How I Personalize 750,000 Cold Emails a Month Using Clay and AI

In the past 30 days, my team sent more than 750,000 hyper-personalized cold emails across 19 different niches for our clients. Every subject line, every email body, every follow-up was customized using an AI process built inside Clay. Since we rolled this out, reply rates have climbed sharply, and the process is more repeatable than anything we've run before.

But here's the thing most people skip: none of this works if your lead list is wrong. The AI can be perfect and you'll still get nothing if you're emailing the wrong people. So let me walk you through the full system, start to finish.

Start With a Targeted Lead List (Under 60 Seconds)

Every client that comes on board fills out a thorough onboarding form. Who's your ICP? What do you sell? What pain points do your customers have? How do you uniquely solve them? Any case studies or marketing collateral?

I take that information and run it through a prompt in Claude to extract the exact company filtering criteria: industries, company sizes, geography, signals of an established business (not a startup), and personas to avoid. That output gives me clean parameters before I ever touch a data source.

From there, I pull the lead list, bring it into Clay, and run it through a waterfall enrichment strategy. For verification, we're currently using Million Verifier and TryKit. They're the most cost-effective and accurate combination we've found for our clients.

Build the Macro Research Layer First

Once the list is clean, I want to understand everything happening in the target industry before writing a single word of copy.

For a recent client, Productivo, who helps US product developers find manufacturing outside of China, the macro context was obvious: tariffs. I fed GPT a detailed prompt asking for a comprehensive overview of the current tariff landscape, tailored to the specific business personas we'd be reaching out to. That output becomes a permanent column in Clay, the macro environment layer.

This matters because it gives the AI enough context to write something like: "With tariffs now up to 50% on steel and a universal 10% on imports, costs must be creeping up fast." A human account executive would need 15 to 30 minutes and deep industry knowledge to write that line. The AI does it in seconds, for every contact on the list.

Go Deeper With Company-Level Micro Research

The macro layer tells the AI what's happening in the world. The micro layer tells it what's happening inside each specific business you're emailing.

I build a custom research prompt in Clay that takes the prospect's website, their LinkedIn company page, and the individual contact's LinkedIn profile, then runs a multi-step research process to surface the specific pain points that business is likely facing. Two things I've learned make this dramatically better:

1. Make the role creation incredibly detailed. If the AI is writing to a supply chain director, tell it to act as a senior supply chain consultant. The more specific the role, the sharper the output.

2. Turn on reasoning. Without it, the model defaults to the cheapest path. With reasoning enabled, it makes inferences like "this company's website looks static, so they probably rely on existing channel partnerships", and that kind of insight is exactly what makes copy feel researched rather than templated.

One more thing: I added a confidence column to the Clay table. Anything rated medium or below gets re-run. The difference in output quality between a high-confidence result and a medium one is significant, especially when reasoning is involved.

Write the Cold Email Copy With the "Why You, Why Now" Framework

Now I have three columns feeding into the copy step: the company overview (who my client is and how they help), the macro research, and the micro research. I pipe all three into a GPT prompt built around the "Why You, Why Now" format.

The structure is simple: make a specific observation, ask a pain-oriented question around it, explain how you can help, and close with a value-driven CTA or a relevant case study. It's short. It's direct. It's the best-performing format I've seen across our client base.

A few things I bake into the role creation for the copywriting prompt that make a real difference:

• Write in the voice of a founder speaking to another founder at a crowded industry trade show

• Be casual and concise

• Short words, short sentences, short sections, one to two sentences per paragraph, separated by a natural line break

• Write at an eighth-grade reading level

• Start sentences with "and" or "but" instead of "therefore" or "furthermore", that's how people actually talk

I output the results as JSON so the subject line and email body land in separate Clay columns, ready to import directly into Smartlead or Instantly. One small operational note: if you're running multiple campaigns to the same contact list, tag each record with the associated campaign. It prevents message overlap and keeps your sequences clean.

📥 Best Cold Email Software 2026

The 7 cold email tools worth your money in 2026 — ranked by an agency managing 25,000+ inboxes.

Get it here →

Follow-Up Emails: The "I Didn't Do a Good Job Explaining" Opener

The best-performing opener I've found for follow-up emails is: "I reached out the other day, but likely didn't do the best job explaining..."

It sounds human. It doesn't feel like a chase. And it gives you a natural bridge to restate your value prop from a different angle.

For the follow-up format, I use the four Ps (picture, promise, proof, push) rather than "Why You, Why Now." The structure: reference something specific from the micro research, restate how you help, use two or three bullet points to show proof, and close with a no-friction CTA. Keep it tighter than the first email, I aim for 600 to 800 characters on most follow-ups.

You can extend this same logic to a breakup email, where you call out a colleague at the company by name. We've found that to be surprisingly effective, though it does burn more AI credits.

Bonus: Competitor Takedown Campaigns

One other use case worth mentioning: if you're running a competitor displacement campaign, source your list using a tool like BuiltWith to find companies actively using the competitor's product. The flow is nearly identical, company overview, persona research, micro research, but you add one more variable: a direct comparison between your solution and theirs.

The copy formula stays the same. Short words, short sentences, short sections. Call out the specific friction the persona feels with the current tool, paint a picture of what life looks like without it, and give them a reason to take the next step.

Key Takeaways

• A bad lead list kills personalization before it starts. Filter by ICP criteria before enriching anything.

• Build three research layers in Clay: company overview (your client), macro environment (industry trends), and micro environment (individual company research).

• Use detailed role creation and enable reasoning in your AI prompts. Both meaningfully improve output quality.

• "Why You, Why Now" is the strongest format for initial cold emails: observation, pain question, solution, CTA.

• Follow-ups perform best when they open with a human-sounding reset: "I reached out the other day but likely didn't do the best job explaining."

• Output copy as JSON so subject lines and email bodies map cleanly to separate columns in Clay.

• Tag every record with its associated campaign when using Smartlead or Instantly across multiple clients or sequences.

Frequently Asked Questions

What AI models does Troy use for this Clay workflow? Troy primarily uses GPT-4.1 Mini for the Clay prompts because it's more cost-effective than Claude's models. He goes back and forth between the two but defaults to GPT-4.1 Mini given Anthropic's pricing. Claude is used separately for building out client knowledge bases and project overviews.

How does the macro vs. micro research structure actually work in Clay? Macro research is a single column built from a GPT prompt about the broader industry landscape (in the Productivo example, the 2025 tariff environment). Micro research is a per-row column where a custom Clay script visits each prospect's website and LinkedIn profiles to surface company-specific pain points. Both columns then feed into the copy-generation prompt alongside the client's company overview.

What email copywriting frameworks does Troy recommend? For initial cold emails, he uses "Why You, Why Now" (specific observation, pain question, solution, value-driven CTA). For follow-ups, he uses the four Ps (picture, promise, proof, push). He also mentions a hook-pivot-proof-solution format he's tested. All formats share the same writing rules: short words, short sentences, eighth-grade reading level.

What email verification tools does Troy currently use? Million Verifier and TryKit. He describes them as the most cost-effective and highest-accuracy combination his team has tested for client campaigns.