1 of 15

Getting a Daily Digest of the So Whats �from a Series of AI Newsletters using relay.app

James Raybould, Turing

2 of 15

Sample of final product

(this is just the 1st 2 emails summarized)��1. Custom email title

2. Custom header for each email summary

3. Overall summary

4. Specific So Whats

5. Link if I want to read the full newsletter

3 of 15

This is the overall set of 5 steps��1. Set schedule�

2. Choose emails to synthesize�

3 and 4 �Choose how to synthesize them using LLMs�

5 Send one email back to me

4 of 15

Step 1: Schedule��In this case daily at 6:30 am PT

5 of 15

Step 2: Find emails��This shows the 1st 2 emails I pull in (I show all on next slide)��You can do 2 emails or 30 emails, it’s just a question of how many OR statements you want to write (and how many API calls you want to make)

6 of 15

Step 2: Find emails

This shows ALL the emails I synthesize - zoom in to 200% or 400% - but the key point is that you can do it for 1 email or 20 emails��[it’s just adding OR statements and knowing your API costs with LLMs will be higher]

7 of 15

Step 3: Use Iterator

Iterators let you quickly process a whole list of items. In this case, you want to do an AI analysis of each newsletter email you found in the previous step

8 of 15

Step 4: Use GPT-4o mini to analyze the email

This is the key step where the AI does work for us! You need to give the AI:

  • The inputs: The email
  • The instructions: the prompt to explain what you want it to analyze
  • The outputs: what data you want to return for each email

You can easily test your prompt on a single email to iterate on it.

9 of 15

Step 4: Use GPT-4o mini to analyze the email

I chose to use GPT-4o mini, but you can any models, including ones Relay.app offers

Under Apps on the left-hand rail you can connect to the various apps; in this case you’d be connecting to OpenAI and adding your API key there [which you get from: https://platform.openai.com/docs/overview]

10 of 15

Step 4: Use GPT-4o mini to analyze the email

Here is the full prompt from �the prior page

11 of 15

Step 4: Use GPT-4o mini to analyze the email

And here is the prompt text itself if you’d like to copy and paste

You are summarizing email newsletters for a busy executive. Focus on extracting the high-level themes, major announcements, and key actionable insights in the newsletter. Structure the response as follows:

[Insert Email Subject using @ or Attach Data]

Summary: Provide me with a summary of the [Insert Email body]

in a single paragraph. Be concise: provide me with just the summary itself, don't add any introductory text or additional fluff.

Focus on the body of the email, but make reference to [Insert Email Subject] and [Email Name] if it adds clarity.

Key Takeaways: Break down the most interesting takeaways, insights, or So Whats into clear, digestible bullet points. Output this is a bulleted list (-). Use no more than 5 bullets. ��

Email link: [Insert Email Email URL]

Name: [Insert Email Name]

12 of 15

Step 4: Use GPT-4o mini to analyze the email

And here is the Output where it defines the format of the email itself

13 of 15

Step 5: Send me the synthesized email

Finally, we need to synthesize all of the individual AI analyses that ran for each newsletter into a final email.

To do this, we can use a special feature called “unrolling” that takes a list of items and combines them into a single, nicely formatted summary.

14 of 15

Reminder of

my final product

You can customize �your prompt and �formatting as much�as you’d like��Note: it may take some trial-and-error, but relay.app does a nice job of letting you test each step��I’d suggest starting with one email first

15 of 15

There are lots of similar use cases we can build with similar logic

  • Meeting Preparation�
  • Email So Whats pushed to Slack

  • Meeting Follow-up and synthesis (across ALL meetings)

  • Competitor news and intelligence tracker

  • Competitor marketing analysis - campaigns, messaging, channels etc.

  • Customer segment intelligence