How this CEO turned 25,000 hours of sales calls into a self-learning go-to-market engine | Matt Britton (Suzy)

0:00With my company, my sales team was consistently telling me that they just didn't know how to find anything. They didn't know how to find what customers were interested in. You had a bunch of salespeople. They said, I need more information to serve our customers better. You realized you had 25 ,000 hours or something of reported customer calls, which are the perfect source of truth for how customers want to be interacted with. You're going to show us a zap now that takes a single recording and does a bunch of stuff. So basically I needed to figure out, well, can I create a feed for Zapier? So he knew the call ID of each new call as occurred.

0:37So the first step is essentially a trigger where a new call comes in. It'll basically scrape the information from Gong. And one of the things Gong will give you is that call ID. So that appended to the URL essentially is all I needed to give Browse to be able to go to that URL and save it a century. It's great. The entire transcript. It wasn't connected. I had to kind of hack it together. I love a CEO that knows how to build it. I love a CEO who knows that no problem is not solvable.

1:10Welcome back to How I AI. I'm Clara Vau, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, we have Matt Britton, CEO of Suzy. Now, normally we show two or three workflows, but today Matt's going to show off the one mega workflow that rules it all at his company. He's going to show you how to take a single asset and turn it into tons of go-to-market goodness, from emails to customers, enriched data sources, and even marketing assets that can be used to source more prospects that are going to be successful with your product. Let's get to it. This episode is brought to you by Brex.

1:48If you're listening to this show, you already know AI is changing how we work in real practical ways. Brex is bringing that same power to finance. Brex is the intelligent finance platform built for founders. With autonomous agents running in the background, your finance stack basically runs itself. Cards are issues, expenses are filed, and fraud is stopped in real time without you having to think about it. Add Brex's banking solution with a high yield treasury account. And you've got a system that helps you spend smarter, move faster and scale with confidence. One in three startups in the US already runs on Brex.

2:29You can too at brex.com slash how I AI. Matt, thanks for coming on how I AI. I'm excited because as I was saying before we started the show, we get vibe coders left and right. And I I know we're going to talk about some custom software that you built, but we just do not get enough on the go-to-market and marketing side of AI automation. So I'm really excited to show what you have to share. So really appreciate you joining today. It's time to be here. So you and I both really love Zapier. And I have to ask, even before the age of AI, was this a tool that you relied on? Why has this specific software become kind of the backbone of so many of your AI-based automations?

3:16So I've always been fairly technical, but I've never been a coder. I sold the first ads ever on Facebook directly to Mark Zuckerberg and Edward Severin in 2005. I bought some of the first Google keywords ever to exist right when I started my business in 2002, my first ad agency. So I've always loved sort of ad tech and getting my knowledge and understanding how these tools work. but at the same time, I've never been an engineer. And as I wanted to get more sophisticated in the tools and solutions I've built for various companies that I've run, I've needed to not just use one tool, like AdWords, but multiple tools to stitch things together to be more efficient.

3:55And I was turned on to Zapier, like most other people, just to a Google search. And I think I wanted to connect leads that were coming in through my website to some type of flow where it automatically emailed the person who signed up. and then I just kind of start to dive into it. But to your point, Claire, it wasn't until Zapier integrated AI when kind of my mind just became blown in terms of what's possible. Okay, so you're going to show us how you take a single asset, and I won't spoil what it is, and turn it into basically a full suite of activities across your marketing, sales, internal company work.

4:29So why don't you pull that up and show us what you built? Before I pull it up, I guess I should say that I think it's all about figuring out what problem that you want to solve. And I think with AI in general, people get so overwhelmed with just the amount of tools and the rate of change that they just find themselves kind of playing around with all these tools, trying to get to the point where they feel like they're comfortable in understanding them, but at the same time, they're not really moving their business forward. And I think the reason that's the case is people don't ever take a step back and think like, what is the core problem I need to solve in my business?

5:02Like what's holding me back from growing faster than I want to, what's getting in my way or what's an opportunity I know is there, but I'm not able to take advantage of it. And with my company, what I was hearing over and over again was my sales team was consistently telling me that they just didn't know how to find anything. They didn't know how to find what customers were interested in. They didn't know how to find, how to speak to people of a certain industry or a certain title in terms of identifying use cases. So just so many unknowns. And so once I understood or put my finger on that problem, I just became very sort of tunnel visioned.

5:43And I was determined to figure out how I can build solutions that can aid my sales and customer success team to be more in the know. So once you've actually identified the problem, the next step is figuring out what data can help you seize that opportunity. And in the case of, you know, understanding our customers and getting that information, it just so happens that since the pandemic, when our company went remote, we've been using this tool called Gong that's essentially attached to Zoom calls that records every single call that we have. So it says this call is being recorded for quality assurance purposes.

6:19And I always knew we had obviously Zoom and I knew that we had Gong. But what I didn't know is that their transcripts were amazing and that we actually had 25 ,000 hours of call transcripts that had been a mess over the last five years. And if you think about understanding information about your customers and your business, there's no better source of truth. So we have since built an entire operating system around this information, not just the historical information, but a variety of different workflows that happen with each new call that occurs. Because it's not just about understanding what's happened in the past, but it's also being able to be highly responsive to what's going on in the present.

6:56So the first thing I'm going to show today is an automation that we have created based upon calls our teams have, either our sales team or our customer success team. And essentially what happens is as soon as that call is over, a series of events happen with that individual transcript. We also do things sort of at large with the aggregate transcripts, if that makes sense. But right now I'm going to show you what happens kind of like real time once a call is completed. Great. So while you pull that up, just to recap for our listeners, you had a bunch of salespeople. They said, I don't know how to find the information that I need.

7:34I don't know how to generate the information I need. I need more information to serve our customers better. You realize you had, and I'm correct me if I'm around 25 ,000 hours or something of recorded customer calls, which are the perfect source of truth for how customers want to be interacted with. And you decided that was going to be the core context for a lot of these actions inside your company. And then you're going to show us a zap now that takes a single recording and does a bunch of stuff. I got a preview of this and it does a lot of things. I tried to give AI to my engineering team to figure stuff like this out.

8:11And it just became overwhelming to them, even integrating the product. And what's been helpful for me was first building things on my own. And I'm not technical enough to be able to build on top of our software product. So the tools like the one that I'm going to show you today was a great way for me to be able to dive into the power of AI because it was on the edges of the, it wasn't the product was sort of on the edges of how we operated. And through that, though, I became far more adept at AI. And now I'm very much involved in our product itself. So often people struggle to find a way in, and there's lots of different ways.

8:48And one way is actually building something for yourself personally, or building something for the marketing organization or somewhere else. And then through that process, you really start to get it. And then you can start to be more, you know, proficient in AI in much more substantive ways within the business. Yeah. And I want all the other CEOs and executives watching this podcast to listen to exactly what you said, because it is not sufficient to instruct your engineers to build AI. You'll go nowhere. Yeah, no, you'll go nowhere. And I've said this a lot. But this is a moment for actual hard skill building in leaders, which is you actually have accessible skills to build in using AI, building AI tools, using these sort of like no code versions of tools to really upskill yourself on the capability.

9:31And that's going to make you a much more relevant leader, much more. Yeah, you're opening up the hood. It's like if you bring your car in and you don't know anything about fixing a car and they tell you$4 ,000 to fix a transmission, you're going to say, okay, because you need your transmission fixed, right? But if you actually just open up the hood and you understand how transmission works, even if you're not a mechanic, maybe you can say, well, it really shouldn't cost$4 ,000 to fix this. It really costs close to$2 ,000. And I think that's sort of the same analogy when it comes to AI. So the first step is building what I call a trigger automation.

10:07And this trigger automation essentially comes from a tool that we've created that we use called Browse AI. So this is Browse AI. And essentially what Browse AI does is it runs like a script where essentially scrapes information from GANL calls. So what you see here is a URL string. You need a URL string in order to identify a call transcript. as it comes in. And GOM didn't have an easy way to do this. So I basically started to bring up a bunch of call transcripts. And what I started to see is they all kind of start the same way and they just end it with this call ID. So the only thing different from call to call was this call ID.

10:47So basically I needed to figure out, how can I create a feed for Zapier? so new, the call ID of each new call as it kind of occurred. So the first step is essentially a trigger where a new call comes in, right? And then what happens is when the new call comes in, what it will do is it'll basically scrape the information from Gong and one of the things Gong will give you is that call ID. So I'm able to actually see the call ID. So if I click here and I scroll over, you'll actually see that there's a call ID that I can identify here, which is right here. And so that appended to the URL essentially is all I needed to give browse to be able to go to that URL and essentially scrape the call transcript.

11:44So it wasn't connected. I had to kind of hack it together. So if you'll see here, it basically knows what to run just based upon what's brought in. And then it will go to this page, which I will show you here, which actually is where the transcript is. And it's able to essentially scrape the entire transcript. So this is the raw transcript that's coming from the GAN calls by Browse AI going to that GAN page and just getting this information. But I had that initiated. So that first step essentially initiates the scrape. And then when the scrape is completed, it starts my next automation. Yeah. Yeah.

12:19And so just to call this out for folks that are trying to build their own thing, it's okay if your tool itself does not expose the data you want. In this age now, you can usually use another tool or an alternative. There's always a way. Yeah. Or an LLM to really pull the data you need out of any system. Yeah. And I could have given up, Claire, like at that point, that probably one step took me the longest. And if I never would have gotten past that step, and I think a lot of people would probably have given up at that step. But after I got over that hurdle, then everything else became so much easier.

12:54And it's really like an analogy for life, like building something like this. And there are other stumbles I've had along the way in building things, but you just have to know that there's a way. And just because a tool doesn't do it doesn't mean it can't be done. And this, like in the rear view mirror, seems obvious. And now if I had a similar challenge, I'd be able to do it right away. Because what will happen is every time you solve a problem such as that, the next time you need to build something, you'll have all these sort of ideas and like hacks in your tool chest, so to speak. And then now I'm at a point where it's like nothing you can tell me to build that I wouldn't know how to build because I just know how all these little things can be solved for.

13:32And you learn coding along the way, like along the way, you learn what JSON means and all these things by having your hands on it and creating the automations. A hundred percent. What I was going to say is this is a CEO that I love. I love a CEO that knows how to build it. I love a CEO who knows that like no problem is not solvable. And I think just even getting hands-on with some of these no-code tools and these AI tools just gives you a little bit more context to be bolder about what you build. Okay, so you have... That's right. So the task is done, right? The call is done. And so this next trigger is true when browse AI successfully scrapes a call transcript.

14:12And the first thing it will do is obviously it'll trigger it. And you'll see here, it'll give me the entire transcript of the call. And that's basically now it's like, OK, now I'm in business right now. I have everything I need. And there's a bunch of other stuff in that gun call transcript that I use to do database lookups throughout that we'll kind of get into. I'll have a delay of about two minutes before pulling in data like this, just because I want to make sure that all the data is brought in, the scrape is done. And I'm just, you're prone to errors, especially if you're running a lot of paths quickly, if you don't put in a delay.

14:45So always like a one or two minute delay is a buffer just to let the system catch up so it doesn't break. So that's kind of self-explanatory. The next thing I do is I run a format or I'm basically removing all the HTML from the transcript. So when you scrape sometimes, it'll pull in the HTML. And I don't want that. I just actually want the actual text. So I run a formatter step where I'm removing all that. I'm pulling out anything I need to that might confuse the analysis. So I'm just essentially getting the raw text. Then what I do is I start to enrich the data with other information besides just the Gong transcript because I have the Gong transcript.

15:23But one of the things I knew I wanted to build was after the call is done, I wanted to be able to tell the salesperson that was on that call what transpired. I wanted to make it easy for them to write a follow-up email. I wanted to be able to identify who their supervisor was, right? But that wasn't directly pulled in through BOM. However, we have other data sources that essentially can connect that information. So we have a Google Sheet here. For example, if you look up this ID, it connects the ID to the brand and the brand to the user, which is a whole separate workflow that we created. So it can kind of connect the dots.

16:01Cause when you're running an automation, you're not always going to get the data from the trigger. Sometimes you have to round it out. And the way you round it out is using things like lookups on Google sheets. So you're pulling everything in. It's almost like you're going down a path, you're on a hiking trail and you want to be able to pull the supplies you need along the way before you get to the destination. And when I started, I had a backpack, but the backpack didn't have water in it. And now I have water, right? Because I grabbed it from here. And you're kind of going along a journey. And I personally, one reason why I love Zapier versus other tools is the way my mind thinks is in a very sequential framework where there's other platforms like NA and, you know, or BotPress, where it's basically like, looks like almost like an octopus, how it's branching out.

16:41I just have a hard time thinking that way. Now, over time, I've had to, because I'm basically describing the difference between automation and agents because agents are not deterministic. Agents have different ways. And my brain has struggled with understanding agents and I'm finally getting there. But basically the progression I see people having to take in AI is you start with using AI as a tool, you know, chat, give me a recipe for lasagna. Then it's okay. Automation is what we're talking about now. And then you get into the world of agents where it's not just always going from step one to two to three.

17:13It might go from step one to three to eight based on what you're trying to accomplish. Well, one tip for you or one tip for the listeners here I found is we'll go through this whole thing. And a good exercise I found is taking a sequential step based automation and trying to use, for example, Zapier agents and just describe that automation in natural which in steps and see how close you can get. Even that replication across modalities can be a good way to test your exercise. Yeah, just test it out. 100%. Yeah. And it's looking up the information. So it's able to basically grab the information.

17:47And then after I feel like I've had all the information, the next thing I'm going to do is this where I'm starting to pull in the LLMs. And an important part here is first and foremost, knowing what LLMs to use. And one thing I've had a hard time with is actually just, we have so many automations now. I think we can do a better job of the organizational design behind it. Because what happens, I built so many things and I don't always do proper handoffs. So for example, here, it should always say use the latest stable version, but it didn't, right? So I'm going to change it now here live on the spot because I want to be using the latest version.

18:22You also want to make sure you're using the best model. I still think GBP4 Turbo is probably a good model for this, but you could see in the platform Zapier, there are multiple different versions that you can choose from based upon, and obviously they all lead up different amounts of coins. But it's pretty incredible in terms of all the models. Now with GPT-5, it's supposed to be able to choose for you, but it's unclear to me how that works in the context of an API. And for some reason, still in Zapier, you're still able to choose. And I spend a lot of time testing. I'll go in the chat GPT and test the sample input in a variety of different models to make sure it's like whatever's the best output, the quickest is what I'll tend to use.

19:02So you're still losing it using classic GPT-4 Turbo, a good old classic classic favorite. Right. AI is supposed to make work easier, but I've been there. Weeks of setup, endless back and forth with engineering, and yet another tool the team never really adopts. That's why I use Zapier's AI orchestration platform. It connects with nearly 8 ,000 apps, so I can finally put AI to work without the drama, without the delays, and without pulling engineering in every time I want to automate something. With Zapier, you can roll out AI-powered workflows that do real work across your whole company in days, not weeks.

19:36I use Zapier every single day. It automatically responds to leads with enriched personalized data. It checks my calendar weekly and offers smarter ways to manage my time. And it even drafts emails for every new request that lands in my inbox. All of that running quietly in the background so I can focus on the work that matters. And Zapier's built for scale. With enterprise-grade security, compliance, and governance, it's trusted by teams at Dropbox, Airbnb, Opendoor, and thousands more. Go to try.zapier.com slash howiai to learn more about how Zapier can bring the power of AI orchestration to your entire org.

20:14Let's talk a little bit about this prompt. So tell me what first kind of summarization exercises you want to do here. Yeah, so basically here, and the reason I can use a model like a GPT-4 is, and part of it, again, is just keeping up with continuing to update the models, but I don't fix things if they're not broken. So this particular Zap works perfectly for us, and it gives us everything we need. And we don't need more rigor in analysis here because it's just some core things that we want to identify. So I'd rather not spend the extra money and even go through it. But at a certain point, if it didn't work, I would look at models.

20:50before they're going fast enough. What's interesting is that older models tend to work faster and faster and faster over time, and they actually error out less. And sometimes the older models get updated as a new models update as well. So it's not like you're driving an 84 Chevy, so to speak. So here, this is a key step. This is called Core Summary Generator. And what this is asked to do is analyze the customer success call transcript between Susie and our client to gauge the health of customer relationships and identify improvement areas. Start summaries with the customer's company name, key participants, and then kind of going through it, ask for the key stakeholders.

21:26And then it gives a call overview. We have to describe the call's purpose, the main topics discussed, and the outcome. Exclude small talk. And then I have a variety of different instructions. Assess the overall customer sentiment. Noting any frustrations or concern. Provide a sentiment score from 1 to 10. Where 10 reflects high satisfaction and 1 indicates potential discontinuation of our services. This is the key thing because it's allowed us to quantify customer sentiment over time. And we actually benchmarked this against actual churn. And it's been highly predictive in terms of if you take the average sentiment score of customer calls over the past year, it's a huge predictor of if the customer is not just going to churn or are they going to upsell if they're really happy.

22:07Also, one great thing the customer successfully did on the call, kind of identify that. And what are some things that they actually could have done better? And then list the next steps. So this is basically just an overarching prompt where it'll take a transcript and identify all this information for me. And then that content I can do a variety of different things with. But it's a huge part of the overall app. So one of the things I want to call out here as I was reading your prompt is in an ideal world, all your best CSMs are doing this after every call in a perfect way with great objective self-evaluation, all this kind of stuff.

22:45And the reality is we're all so busy that, you know, your customer success folks, your sales folks are probably going meeting to meeting, meeting and at the end of the day trying to figure out their notes and put little things in. And I just think what's nice about this is you can make the customer success or account manager's job a lot easier and let them be exceptional at their job by automating some of the work that they would do. And so I think it's a really good hygiene step for anybody to think, you know, after I'm coming out of a meeting, if I were to do my best job possible, what are the five things I would do coming out of each meeting?

23:21And then just automate that for yourself. And then you know that every time you're going to be doing that. Yep. So the bunch of other steps I have, and I'm not going to go through all of them because there's a ton of them. But basically, it looks up the user on Slack. So I understand the user meaning our employee on Slack. It identifies the people who aren't from our companies who can kind of exclude them. It's able to find the user here. So I use Slack as a lookup sometimes because our company's entire directories on Slack. So if I'm trying to get someone's email address in an automated fashion and I have their name, I can actually use Slack as a lookup tool in Zap without even actually posting anything to Slack.

23:59So sometimes these tools actually can be used for other purposes. That's not their core purpose. And then basically, one of the main things I do from this is I send a channel message. So basically, after the call is done, you can see new customer success caller as the account, the opportunity, the leader of the call from our company, the date of the call. And it basically has that summary that gets sent out. So we have a channel that's a constant feed that I, obviously, the CEO, I'm very attuned to, and I'll share it right now, where basically every time a customer call is done, it just pops up on Slack.

24:36And I'm able to really, you know, we have 300 employees at our company, and I'm really able to get a sense of the kind of pulse of the company, what customers care about, just based upon looking at something like this. And it's, you know, that alone, if that was the only invention that came out of a high, it would be pretty incredible if you think about it. And this is just like one of many things that we do. So I'm going to pull up and slap right now. And you can see here, this is the sample call. And it shows who the key stakeholders were, what the call attempted to establish, what was the sentiment score, got an eight, right?