A Product Market Fit Show | Startup Podcast for Founders
00:25:06 · Sean McCarthy
[00:25:06] S0: Yeah, exactly. So they were already going to do it. They can hit record and walk through, and as they're walking through, of course, they can verbally articulate anything that they want. From there, once we have it, we'll process-mine it. So we'll actually make sure that it's the fastest route to get from A to Z. What we see is that historically, maybe they're pulling down an Excel spreadsheet, doing some sort of manipulation, and uploading it somewhere else. Where obviously they don't need to do that anymore. So we'll make sure that it's efficient. After it's built, it'll go to staging, and then they can actually push it to production. Really where that pulls ahead is, let's say that there's two different things. Number one, let's say the use case is in flight, but then there's an edge use case, and they're like, oh, I totally forgot to tell BackOps that when customer A, B, C has this problem, we do D, E, F. They can just record a two minute of this is how you handle this specific case, right? Then we can very quickly adapt the agent and then the second part is that on that enterprise-grade side. Now if it's that claim and finance wants to track it, they can see what the current process is. So there's no more silos and they can say, hey, well, every time that happens, we have to do our process. So can we just attach our process on? And the answer is yes, they can just attach that on via the process center. And so that was a huge unlock for us to also get a wish list of customer wants after the fact. Because they're just kind of, in the example that you said, going through their day and now they're just recording a bunch of these use cases, right? And wanting to give those over to us.
00:26:28 · Pablo Srugo
[00:26:28] S2: You love this show, you don't want to miss the next episode, why would you? So hit that follow button. Trust me, it's in your own best interest. One of the things I'm hearing a lot about right now is you got to think about building AI, like you're building a worker, right? The same way you would train and onboard someone is the same way you want to onboard AI. And this is very much in line with that, right? Which is, if you hire a new person, you're going to show them. You're probably not going to send them, here's like a hundred-page SOP, you know, go read it. You're probably going to be like, okay, this is the thing you're going to do today. Here's how you do one, go do it for the rest of the day and let me know when you run into problems, and I'll tell you. That's basically how I train someone. The one question I have is why have this friction of like hit record, the natural thing would be like you're just always on recording and you'd like prompt them, and you're like, hey, I just saw you do something. It seems annoying. Do you want that to be a workflow? What's the thinking there?
00:27:19 · Sean McCarthy, Pablo Srugo
[00:27:19] S0: Yeah, I mean, you're seeing a lot of our roadmap already. Already today, we're pairing the recordings with ticket data and email data to also measure some of those things. The always-on recording is where a lot of this process center goes. However, we're still in the early days, I think, of some of the comfortability of AI watching what they're doing and we found that most customers have asked that we started this way. And eventually would be okay with more of a screen watching type scenario. But for right now, have felt the vast majority think it's a little too much too soon. [00:27:52] S2: You know, it's an important thing, especially in enterprise. The security and data piece is obviously huge. And I've seen this before, where if your initial product can be simple from a security perspective, and simple usually means something always-on recording is definitely not simple. Because the access that you have is uncontrolled. Also, if you don't touch customer data, like if you can have something that somehow sits on the side, and you sometimes can't. But if you can, you're going to get in way faster and if you can at least add some controls, that's also going to help. And then, you know, you work your way to full visibility on everything. It might sound like the best place to end up, but if you try to have that as a line. You just won't get into these enterprises to begin with.
00:28:33 · Sean McCarthy
[00:28:33] S0: Totally agree, and there's a reason the system that's built today works. Because they're picking and choosing what systems access, and what information we're getting, and what workflows we're getting. And obviously with an always on scenario, either a human would need to toggle that off if they're going to do something sensitive and remember to do so. Or you have always on, you might have recorded something that didn't. I think that the technology that we're building also still has some ways to improve to make sure that those safeguards are in place at the enterprise level. But I do think that something similar will become the norm as it relates to gathering inefficiencies within a large organization and being able to have better visibility into what those tasks, and workflows are.
00:29:14 · Pablo Srugo
[00:29:14] S2: You know, another thing that you mentioned there is this idea that nobody wants to have ten AI vendors. In the old world, like five years ago in the SaaS world. People would talk about nobody wants to have ten, fifty SaaS apps, you'd rather have all in one. It sounded right, but it wasn't true, except for the enterprises that bought Microsoft. In most cases, you did have this world of best-of-breed apps. Everybody's in a silo. They're great at what they do and you end up with a large enterprise would have fifty of these apps. And that's just the way the world works. And so many different people try to come in and stitch them together. It's not a thing, again, unless you're Google or Microsoft. In this world of AI, that seems to have changed. You know, we had, for example, Serval on and Jake was telling me about how he's rebuilding ServiceNow. This entire behemoth of a twenty-year-old company, instead of building on top. In the past, you would say, okay, take ServiceNow, you're not going to move that. I'm going to build AI on top and yeah, it's a dependency, that's annoying, but what else am I going to do? And now you're like, I'm going to rebuild ServiceNow. In your case, something similar where like, you know, you started with this Slack bot and maybe, you could have verticalized and stayed, okay, well, we're just going to own operations, somebody will do finance later and yet what customers are demanding is, no, you need to be across. Maybe tell me more about what you're feeling on the field, what you're hearing from customers around this concept of one platform that can kind of do it all.
00:30:26 · Sean McCarthy
[00:30:26] S0: Yeah, I'd be happy to and I think there's probably a lot of synergy with the way that Jake looks at it as well. I think what we're hearing from our customers is if this thing is going to be really effective, it has to take the end-to-end of that entire workflow and so if I have to insert a human in the beginning to take the phone call and understand what the tracking ID is, and whatever the problem is, and then at the end. Finance or whatever department has to go do all of these additional actions, it's not that beneficial. You're actually selling or solving for a very small sliver of minutia that is not that impactful across the company, and that means that I'm going to have to go get either more humans or other vendors to service the rest of that. So what they've been adamant on is like, not that you necessarily have to, let's say it's a forty-step process. Not that you have to take the entire forty-step process across five business units right away. But the ability to maybe start from the beginning and build a classification system, and then kind of take that piece by piece is important. Because they are trying to establish relationships with AI vendors that they trust that are taking more and more as it relates to not only complex workflows, but maybe there's opportunities like the claims where only twenty percent are getting filed. They're looking to hand those off to a reliable person that is going to take the end-to-end and so for us, that has been something that we've really indexed on and making sure that is it possible for us to actually take that entire workflow. And if it's not, we'll tell the customer too and say, hey, there's actually a dependability here or there's something that might block this. But that's been super impactful and I think that that's really where it goes is all of these point solutions will become a commoditized layer here very quickly. And you have to think the simple ones are going to be able to be solved with things like cloud code, and we're seeing a lot of that already today. And so, I think that where a lot of our moat and where the customers look at us as a more of a one-stop shop is the fact that we can go deep, and we don't need API access as an example. And we can call and email these carriers, and gather documentation, and do a lot of the things that before they had to have humans do.