One of the most underused opportunities inside companies is taking the data you already have and turning it into something useful. Qualitative data used to be hard to classify at scale, but with AI you can now structure sales call transcripts and convert them into product, marketing, and commercial insights using a no-code workflow in Relay.

This walkthrough shows how to connect Google Drive, Google Docs, ChatGPT, and Google Sheets to automatically read meeting recordings, extract recurring questions, and surface coaching recommendations for your sales team.

What problem does this AI workflow solve?

Sales teams generate hours of conversations every week, but most of that knowledge stays trapped inside transcript files nobody reads. Automating the analysis lets you spot patterns across dozens of meetings without manually reviewing each one.

The goal here is to take transcripts sitting in a Google Drive folder, run them through AI, and push categorized findings into a spreadsheet you can later turn into a dashboard.

What kind of data can I analyze with this flow? Any qualitative text source: sales call transcripts, WhatsApp chats from an e-commerce, support tickets, or onboarding interviews. The logic stays the same; only the input changes.

How do I set up the trigger in Relay?

Relay is the automation platform powering the workflow, and it has a solid free plan to start. After logging in, you create a new workflow and lean on the built-in AI assistant to scaffold the steps [00:55].

The initial prompt describes the goal in plain language: analyze meetings with potential clients, identify recurring questions, generate insights for content, marketing, and product, and coach how to improve commercial conversations. The assistant then proposes the steps one by one.

The trigger fires whenever a new file lands in the Meet Recordings folder inside Google Drive. That single event kicks off the rest of the chain.

Why use Find Documents instead of a PDF reader?

The AI assistant initially suggested extracting the transcript as if it were a PDF, but Google Meet saves recordings as Google Docs. That mismatch breaks the flow.

The fix is adding a Find Documents step from the Google Docs integration [04:30]. You configure it to:

• Search by parent folder, pointing to Meet Recordings.
• Return a single document per run.
• Sort by last modified in descending order so it always picks the most recent file.
• Pause the automation and notify you if no document is found.

This step has to sit before the AI analysis, otherwise ChatGPT has nothing to read.

How does ChatGPT classify the insights?

The ChatGPT step receives the document content from the previous node and returns a structured list of findings. Because a single meeting can produce multiple insights, the output is a list, not a single object.

To handle that, you add a Loop over a list node [06:45]. The loop iterates through each insight and sends it individually to Google Sheets, so every finding becomes its own row.

Why do I need a loop here? Because one meeting can surface five or ten different insights. Without the loop, only the first one would reach the spreadsheet and the rest would be lost.

How do I match AI fields with spreadsheet columns?

The first test run failed because the Google Sheet had no headers and Relay didn't know where to drop each value. The shortcut is asking ChatGPT itself to suggest the right columns based on a sample AI response [08:20].

The resulting structure includes:

• Categoría: high level theme of the insight.
• Subcategoría: more specific tag inside that theme.
• Hallazgo o pregunta: the actual finding or recurring question.
• Recomendación o acción: suggested next step.

Once the headers exist in the sheet, you go back to the Google Sheets node in Relay and map each AI field to its matching column. Skipping this mapping is the most common reason rows arrive empty.

What can I do with the categorized output?

After publishing the flow and running it again, every new transcript dropped into the folder produces clean rows in the spreadsheet, ready for analysis.

From there you can extend the system without rebuilding it. Useful iterations include:

1. Capture the meeting name and date to track insights over time.
2. Add the participant or account name to segment by client profile.
3. Pipe the sheet into a dashboard to visualize patterns across product, marketing, and sales.

The deeper value is using these conversations as a knowledge base to identify growth bets you wouldn't spot from internal brainstorming alone. Real customers, real language, real friction points.

Building AI flows rarely works on the first try, and that's part of the process. The point is that you now have a repeatable pattern to classify qualitative data at scale, and you can adapt it to your own use cases. Tell me on LinkedIn what you built with it.