Churn, the bane of any subscription business. What if GenAI could help you with this? Since I’m asking what is clearly a rhetorical question, we can be pretty sure that the answer is “wait no more! It can!”

This is the first in what is going to be many useful LLM/Chatbot prompts for business use. What I have been seeing is that there’s all kinds of prompting guides, but they tend to be fairly generic and high-level in scope. This isn’t necessarily bad it’s useful to understand what a GATE prompt is and when you should use it, but I wanted to take a much more actionable…”will this help me solve my problem?” Perspective.

But so differently, I want to start from the solution and then give you the prompt.

In this case, I just want to help you analyze churn.

The Prompt

Analyze this customer data and identify the top 5 warning signs that predict churn within 90 days. Then create a retention playbook with specific actions for each warning sign, including email templates and intervention timing.

This prompt includes two sections to it. The first section is, here’s what I want the AI to look at, here’s what I want to be extracted. And then the second section is describing how I want you to help me, what I need in order to take action. You could do the first part with relatively traditional tools, but that second part, the recommendations, is the realm of a large language model.

The Data

In order to make this interesting I had to have some data to analyze and it’s hard to find real data sets that allow for this kind of analysis. So I had a different conversation with Claude, where I asked it to produce synthetic data that would allow for this sort of analysis. That data is available for download here so that you can give this a try yourself using the exact same data to be fair. The status is structured in order to make this pop a little bit more quickly. But this will absolutely work with your data.

The Comparison

I thought it would be interesting to compare the results between chatGPT, Gemini, and Claude. Gemini was having some sort of weird file upload issue so I was only able to do ChatGPT and Claude. The high level summary is that cloud gives a much better playbook and with a much more interesting analysis. They both get to basically the same point but Claude completely out strips ChatGPT with the recommendations however ChatGPT provide provides something very interesting, which is an offer to help build CRM templates and I will show you that conversation and discuss what that is.

ChatGPT Differentiation (Automation Templates)

I can help you create automation-ready templates for:
	1.	CRM tools like HubSpot or Salesforce:
	•	Trigger: Churn risk criteria met
	•	Action: Auto-email, task creation, and follow-up logging
	•	Format: JSON, YAML, or CSV for Zapier, Make, or native automation
	2.	Zapier / Make (Integromat) flows:
	•	Trigger: New entry in Google Sheets / CRM
	•	Filter: Check if criteria exceed thresholds
	•	Action: Send templated email via Gmail / Outlook or create Slack alert
	3.	SQL + Retool / internal dashboard alerts:
	•	Periodic churn-risk scan → flag customers → show in dashboard

Would you like a specific one of these generated for you? If yes, which tool are you using? 

Claude Differentiation (The Playbook)

Claude’s playbook is several pages of multi-stage reactions to each condition cropping up. ChatGPT’s playbook seems positively anemic in comparison. I highly recommend looking at the two and comparing for yourself. (See below, down below the summary.)

The Summary

Claude and ChatGPT came to basically the same conclusions (good), but took very different paths in terms of how to help the prompter with solving the problem going forward. ChatGPT opted for direct action, with helping the user develop templates in the CRM that would act to find these situations going forward. Claude opted to provide much more detailed responses for use in each situation. This is an excellent representation of each company – Anthropic with their skill and finesse in writing, and OpenAI having a more clinical, technical bent.