Best-Rep Email Cloning
Feed the agent your single best rep's best email, then let it AB-test variants—that's how AI beats your mid-pack.
- Difficulty
- Easy
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 90%
A method for making AI-generated outbound emails good instead of spammy noise: use your best salesperson's or marketer's actual best-performing email as the training template, then let the agent generate and AB-test variants with light personalization from your data. Use it when your AI outbound is producing generic garbage.
Origin
Asked why so many AI emails are terrible, Lemkin says the fix is to train on your best human's copy rather than let the agent invent from scratch. He notes bad AI email is a symptom of a poorly trained product, and that most human reps' emails are worse than people assume anyway.
Core principles
- 01Anchor the agent on your best human's best email — don't let it start from nothing.
- 02Agents excel at generating variants and AB testing; that's the one job to hand them.
- 03Light personalization from real data sources (CRM, website visitor tracking) lifts quality further.
- 04The bar to beat is your mid-pack rep's email, not your single best artisanal writer's.
How to run it
- 1
Harvest your best copy
Take the best email your top salesperson or marketer ever sent — the one that actually worked — as the template.
- 2
Train the agent on it
Upload that text and train the agent on it. Let it iterate and AB-test from that starting point rather than inventing tone from scratch.
Pro tip You can sanity-check by asking Claude/ChatGPT for three variants of your best email — they'll be pretty good.
- 3
Wire in a data source
Give it a data source (as simple as Salesforce). Many agents also track website visitors, so they can lightly personalize each email with real context.
Pro tip Website-visitor tracking lets the agent personalize based on what the prospect actually did.
Watch out Keep personalization light — the goal is a good facsimile of your best rep, not a fabricated deep-research masterpiece.
- 4
Send without hiding the AI
Testing showed disclosing vs hiding the AI didn't matter across hundreds of thousands of sends — people don't care if the email adds value and responds instantly.
Pro tip Founders often reply 'I can tell this is AI but it's pretty good, can we meet?' — which tells you it works.
In the wild
When Lemkin's startup was acquired by Adobe, sales leader Sam Blonde read the inherited reps' emails and called them the worst he'd ever seen. Lemkin's point: AI trained on your best beats emails like those.
→ Illustrates that the mid-pack bar AI must clear is low.
Common mistakes
Letting the agent write from scratch
Untrained agents produce the terrible emails everyone complains about. The template must be your best human's proven copy.
Manufacturing excuses not to deploy
Debating whether to disclose the AI is often an excuse to avoid the work; testing shows it doesn't move outcomes.
Is it for you?
Best for
GTM teams whose AI outbound is producing generic, low-response emails
Not ideal for
A small set of high-dollar enterprise logos where a top human should hand-craft each email
From the transcript
“take your best person on your sales team, the best marketer you have, take their email copy and use that as a template for your…”
“Agents are really good at AB testing. They're really good at creating variants.”
“If you're getting terrible emails, it's a poorly trained product from a bad vendor.”
From the episode
We replaced our sales team with 20 AI agents—here’s what happened
Jason Lemkin (SaaStr)