B2B email marketing: the list matters more than the copy

B2B email marketing lives or dies on the list, not the copy. A sharp message sent to the wrong account gets ignored. An ordinary one sent to an account with the exact problem you solve gets a reply. So the work worth automating sits upstream: finding the accounts showing signs they need you, ranking them, and putting a person in front of the best ones.

Most teams are automating the other end. LLMs are everywhere in outreach now, but almost entirely in the copy. Most of it reads as slop, and recipients can tell within a line.

The more useful job for an agent is research. Agents can chain context across sources and read a company's website code, which makes a kind of filtering possible that no sales rep ever had the hours for. We call the method signal stacking, and the rest of this article walks through it.

Why bought B2B email data lists don't get replies

A bought list is the same list your competitors have. Filter a data provider by industry, headcount and funding stage and you get a perfectly reasonable set of accounts. Every other company selling into that market can run the identical filter in an afternoon, and plenty already have. Those inboxes are getting the same pitch from several directions.

Firmographics describe a company, not a problem. Headcount tells you a business is big enough to buy. It says nothing about whether they need what you sell this quarter. That gap is why list-first campaigns lean on volume: if you can't tell who needs you, you send to everyone and hope.

The filter is the baseline, not the strategy. Keep it. It cuts the universe down to companies that could plausibly buy. It just can't do the second job, which is finding the ones that should buy now.

Where LLMs actually belong in outbound

Writing the email is the part LLMs do worst for you. Generated copy has a shape people now recognise on sight. Personalisation that only pretends to be research ("I noticed your company is growing fast") costs credibility rather than earning it.

Reading is the part they do best. An agent can open a homepage, inspect its source, scan a careers page and return a structured answer for every account on a list. Work a rep would do one browser tab at a time runs across the whole list in a single job.

Chained, those reads become filters that didn't exist before. One signal is a hint. Three signals pointing at the same problem are close to a confession.

Signal stacking: a worked example

Take a simplified case: a company selling software that helps businesses set up their analytics properly. Three passes, each narrowing the last, then a ranking step.

1. Filter by industry, headcount and funding stage. This is the baseline list from above. Your competitors have it too. Its job is to define the universe, not to pick from it.

2. Check each homepage's code for Google Analytics or Google Tag Manager. A site running the Google tag loads a script from googletagmanager.com, and that reference sits in the page source where an agent can find it. No tag means no tracking. That company has the exact problem the software solves, and their own website just told you. It's a bigger gap than it sounds, which is why turning on Google Analytics is the first marketing step for most businesses.

3. Check their open roles. A fuzzy match on titles like "data engineer" or "data analyst" is a high-intent signal: the company has decided the problem matters and attached budget to it. The match has to be fuzzy because titles vary, and "analytics engineer" and "BI analyst" carry the same intent.

4. Have Claude rank each account against your ICP. Pass in the stacked signals plus a written definition of your ideal customer, and ask for a score and a one-line reason per account. The reason is the useful part. It becomes the opening of the message, because it's a trigger the account will recognise as true.

That's what turns the slow work of a sales rep into the repeatable work of an engineer. Thousands of accounts, not 20 a day.

A note for Australian markets. Funding stage is a weak filter here. Most Australian B2B businesses never raise a round, so a funding filter quietly empties the list. Swap it for signals that exist in quieter markets: ad library activity, review velocity, booking friction on the site, recent site changes, role changes.

[IMAGE 2: Screenshot of the ranked output table: account, signals found, score, one-line reason, tier]

Caption: The ranked list, sorted by score

Alt text: Table of B2B accounts with website tag check, matched job titles, an ICP score and a one-line reason

Why the top tier still gets a human

Ranking decides who gets your time, not whether a person writes. Once the list is ordered, a human takes over. The top tier gets a handwritten message from the founder, opening on the signal rather than on a pitch.

Tier the treatment. Write 1:1 for the highest-signal accounts and run a lighter batched sequence for the broader list. The engineering exists so the founder's limited writing time lands on the accounts most likely to answer.

That's the whole pattern. Machines do the research and the ranking. People write the message. Not end-to-end LLM spray and pray.

Engineers build the list. Founders write to the top of it.

What it does to your cold email reply rate

Reply rate is the honest scoreboard for B2B email marketing. It's the one number that reflects whether the recipient cared. Untargeted cold email sits low: in our cold email outreach system we work from a baseline of 1% to 3%, which is roughly a hundred sends for one conversation.

Signal stacking moves the number through relevance, not volume. A message that opens on something true about the account (no analytics tag, an open data role) reads as research rather than a template. Fewer sends, each one more likely to land.

Measure per tier and per signal. If the no-tag accounts reply and the hiring accounts don't, that's a finding about your market. The next run's filters should change because of it.

Why outbound is turning into an engineer's job

Clay, the data and workflow platform a lot of this gets built on, raised $115m at a $7.1bn valuation in September. Alongside the round it's funding a $1m scholarship to train what it calls GTM engineers: people who build revenue systems rather than send emails one at a time.

The job title tells you where the work has moved. Winning at outbound is now less about writing volume and more about designing filters, wiring data sources together and judging what the output means.

Most Australian B2B founders haven't properly adopted this yet, which leaves a first-mover window open for now. It won't stay open. Once every competitor stacks the same signals, the stacked list becomes the new baseline.

When this isn't worth building

If your whole market is 40 accounts, skip the stack. Research them by hand over a week and write to each one. The engineering pays off when the universe runs into the thousands and the signals that separate a buyer from a lookalike are buried where you can't read them manually at that volume.

The same goes if you haven't settled who you sell to. A ranking prompt needs an ICP to rank against, and if that definition doesn't exist yet, start with the GTM foundations before building anything.

Building it yourself

The spec matters more than the tool. Here's a starting version you can adapt and run in whatever you build with:

Build a signal-stacked account list for [product].

Universe: companies in [industry], [headcount range], based in [country].

For each account:

1. Fetch the homepage and check the source for "googletagmanager.com", "gtag(" or a "GTM-" container ID. Return yes or no.

2. Fetch the careers page or job listings and fuzzy match open roles against [role keywords]. Return any matched titles.

3. [A third signal specific to the problem you solve.]

Then score each account 1 to 10 against this ICP: [paste your ICP]. Return the score, a one-sentence reason naming the strongest signal, and a tier (1:1 or batched). Output a table I can sort by score.

Run it on 50 accounts first and read every reason line. If the reasons are wrong, the ranking is wrong, and no amount of copy will fix that downstream.

Common questions

Are B2B email data lists worth buying?

As a starting universe, yes. As the whole strategy, no. A bought list is the same one your competitors are working from, so treat it as the baseline you filter down from rather than the list you send to.

How do I get B2B leads without buying more data?

Add signals to the data you already have. Look for evidence on each account's own website and job listings that they have the problem you solve right now, then rank accounts by how many of those signals line up.

What's a good cold email reply rate?

Our working baseline for untargeted cold email is 1% to 3%. The more useful benchmark is your own last run: track reply rate per tier and per signal, and judge every change against that.

Do I need Clay to do this?

No. Clay makes sourcing, enrichment and chaining faster to wire up, but the method is a handful of questions asked of every account plus a ranking step. A script that runs Claude across a list of domains can do the same stack.

Should B2B email outreach be fully automated?

Not for the accounts that matter most. Automate the research and the ranking, then have a person, ideally the founder, write to the top tier by hand.

Automation Corner is where I break down growth systems worth automating, and how to actually build them. If you'd rather have your outbound list built for you, book a 30 minute call.