email personalization at scale

Email Personalization at Scale: How to Personalize Thousands of Cold Emails Without Burning Your Domains

In this post:

Email personalization at scale usually starts with one instinct: find a signal for every account. You then map the market, list the pain points and build a Clay table for each signal, like funding, hiring, new leaders, tech changes and news.

We've built that setup for clients, and it gets out of hand quickly. A market of 50,000 or 100,000 companies ends up with 20 different signals and a web of cross-referenced tables, and around 80% of the companies still don't match any signal at all.

So the personalization you worked hardest on reaches a small slice of the list, while the rest gets a template, and the team spends its time maintaining tables instead of improving the message.

The setup we use now starts from segments built on company characteristics, writes each segment's message around a pain point you already know it has, and adds signals and AI-written lines as a layer on top.

We'll walk you through that system in this step by step guide, from the segments and context documents to the review, the sending setup and the numbers that tell you what's broken.

TL;DR

  • Personalization at scale should start with segments built on company characteristics, which tell you the pain point before you look for a single signal.

  • Most of your market won't have a usable signal, so you need to add signals as a layer where they exist and tier your list so the top accounts get hand-written emails.

  • Let AI write personalized email lines only from your columns and context documents, with web search off, a fallback for every signal and a person who checks every pitch before it sends.

  • Send at a volume your domains can handle, with small, targeted batches, so your spam complaint rate stays well under 0.1%.

  • Measure meetings by segment, signal and tier, and use the reply pattern to spot targeting, messaging or infrastructure problems.

Why personalization breaks when you scale outbound

From our experience, most teams usually fall into one of four traps when they scale outbound, and each one also hurts their domain reputation.

  1. Merge-field templates look personalized to the sender and generic to the buyer. In Hunter's survey, 61% of decision makers said cold emails "suck" because they aren't relevant.

  2. Hand-writing every email keeps the quality high and caps your outbound at however many emails your reps can write.

  3. AI blasting results into sending invented facts and the same flattering line to large lists, and the spam complaints that follow damage your domains.

  4. Signal overload builds a table for every signal until nobody can maintain the setup, while most of the market still has no signal to personalize on.

As well, data from Gartner show that 73% of B2B buyers "actively avoid suppliers who send irrelevant outreach", and Google tells senders to keep spam rates below 0.1%. That means the fix has to make every email more relevant and keep the volume under control.

Gartner press release: 73% of B2B buyers actively avoid suppliers who send irrelevant outreach

Source: Gartner, Sales Survey 2025

How to safely scale your email personalization and your outbound campaigns

Here's a step by step guide to help you safely personalize your emails at scale.

Step 1: Segment your market by company characteristics

Most of the lift in email personalization at scale comes before anyone writes a personal line. Who you contact and which problem you lead with decide most of your replies.

That means your outbound message/email being relevant to the recipient is way more important than any personalization you can ever write.

Start by mapping your total addressable market, then cut it into segments based on characteristics you can find for every company. Lookalike tools like Ocean.io or AI Ark help with the mapping.

Say you sell paid advertising services to SaaS companies in the Netherlands. You could split that market by two characteristics: whether a company raised funding recently, and whether it has a CMO.

  • Recent funding and a CMO: the company wants to scale, and someone already owns marketing.

  • Recent funding and no CMO: the company has money to grow but nobody in-house to set up paid advertising.

  • No recent funding, with or without a CMO: budgets are tighter, so the message needs a different reason to act.

Each segment points to a different pain point, and you know it without finding a signal for a single company. That's relevance at the segment level, before any personalization starts.

Once you know the pain point for each segment, write it down as an angle. An angle is the short brief every email in that segment is built on: who you’re writing to, the one problem they have, why it matters to them now, the proof that you can solve it and one low-pressure ask.

That brief does most of the work. Every email in the segment will use it, so the only thing that is going to change from one account to the next is the opening line, and that line just has to make the angle feel specific to the company.

We saw how much this matters with one of our clients, Crisiscentrale. They’re a crisis management firm that sends crisis managers, communication advisors and psychosocial care providers, day or night, when a fire, a cyberattack or a misconduct case stops an organization from working.

Instead of one campaign for every kind of organization, we built a separate campaign for each sector they serve, like provinces, water boards, municipalities and healthcare, and wrote each one around the risks that sector deals with.

Those sector campaigns got 203 replies from 1,881 cold emails, an average reply rate of 10.8%. For comparison, Instantly’s 2025 benchmark puts the average cold email reply rate at 3.43%, so the angle carried most of the result.

Step 2: Enrich every account with the data the email will use

If the facts aren't in the record, no writer and no AI model can write a relevant line.

Decide which fields your segments and emails need, like company size, funding, leadership roles and tech stack, plus the right contact with a verified email address. We fill them in Clay with an enrichment waterfall, and every fact keeps its source URL so anyone can check a line before it sends.

A thin CRM record next to an enriched one, where every fact for email personalization keeps its source.

While you're at it, only collect what the email will use, because GDPR limits personal data to what's "necessary" for the purpose.

Step 3: Add signals where they exist, then tier your list

A signal is a public event that gives you a reason to write and reach out now. Signals turn a segment angle into a personal email, because the email can name the event and connect it to the problem in the same breath.

You won’t find a signal for every account, so don’t try. In a market of 50,000 or 100,000 companies, even 20 different signals can leave most of the companies without any of them. And every extra signal adds another Clay table to connect to the rest, so the setup gets harder to manage each time.

So pick the few signals that fit each segment’s pain point. When an account shows one, the email can mention it, and when it doesn’t, the segment angle still does the job.

Signal

Why it's a reason to write

Example opening line

The company is hiring for sales roles

New reps take months to ramp, and someone has to fill their pipeline in the meantime

"I saw you're hiring [number] SDRs. Getting new reps productive quickly usually depends on having pipeline ready for them."

The company exhibits at or organizes a large event

A big event creates a lot of work in a short window

"Having 350 exhibitors at [event] is a real operational lift. How are you handling booth sales?"

An office move or renovation

A move is a project with a fixed deadline

"Moving into a 10,000 square meter office in October is a big project on top of the day job."

Someone from the company visits your website

They're already researching the problem you solve

"Someone on your team was reading about contract management on our site recently. Would you be open to exchanging ideas?"

A live risk in the news

The risk is on their mind right now

"With protests planned in [city] this week, how prepared is your team if things escalate?"

Drop stale signals, then tier your list.

  • Top accounts: a person researches and writes each email by hand, because one meeting with these accounts could change your quarter.

  • Accounts with a signal: AI writes a personalized line from the signal, and a person checks it.

  • Accounts without a signal: strong segment copy with no person-level line, which still beats a generic template.

Don't automate a step just because you can. If a job only takes a few minutes by hand, an automation can cost more time to build and maintain than it saves.

Tag every lead with its tier and the signal that qualified it. You'll need those tags in Step 7, and our post on signal-based outbound campaigns shows how we set these campaigns up.

Step 4: Let AI write the personalized lines, within rules

AI is what makes the middle tier possible, because it can write a line for every account with a signal in minutes. Without rules, it invents facts, flatters, opens every email with "I noticed" and picks the newest fact instead of the most relevant one.

Give the AI context documents instead of the open web

The most reliable way we've found to write personalized openers is what we call context engineering. The AI step in Clay works with web search switched off, so it can only use the inputs you give it.

Those inputs are the columns in the row, like the segment and any signal you captured, plus a few context documents:

  • A writing guide with your tone, length, banned phrases and a handful of good example lines.

  • A company document that explains what you sell, who it's for and the proof you can use.

  • A persona document for each buyer type, with their responsibilities, KPIs and the pain points they deal with.

We believe the best material for the persona documents should come from real conversations. Recordings and notes from sales calls show the pain points each industry brings up, in the words buyers use, and those words make the opening lines sound like someone who knows the job.

Here's the shape of a prompt we'd use for an opening line:

  • Role: You write the first line of a cold email for [your company], which [one sentence on what you sell].

  • Inputs: company name, segment, persona, signal, signal date and signal source, taken from the row.

  • Context: the writing guide, the company document and the persona document for this persona.

  • Task: write one sentence under 25 words that connects the signal to a pain point from the persona document.

  • Rules: use only the inputs and context, follow the writing guide, and never open with a compliment, "I noticed" or "I hope this finds you well".

  • Fallback: if the signal is empty or too old, return EMPTY so the segment copy is used instead.

When the AI can't search the web, it can't pull in a wrong fact about a similarly named company, and the EMPTY fallback stops it from inventing one. You can read our post on prompt engineering for GTM for more imformation on this.

Our post on how we build Claygents covers the research side, for the cases where you do want an agent to look something up. Keep the rest of the email short, too: Gong found the best cold emails land between 50 and 100 words, and that pitching cuts reply rates by as much as 57%.

Also, buyers notice when the AI did all the work. In Hunter's survey, 69% of decision makers said it bothers them if AI was used to write an email, "unless the output feels genuinely human".

Hunter survey result: 69% of decision makers are bothered by AI-written emails unless they feel human

Source: Hunter, The State of Email Outreach 2026

Step 5: Check every pitch before it sends

We check every pitch before it sends, because that rule catches most of the problems above.

The check is quick when you know what to look for:

  • The fact in the personalized line matches its source.

  • The signal is still current on the day you send.

  • No merge field is empty, which is how emails end up starting with "Hi ,".

  • The email ends with one low-pressure ask, with no calendar link in the first message.

A cold email draft checked against four rules before it sends, catching an empty merge field.

Anything that fails goes back to the fallback or gets rewritten by hand, and repeated mistakes tell you which context documents to fix.

Step 6: Send at a volume your domains can handle

Personalization only pays off if the email reaches the inbox, so spread the volume across several sending domains and mailboxes, and keep each mailbox well under the hard limits.

Sending tools recommend around 30 cold emails a day per mailbox (Instantly) and no more than 50 (Smartlead), far below Google Workspace's limit of 2,000.

How many mailboxes you need

The maths is simple once you pick a daily volume. To send 1,000 cold emails a day at around 30 per mailbox, you need about 34 mailboxes.

Instantly recommends three to five mailboxes per domain, so those 34 mailboxes need somewhere between 7 and 12 sending domains. Keep cold email off your main domain, because Google counts mail from a domain and its subdomains together toward its bulk sender threshold.

Mailbox maths for 1,000 cold emails a day: about 30 per mailbox, 34 mailboxes, 7 to 12 domains.

New mailboxes need warming up before they carry that load. Instantly recommends at least two weeks and lemlist three to four, so plan the extra domains a month before you want the volume.

Keep each send small and targeted

Smaller, targeted sends also get more replies. In Hunter's data, sequences with 21 to 50 recipients averaged a 6.2% reply rate, against 2.4% for sequences with more than 500.

Hunter data on reply rates: 6.2% for sequences of 21 to 50 recipients against 2.4% above 500.

Skip spintax as a trick for getting past spam filters. Deliverability expert Al Iverson calls random variation used to dodge filters "hashbusting", and warns that if filters detect it, "it is a clear signal that you are trying to manipulate the system."

Al Iverson on Spam Resource explaining hashbusting, random variation that tries to get past spam filters

Source: Spam Resource, DELIVTERMS: Hashbusting

Match your sending setup to who you're emailing

Who you email changes how hard it is to reach the inbox. Larger companies usually protect their inboxes with email security gateways like Proofpoint, Barracuda or Mimecast, which are much harder to get through and quicker to block a sender.

If most of your market uses Microsoft Outlook, send from Outlook mailboxes too, instead of only Google Workspace. For enterprise targets with thousands of employees, plan for domains that have aged for a year or more before they carry cold email.

Watch the names of your sending domains as well. We avoid sending domains that copy the company name one to one, because in our experience, when one domain in a set gets blocked, others with the same name pattern tend to follow, so we use names that describe the service instead.

We cover the domains, authentication and warm-up behind all of this in our email deliverability guide.

Step 7: Measure by signal and tier, then feed the results back

Measure personalization by the meetings it books. Use the tags from Step 3 to compare positive replies and meetings by signal and by tier, and stop using signals that don't book meetings.

To check whether the personalized lines beat your fallback copy, hold back a random share of each list and send them the template version. Our glossary post on the AI personalization engine explains more about this.

Read the reply pattern before you change anything

When a campaign underperforms, the problem is almost always one of three things: infrastructure, targeting or messaging. The mix of replies tells you which one to look at first.

What you see

Likely problem

Where to look

Very few replies of any kind

Infrastructure, targeting or messaging

Start with deliverability, because emails in spam get no replies at all

Plenty of replies, few of them positive

Messaging or targeting

The angle, the proof and whether the segment really has the pain

Few replies, but most of them positive

Infrastructure or targeting

Inbox placement, and whether the list is too narrow or too broad

Feed what you learn back into the first steps: better signals get more accounts, weak angles get rewritten, and the lines your reviewers keep fixing become new rules in the writing guide.

Personalization at scale and privacy in the EU and the Netherlands

Personalizing at scale means researching a lot of people, so European privacy rules apply, and you should check your setup with an expert.

When you contact someone using data you didn't get from them, GDPR Articles 14 and 21 require you to tell them who you are, where you got their data and that they can object, "at the latest at the time of the first communication".

GDPR Article 14 listing what you must tell people when their data was not collected from them

Source: GDPR Article 14, Information to be provided

In the Netherlands, the Telecommunications Act also requires prior consent for commercial email, unless the business published the address for that purpose. Emailing a named person's work address isn't automatically allowed, so get advice before you scale outbound into the Dutch market.

If you want to scale outbound without losing what made your first emails work, book a call with us.

What else people ask about personalizing emails at scale

What is personalization at scale, and how is it different from a mail merge?

Personalization at scale means sending relevant, individual emails to thousands of prospects without writing each one by hand. A mail merge swaps a name into the same email for everyone, while personalization at scale changes the reason for writing, based on the segment and any signal you found.

Do you need a signal for every account?

No, and trying to find one usually makes the setup too complicated to maintain. Segments built on company characteristics already tell you the pain point, so accounts without a signal get strong segment copy and move up a tier when a signal appears.

What is context engineering for cold email?

Context engineering means giving the AI everything it needs to write a line, like the row's data, a writing guide, a company document and persona documents, with web search switched off. The AI then writes only from what you've given it, which keeps the facts accurate and the tone consistent.

Can AI personalize emails at scale without sounding like AI?

Yes, if it writes one or two lines from real facts, follows a writing guide with banned openers and falls back to segment copy when there's nothing useful to say. Keep a person on the final read.

What is a good reply rate for personalized cold emails?

Instantly's 2025 benchmark put the average cold email reply rate at 3.43%, with the top quarter of senders at 5.5%.

What tools do you need for email personalization at scale?

You need a data and enrichment tool like Clay, an AI model to write the lines, a sequencer like Instantly or lemlist, a CRM and enough warmed sending domains and mailboxes. Start the domains first, because new mailboxes need two to four weeks of warm-up before they carry real volume.

Revenue tips, Weekly

Workflows, automation strategies, and GTM insights delivered straight

Ready to build your
GTM engine

Let's talk about your sales challenges and how we can help you scale

Nebor brand chevron in purple, section accent
Nebor brand chevron in orange and purple, section accent
Revenue tips, Weekly

Workflows, automation strategies, and GTM insights delivered straight

Ready to build your
GTM engine

Let's talk about your sales challenges and how we can help you scale

Nebor brand chevron in purple, section accent
Nebor brand chevron in orange and purple, section accent

Stay ahead with Nebor

Stay ahead with Nebor

Get insights on workflows, GTM strategy, and automation delivered to your inbox.

Get insights on workflows, GTM strategy, and automation delivered to your inbox.

By clicking Sign Up you're confirming that you agree with our Terms and Conditions.

Services

Case Studies

Resources

About

Services

Case Studies

Resources

About

Nebor logo, a go-to-market and RevOps agency based in Amsterdam

Services

Case Studies

Resources

About

Nebor logo, a go-to-market and RevOps agency based in Amsterdam