Voice to Value: I Used ChatGPT Voice to Help Build an Automation | Episode 099

Today we're going to use ChatGBT in voice mode to help us automate. Once again, welcome to the Digital Customer Experience podcast with me, Alex Turkovich. So glad you could join us here today and every week as we explore how digital can help enhance the customer and employee experience. My goal is to share what my guests and I have learned over the years so that you can get the insights that you need to evolve your own digital programs.

If you'd like more info, need to get in touch, or sign up for the weekly companion newsletter that has additional articles and resources in it, go to digitalcustomer success.com for now. Let's get started. Hello and welcome.

My name is Alex Turkovich, and this is the Digital CX Podcast, the only show where we talk about digital in CX on a weekly basis, roughly. Um welcome back to this is episode 99. Super exciting because, well, the next one's 100, which some would argue is an arbitrary number, but uh still we gotta celebrate our milestones a little bit here and there. Um, today I have a pretty cool episode for you that is kind of inspired by one of my many workflows.

Um, specifically with regards to voice mode in Chat GPT. Sometimes when I'm on a walk or just doing other things, I like to put Chat GPT in voice mode to help me get some things accomplished. And one of the things I've been doing lately is having it help me work through um partially in my brain, but also get advice on how to automate certain things. And um, I was doing this the other day on a walk.

I was talking, I was chatting with ChatGPT about how to automate a certain process for this podcast, actually. And it dawned on me that um some of you may not be super familiar with that mode. And if you are, awesome. If you aren't, welcome to the world um of talking with AI in real time.

But what I figured I would do is actually have one of these chats live on the podcast uh with you listening in. So the scenario that I'm putting forth uh forward with ChatGPT is I want it to help me create a automation that pulls chat transcripts or or voice call transcripts, customer call transcripts into a database so that our product team can query it for product feedback. Um which is actually a pretty cool use case, right? You you could theoretically I mean you can set up a a chat bot for product teams to use when they want feet like kind of quote unquote live customer feedback on specific modules and things that are not working well with them, things that are working, um, and those kinds of things, right?

So um you'll hear that whole scenario that I lay out with the chat. A couple things to point out. Do listen for how I'm prompting it, especially in the beginning. I'm giving it very specific instructions for what expertise I want it to have.

I'm also explicitly telling it not to make stuff up and do some research on the tools and the capabilities. Um, a lot of times when I do this, I go uh a little bit more into detail, but I didn't I didn't want to present you with like a 20 or 30 minute long chat, right? I wanted to keep it under 15 minutes for you. Um but sometimes I will go further into detail about you know the specifics of setting a certain part of the automation up.

Um and in this case, what I asked it to do at the end was to spit out a PDF kind of guide for how to set that up. And I will link that PDF in the show notes if you do want to um look at it afterwards and see actually what it came up with. Now, I uh you know, sometimes you have to take those things with a little bit of a grain of salt. Uh you know, it it it won't be hyper accurate, especially if you haven't given Chat GPT uh or Claude or Gemini for that matter the specifics of what it is you're doing, what environment you're in, what your role is, who you are, who the company is.

You know, but if you have spent enough time with any of those models or tools, they'll have a general context for who you are and what you do. And so the the results will be a little bit more tailored towards you. Um now I think we all know when we ask, you know, Chat GPT, especially to pump out like a PDF or whatever, it's gonna give you the bare bones, right? It's not gonna give you anything highly formatted or whatever, but um I think Claude especially is making a lot of inroads into making that a little bit better, especially with regards to PowerPoint generation and those kinds of things.

Anyway, I'm rambling. What follows is my roughly 10, 12 minute um conversation that I just had with Chat GPT about setting this thing up. I hope it's useful to you, and I would love your feedback on maybe some things I could have done differently or different ways of, you know, prompting it to get a better result and those kinds of things. I'm always open to feedback, right?

So if you guys are are working on this kind of stuff and doing this on the daily basis, you know, we can all learn from each other in terms of better prompting. But I hope that's insightful and I hope that you can apply this to your own role, right? The end goal is for you to also then do this kind of stuff so that as you're doing the dishes or as you're taking the dog for a walk or as you're driving home from work, whatever it is you do, you might as well be a little productive at the same time.

Now, the trick is after you have these conversations, going back and actually executing on them, right? I've had a lot of these types of conversations with Chat GPT, and then after I get home or whatever, I just kind of forget about it and I don't go back to it, right? So whatever you do, make sure you make give yourself a mental note to go after it. And in fact, one thing that you can do at the end of these, um at the end of these chats is you can have ChatGPT send you a reminder at I don't know, 8 p.

m. that evening or you know nine a.m. the next morning that you need to action on what it is you chatted about, right?

So you can actually have it remind you. Um especially if you're working with calendar integrations and whatnot. But anyway, here's the conversation. I hope you enjoy it.

Okay, so I'm hoping you can help me with a little project, ChatGPT, where I need you to essentially act as a digital automations expert, specifically AI automations expert with a deep knowledge in customer experience best practices. Absolutely. I can put on my AI automations expert hat for this. Just let me know the details of what you're working on and what kind of Bidens or setup you need.

We can dive right in from there. Okay, the other thing I want you to do is to make sure that as we're discussing these things, you're conducting some deep research on best practices and tooling capabilities, specifically with regards to API availability. And then also don't make stuff up. So I want your answers to be grounded in nothing but reality.

Absolutely. I'll keep everything as real and grounded as possible. So as we go along, I'll do the deep research on those best practices and the tooling details, especially around APIs. And of course, I'll stick strictly to what's actually true.

No fictional stuff, just reality. Let's get into it. Okay, sounds good. So my goal is to essentially create an internal chatbot for internal use only, primarily by the product and engineering group.

Um, and this chatbot should be trained on customer experience team call transcripts for the intent of giving product teams a way to get instantaneous product feedback across all of the calls. So the idea would be that a product manager could essentially go to this chat bot, ask a question about maybe a specific product or specific module, maybe a specific frustration, or things that customers are challenged with or things that customers are really exceeding at and be able to get instantaneous feedback based on those call transcripts, but then also to get links to the specific calls where those um where those things took place.

Now, in order to do this, um, I can give you what our tech stack is like, but then I would also like for you to ask me clarifying questions one by one until you have 95% um until you have 90 95% clarity on what my end goal is and what we need to build. Absolutely. That all makes perfect sense. And I love the approach of doing this step by step with clarifying questions.

So let's dive in. First, a quick clarifying question, just so I have a solid foundation. Can you let me know what your current tech stack is, especially the platforms or tools you're using for storing these call transcripts? For example, are they housed in a system like Gong or in a CRM or just in a database?

Once I know that, we can figure out the best ways to integrate and make that chatbot a reality. Call transcripts, all inbound and outbound, are stored in Gong. Um and my initial intent was to perhaps create some sort of export automation into something either like Notion or Airtable or Google Sheets that this chatbot can then reference. And I would love your feedback as we talk on if you think that's the right approach.

Perfect. That gives me a great starting point. So we know we're working with Gong for the call transcripts, and we want to create an automation that exports that data into a platform like Notion, Airtable, or Google Sheets that the chatbot can reference. So here's my next clarifying question to get us closer to that 95% clarity.

Among those options, Notion, Airtable, Google Sheets, do you have a preferred tool that your team is already comfortable with? Or are you looking for a recommendation on which one might be easiest for the chatbot to query and for you to maintain? Let me know your preference and we'll keep going from there. Looking for a little bit of a recommendation, but I think from what I know, Airtable is probably the best suited for that, just because it does play well with integration tools like make.

com or Zapier. And it also is a lot more scalable when it comes to feeding it tons of data and and making it easily organizable. Um that said I'm open to Notion or Google Sheets as well, if those are things you'd recommend over using Airtable. Absolutely.

Airtable is a great choice. It's really flexible. And like you said, it plays nicely with integration tools like make.com or Zapier.

It's also easy to search and organize, which is going to be really helpful for a chatbot that needs to pull in customer feedback details. So I think that's a solid route. Now that we have that piece clarified, let me ask another question to get us closer to that full clarity. Do you have a preference for how the chatbot itself is built or hosted?

For example, are you thinking of using a specific chatbot framework or platform? Or are you open to recommendations on that as well? Let me know. We'll keep refining.

I'm very open to recommendations on the chatbot interface, whether it's something we built from scratch or whether it's an existing tool that we can use. Um preferably something that is easily integratable with Zapier, for instance, or make.com. Absolutely.

That helps a lot. If we want something that's easily integratable and works well with Zapier or Make.com, one strong recommendation would be looking at something like a no-code or low-code chatbot platform. A tool like LANBot or even a more flexible platform like bot press can be really good because they let you integrate with external data sources and connect with webhooks or APIs to something like Airtable.

So I'd recommend using a platform like LANBot for the chatbot interface. It's pretty user-friendly and it can call out to Airtable through Zapier or make.com to fetch those call transcript details. That way, when a product manager asks the chatbot something, it can just pull the relevant info right out of Airtable and even link them back to the original gun call if needed.

Let me know if that sounds good. And if it does, we can talk next to us. I have a question for you about data privacy. If we are to use one of these external chatbot tools, are there data privacy concerns that we need to be careful of in terms of vetting that with this vendor?

And if there are, are there options for maybe native chatbot capabilities within like Zapier or make.com, for instance? Absolutely. And that's a really smart question to ask.

So in terms of data privacy, whenever you're using an external tool, especially one that's handling potentially sensitive customer data from call transcripts, you definitely want to vet that vendor for their data handling policies. Look for things like GDPR compliance, CCPA compliance if you're in California, and generally how they store and secure the data. So that's something you don't have to do a little due diligence on. Now, if you're looking for a more privacy-controlled option, there are indeed some native capabilities.

For example, you can actually build a simple chatbot directly using Zapier's interfaces, or you could use make.com with a webhook and a custom front end that you control. That way all the data stays a bit more within your own ecosystem and you have a little bit more control over privacy. It might take a tiny bit more setup, but it gives you that extra layer of assurance.

So overall, you can definitely do it with a vendor tool as long as you vet them. Or you can go the more native route with Zapier make. Let me know which one feels like a better fit, and we'll keep going from there. Okay.

Since we are already a Zapier customer, we'd probably prefer to build that chatbot interface within Zapier. Um and so the flow now, as I understand it, would be that we would feed the call transcripts from Gong into Airtable and then use Airtable to train the chatbot. Can you confirm that? And are there any other things that you would need clarity on before building out a whole infrastructure document for this?

Yes, absolutely. And I can confirm that that flow sounds exactly right. So essentially you're going to be pulling those call transcripts from Bung into Airtable. Airtable is going to be your database that the Zapier based chatbot will query.

And then Zapier will act as the logic layer to let the chatbot retrieve the right information and show it to your product team. So that's a solid plan. I think we're pretty close to that 95% clarity. The only other thing I might want to know, just to finalize everything, is if you have any specific requirements for how the chatbot should respond.

For example, do you want it to give a summary of customer sentiment or just list out the call or offer a combination of both? Once we know that, we can put together a nice infrastructure document and you'll be good to go. I'd like the chatbot to respond pretty directly with specific feedback based on the query from the product team, which also means that I would ideally want to strip some of the call transcripts of any kind of small talk or personal details and things like that as well.

So I would like your input on how we could potentially do that in an automated fashion because it's unrealistic for us to go and scrape hundreds and hundreds of call transcripts. Absolutely. That all makes perfect sense. And I love the focus on keeping it concise, but also allowing deep links for more context.

So let's break that down into the final pieces. First, on the automated stripping of small talk or personal details. One approach is to use a natural language processing step in your automation. For example, you could have a step in Zapier that runs the transcript through an AI text cleaner.

There are some pre-built tools or even just a simple custom GPT model that can identify and remove filler conversation. That way you're only keeping the core feedback. We can definitely outline that in the infrastructure doc. Second, on the output side, it's totally doable to have the chatbot give a concise summary and then provide a link directly to the relevant call in Gong.

That way, product managers can click through if they want to hear the full context. We'll just set up the chatbot responses so they have a short summary plus a listen to the full call link. With those final pieces in place, I think we have everything we need. I'll go ahead and draft out an infrastructure document that covers the entire flow, pulling data from Gong, cleaning it up, storing it in Airtable, and then using Zapier to power the chatbot interface.

If that all sounds good, we'll get that ready for you. That sounds good. When you create this document, um ideally please um provide it in PDF format. And I would like for you to be as specific as possible when it comes to the setup of each step.

So for example, in Airtable, I would want you to be very specific in terms of which columns we need to set up for the data being pulled. For instance, a link to, you know, um the call transcript itself and those kinds of things. So be very specific in those steps and essentially treat it as if you're teaching a beginner how to set up this kind of automation with each tool. Absolutely.

I'll make it as beginner-friendly and detailed as possible. I'll include all the specifics for each step, like how to set up your air table columns, where to link the call transcripts, and how to configure each part of the Zapier flow so that it's all crystal clear and a nice PDF for you. So I'll put that together and we'll have you all set up big go. Great.

Thank you. You're very welcome. I'm glad we could get everything ironed out. I'll get that detailed PDF ready for you with all the step-by-step instructions.

And if you have any more questions along the way, just let me know. Happy to help. Okay, so that is essentially the conversation I had with it. Again, I will put the outputs, that PDF output of what it came back with in the show notes below so that if you want to go look at it, you definitely can.

Uh again, would love your feedback and would love to hear from you what kinds of things you're chatting with ChatGPT about on a regular basis. But that'll do it for this episode. I want to thank you for joining me today and uh hope your week ahead is an awesome one. We'll see you next week.

Thank you for joining me for this episode of the Digital CX Podcast. If you like what we're doing, uh consider leaving us a review on your podcast platform of choice. If you're watching on YouTube, leave a comment down below. It really helps us to grow and provide value to a broader audience and get more information about the show and some of the other things that we're doing at digitalcustomer success.

com. I'm Alex Turkovich. Thanks so much for listening, and we'll talk to you next week.