the constraint is sending capacity, not list size. A healthy cold inbox sends 25–40 emails per day, and each prospect receives a sequence of 4 emails over 14 days. Work that backwards: one inbox processes roughly 200–350 new prospects per month at sustainable volume.

So a database of 1,000–2,000 well-verified prospects supports several months of campaigning on a small inbox setup. A database of 20,000 loose records supports nothing except a deliverability problem. Depth of fit beats breadth of coverage every time.

How quickly does the data decay?

Continuously. People change jobs, companies merge, domains lapse. Industry estimates on B2B data decay vary, but it is commonly put at a rate that makes a list materially degraded within a year — which matches what we see in practice.

The operational answer is a quarterly refresh: re-verify every address, re-check that key contacts still hold their roles, remove departures and add new entrants that now match the ICP. A database is not an asset you buy once; it is an asset you maintain. Treat it like the pipeline infrastructure it is.

How do you know a finished database is good?

Acceptance criteria, checked before anything is sent. A database is ready to campaign on when:

• Every record matches the written ICP. Spot-check a random sample of 30–50 records against the original filters. If more than a handful fail, the sourcing stage was loose and the sample is telling you the whole list is.
• Every address passed verification. Not "most" — every one. Catch-all domains are held in a separate segment with lower sending priority, not mixed into the main list.
• Each company carries one or two named contacts, each with a confirmed current role. Five contacts per company is not thoroughness; it is five simultaneous cold emails to colleagues who sit near each other.
• The list is segmented by sub-vertical, so each segment can receive copy written for it and report its own numbers.
• The build date is recorded, because the quarterly refresh clock starts now.

If a list you have been handed — bought, inherited or built in-house — cannot pass these checks, it is not a database yet. It is raw material.

What does a sensible first step look like?

If you want to test the quality of the approach before committing to a full outbound build, the database is the right place to start, because it is the foundation everything else sits on. We offer a standalone B2B database build at £950: your ICP in, a verified list out, built through the exact pipeline described above. It works as a proving step — you see the data quality before deciding whether to run campaigns on it.

If the campaigns themselves are the goal, the Outbound Engine is the full system: database, sending infrastructure, copy and sequencing, live in 30 days.

---

Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks and what to build first.