Introduction

Hello, I'm Yannis Kargiannidis.

I'm a performance marketeer at Yodec.

I don't know how to code at all.

From Guesswork to Voice of the Customer

So I'm going to show you an example of how to turn basically

In our organization, we record calls with customers, always GDPR compliant, and then we transcribe them.

So we get the transcript, the text of a certain

Campaign Underperformance Spurs Investigation

So the idea was, some time ago we had created a campaign and we had a lot of demos, but we didn't have sales.

So I decided to dig deeper into the campaigns and I actually went inside the calls and started checking what the demos were actually like.

Leveraging AI to Mine Objections and Angles

I took the transcript and I put it inside ChatGPT and I sent it to analyze in order to find objections or angles we could proceed to create new ads.

So it will be more relevant to the exact persona.

Usually marketing tries to guess what the user thinks and to create ads based on guessing.

But with AI and with demo calls, you can actually stop guessing and get the actual voice of the customer.

Early Results and the Bigger Idea

So I did that, it had pretty good results.

We increased our sales and then came an idea.

Scaling Insights Across the Organization

What if we could do that in every call of our organization?

So customer success, sales, support, et cetera.

And

That's how the idea came to build an extraction machine.

Building the Extraction Machine

Data Flow: From Call to Spreadsheet

How this works is that we get a transcription from Fathom.

Fathom is an AI tool that records conversations with Google Meet or Teams or whatever.

And then we send this data into NA10 with Zapier.

And after that, NA10 uses ChartGPT to analyze the data and finally organize them inside a Google Sheet.

Live Demo Setup

So we can now do a demo, hopefully.

I will save this tab.

Perfect.

So there's basically this one.

Running the NA10 Workflow

This is the NA10 workflow.

And now we're going to execute it.

It will wait for us to trigger the automation.

The trigger is basically when transcription is happening.

Now we're going to replicate that.

And we will send the trigger.

Trigger, Analyze, Populate Sheets

we will see that it got it, and now it will pass it to ChatGPT in order to analyze it, and then put it in different Google Sheets.

Reviewing the Sheets While It Runs

So while this cooks, let's go to the sheet.

Ah, I haven't said this one.

Wait a second.

Yeah, it got the webhook, and now it's analyzing it with ChatGPT.

So while this is working, let's go to our database.

Perfect.

OK.

Reading the Output: What the Sheets Reveal

here you can see different marketing departments and the product department as google sees at some point we might see sell row 88 filled with what the agent doing so here we see the meeting url the industry and some insights for every different marketing department or product in here i i have

Perfect.

Okay.

Here I have checked some in order for us to see what the insights are.

Turning Transcripts into PPC Assets

So from a transcript, we got a possible PPC ad angle.

So how we can approach the customer in order to convert him.

in a Google Ad.

And we see that this customer wanted the seamless migration from legacy digital signage systems.

This can be an ad copy snippet, as you can see here.

So basically a headline for an ad that says, migrate from legacy signage without the headaches.

This is ready to use.

You can push it to Google Ads.

Creative and Targeting From Real Customer Language

also we see some creative recommendations for this particular one and targeting recommendations which are not guesswork it's the actual customer we had so if we do a filtering here we can better target potential customers inside linkedin google ads etc

Let's go back, still working, okay.

Product Team Insights From Calls

So I will show you another example for the product team.

Here we see that this particular client had some feature requests, some usability feedback, confusion points about the product,

also desired integration for some of them and here we have some product improvement ideas for example offer custom roles as an add-on to lower tiers because this particular customer had a problem with granular roles here you can see many of those things

And that again, if we tag them correctly and create a dashboard, they could be a database for the product to create backlogs.

Signals for Content, Design, and Product Marketing

Also we see for content design, product marketing, we have things like competitor mention.

So here you can see if someone mentioned a competitor of yours, et cetera.

Verifying the Automation Run

Let's go back to NA10.

okay it worked as we can see here if we go now to the transcript we will see a new cell it was 88 i think okay probably or 87 here fixed okay um so i have a question

Q&A: Categorization and Prompting Strategy

How did you distinguish these categories, the design, the producting, the marketing?

What were the criteria?

Okay.

The Prompt: Acting as an Analyst

When I take a transcript, I run it through a prompt.

I can show you.

Mm-hmm.

The prompt is this one and basically I tell him to act like an analyst and get insights based on what he gets and get a certain output.

So I tell him like get an output for an objection based on what you read from the transcript.

Collaborating With Departments to Define Outputs

And now, in order to create the Excel, this is a better, let's say, product.

And you need to talk with each department to basically optimize it and ask them, OK, what do you want as feedback from a transcript?

And then create those rules to have a better segmented and organized info.

Okay.

So you gave him an idea, but not criteria for these categories, right?

What do you mean with criteria?

I'm not the product person.

When you find something like this, then you cross it to the design team.

When you find something like this.

Okay, yeah, yeah, I did that.

Okay.

Routing Insights to the Right Sheets

And it's basically with the sales name.

So if you go here,

you have some ifs, and basically what you tell him is to match this field with a certain Google Sheet field.

So ChargeGPT gives you everything into one big text, like product insights, PPC insights, everything, and then it's up to the N810 system, when you build it, to figure out

where everything goes.

Okay, I'm not sharing this.

Let me say.

Mapping Fields From AI Output to Google Sheets

Yeah, so this one gives us a big file of text and after that we match it with each like sheet, SEO, PPC, design, etc.

based on the field name on the Google Sheet.

Okay.

Department Use Cases and Outcomes

And so here you can see some use cases for different departments, like the PPC can turn buyer phrasing into ads, and we get triggers, angles, and objection of the customer.

And the action output could be headlines, descriptions, targeting, etc.

The SEO, which is very important, actually gets the voice of the customer, so it can write articles that directly talk to the customer with his languages, so it doesn't guess that this customer knows this information.

Content can get newsletter ideas, product marketing can sharpen its positioning, get objections, competitor mentions, etc.

Thank you.

Roadmap and Conclusion

This is a beta product.

After that, we were thinking about doing dashboards for each one of the things.

with the Google Sheet, then connect it to another, like, Looker Studio, and visualize the whole information.

Thanks a lot.