How to Train an AI to Write Sales Emails

Tools like AiSDR shine at creating and scaling high-volume sales content like emails and text messages.

However, before any AI sales rep can start writing like a pro, it needs enough information and context so that it knows what you want.

Here’s a closer look at how you can train an AI SDR to draft sales emails that sound like you wrote them.

TLDR: 

• The goal: Train an AI SDR to write emails for you
• The tactic: Convert your best email into an email template
• The result: AI SDRs generate emails that consistently meet your expectations 

Step 1: Find your best email

Your first step should be to review your campaigns and select the best emails with the strongest results.

The Goldilocks zone for generative AI is 3-5 good emails. Too few and AI SDRs won’t be able to identify a pattern to mimic your writing. Too many and it will overtrain the AI, leading to inconsistent results.

Here’s an example of an email you might use:

Notice that the language itself is easy for AI to digest and reproduce. It’s concise, specific, and doesn’t use any weasel words.

On a side note, if you plan to tell an AI SDR how to write follow-ups in the same prompt, we suggest limiting yourself to 3 initial emails. This is because you’ll be including 3 examples of follow-up emails and too many emails could throw off the AI SDR’s performance.

Step 2: Convert the email into a template

Once you’ve picked an email, you’ll need to turn it into a template by removing identifying information and inserting placeholders.

Importantly, each placeholder should contain a clear, short summary of what information is expected. 

For instance, you might replace “4 weeks” with the word timeframe, period, or duration. Or if you’re adding personalization placeholders, you’ll likely use first_name, position, company, and pain_point.

Here’s what this might look like.

Alternatively, if you don’t want to create 3 individual templates, you can mark up your best-performing email using spintax. This will, for all intents and purposes, achieve a similar result.

Step 3: Anticipate frequently asked questions

The end goal of your email is to get a positive response. Even better if it’s a meeting request. But before you get there, you might need to answer some questions.

Fortunately, there’s no need to make assumptions. Since you pulled your best email, you should have plenty of data regarding questions that leads typically ask. Simply identify the 6-8 most common questions, then create a list of those questions and their answers.

In case you don’t have such a dataset, then you can pull frequent questions from your demos and customer interviews. You can always fine-tune the list after a few campaigns.

Here’s an example of what a question-answer pair might look like:

Step 4: Build the email into your AI SDR prompt

Generative AI responds better when you use a markup language like XML to structure and format your sales persona as an AI prompt. (Whether you use a prompt chain, prompt chunk, or one large prompt is up to you.)

In this case, you might use <sample_initial_email> and <initial_email_template> to tell the AI SDR which is which. It also simplifies the rest of the prompt as you can use terms like sample initial email and initial email template and the AI SDR will know what you’re referring to.

And for your FAQ, you can simply number each question like <question_1>, <question_2>, etc.

If you put everything together, here’s what this might look like:

If you run into any issues at this stage, you can follow the best practices for debugging AI prompts.

The Result

And voila!

Repeat this process 2-4 more times, and you’ll have a large enough body of work for your AI SDR to write sales emails the way you want them.

On a final note, you’ll want to include other information like prohibited words and social proof. This will help your AI SDR really fill the role of a typical member of your sales team.