Automation templates

How to Analyse Client Calls with AI to Spot Churn Early

You can analyse every client call automatically by sending each transcript from an AI note taker such as Fathom into OpenAI through Make.com. The AI sorts each call into one of three paths: no problem, feedback suggested for the CSM, or escalation necessary. It then logs the call in Google Sheets and sends the right Slack message, so problems and opportunities are not missed.

Ask your AI about this guideChatGPTClaudePerplexityMarkdown
Key takeaways
  • An AI note taker gives you the transcript, and OpenAI classifies the call into one of three paths.
  • Ask the AI for JSON output so the router can read a consistent category.
  • Feedback goes to the CSM by direct message, and escalations go to a shared success channel.
  • Every call is logged in Google Sheets with its path, so nothing is lost.
  • Treat the example prompt as a starting point and write your own using your agency's processes and churn indicators.

What does an AI call analysis system do?

When success managers are on many client calls, things get missed. A client may complain, or mention an underlying issue so lightly that the CSM slips past it. This system analyses every call with AI and flags negative moments, so your success team can follow up and fix the issue rather than ignore it.

It also spots opportunities, such as a chance to upsell, to run training for clients or CSMs, or to ask for a testimonial. The aim is to solve problems before they grow and give clients a better experience, which supports retention.

What do you need to build it?

You need a no code automation platform, which connects APIs. Zapier, Make and n8n all work. We use Make.com here. You also need an AI note taker with a good API that integrates with other systems, such as Fathom, Fireflies or Otter, plus OpenAI, Google Sheets and a messaging tool. We use Slack, but Teamwork or Discord would work too. The Fathom connection works on a free account.

How does the call flow work?

A call happens on Zoom, Google Meet or any call software. When it ends, the AI note taker produces the transcript. AI then analyses it and the call goes down one of three paths.

  • No problem: nothing else is needed. Log the call.
  • Feedback suggested: the CSM could have handled the call better, for example by missing a step in an SOP or script. Find the CSM, send them feedback, and log the call.
  • Escalation necessary: the client is unhappy, something was not answered, or there is an upsell, testimonial or training opportunity. Send a message to the general success channel with the call link, and log the call with that flag.

How to build it in Make.com

  1. Create a new scenario and add the Fathom module with the Watch New Recordings trigger. Connect your account. Each new recording now triggers the automation. The output includes the transcript, meeting title, recording ID, URL, action items and who recorded it.
  2. Add a Set Variables module for the transcript, meeting title, recording ID, recording URL and the recorded by email. The email tells you which CSM to message. While building, fill these with test values so you do not need to wait for real calls.
  3. Add an OpenAI module using Generate a Response. Open advanced settings and set the output format to a JSON object.
  4. Write the prompt. Tell the AI it is analysing CSM calls for your agency, to summarise the call, and to give one outcome. Name the three categories exactly: no problem, feedback suggested and escalation necessary. Tell it to assign only one, and include an example of the JSON format with a call category and a summary.
  5. Add a Router with three routes. Set a filter on each, where the call category equals the matching category. Matching names exactly is why consistent wording matters.
  6. Build each route, test it, then log every call in Google Sheets.

What does each route do?

For no problem, add a Google Sheets Add Row module. Map the recording ID as the call ID, the recording URL, the transcript, the assigned CSM from the recorded by email, the call category, the date and the summary as notes. The sheet can also hold client name.

For feedback suggested, use Slack's search for a user by email to find the CSM and get their user ID. Then add a second OpenAI module that writes feedback for the CSM. In that prompt, remove the JSON requirement and ask for text. Send it as a direct message using the user ID. Then log the row. A Block Kit message gives better formatting than plain text.

For escalation necessary, add an OpenAI module that identifies the specific escalation and writes it as a task the success team can act on. Post it to the success channel with the CSM email and the call link, then add the row to the sheet.

How do you test and improve it?

Use ChatGPT to generate three example transcripts, one for each path. Paste each into the variables and run the scenario. Always test every route, or the automation can fail in practice. Check that the right message arrives and that the sheet has a new row each time.

Then improve it. Rewrite the prompts for your agency, adding your own SOPs for calls, the points CSMs often forget, such as asking for feedback or setting expectations, early churn indicators, and which escalations matter. You can also link escalations to a project management system. The video above walks through the full build.

Questions people ask

How can AI analyse client calls automatically?

An AI note taker produces the transcript when the call ends. Make.com sends it to OpenAI, which returns a category and summary as JSON. A router then logs the call and sends feedback or an escalation message depending on the category.

Which AI note taker should I use?

It does not matter much. Fathom, Fireflies and Otter are examples. Choose one with a good API that integrates with your automation platform.

Why ask OpenAI for JSON output?

The router needs a consistent category to decide which path to follow. Asking for a JSON object with a set list of category names means each filter can match exactly.

Do I need to use Slack?

No. We use Slack because it is common, but Teamwork, Discord or any messaging tool your team uses can receive the feedback and escalation messages.

Rather have it built for you?

Fluxomate builds internal tools and AI agents for agencies. On your stack, owned by you.

See if you qualify