Subscribe
Share
Share
Embed
Enterprise leaders discuss how fast AND accurate document intake enables them to make data-driven decisions, increase capacity, and drive top line revenue.
Welcome to Unstructured Unlocked.
Christopher Wells:A podcast where listeners discover how enterprise leaders are confidently automating document intake and accelerating their workflows to increase capacity and drive top line revenue.
Michelle Gouveia:I'm co host Michelle Gouvea.
Christopher Wells:And I'm co host Chris Wells. Welcome to the podcast. Welcome to another episode of Unstructured Unlocked. I'm your co host, Chris Wells.
Michelle Gouveia:I'm co host, Michelle Govet.
Christopher Wells:And today, we're joined by CEO of Tribble, Sunil Rao. Sunil, how are you?
Sunil Rao:I'm doing great. Thank you for having me.
Michelle Gouveia:Of
Christopher Wells:course. Yeah. I'm excited about this one.
Michelle Gouveia:Yeah.
Christopher Wells:Why don't you tell us a little about about who you are and what it is that you do over at Triple?
Sunil Rao:Great. So who am I? Sunil, nice to meet you. And I, I'm the CEO of Triple. We're a company that started a year ago in April, last year, 2023.
Sunil Rao:And, the thesis for the company has been, how do we automate the function within companies that ends up being the bottleneck in a lot of sales cycles when it comes to technical sales. So for those of you familiar with b two b SaaS companies, a lot of the time, you have this role called the sales engineer. And my cofounder and I, Ray Ray Shipley, we were SEs in a previous life, and we've kind of experienced the pains of being this this role that's always in demand, but there are never enough people in that role within the company. And that's because they usually are the ones that really understand your product very, very well, and they're the ones that don't scale very well internally. So the the thesis was, hey.
Sunil Rao:Can we scale that type of function at most companies? And if we look at the roles that those people have and the kinds of jobs that they do, you can kind of look at a lot of the tasks that they complete within sales cycles and start seeing, hey. Like, what what parts of this are automatable? What parts of this should be automated? So that very much was kind of like the opening act, if you will, of, like, let's automate this role.
Sunil Rao:And then we very much narrowed the focus over the last course of the year, to focus in on very specific workflows that these people participate in, like RFPs and information security questionnaires, which typically inundate these teams across these companies with with this workload that comes in that's beyond what they typically do, which is helping salespeople close deals. So that's Tribble in a nutshell, and there's a much longer version of that, but I'm sure we can talk about it as we discuss.
Christopher Wells:Sounds great. Michelle, you ever see any, really massive RFPs or security questionnaires in the insurance space?
Michelle Gouveia:We don't use paper questionnaires or anything like that. Fully digital now in the insurance space.
Christopher Wells:Yeah. I'm sure. Yeah.
Michelle Gouveia:Yeah. There there are so many places where I think something like trouble could add efficiency. In there. So I'm so like looking forward to hearing more of like how you're thinking about the business where you're focused on now and then, jumping into some of the insurance use cases that sound very similar to what you're doing or maybe even that you have, identified yourself. Yeah.
Michelle Gouveia:The insurance industry is ready and eager to have, I think, more, trustworthy, automation solutions.
Sunil Rao:You know, it's funny. I I had a meeting, Michelle, just a couple of days ago with, with an insurance company, pretty pretty large insurer, and the framework for the conversation was very much this. It's like, this is probably a high leverage generative AI use case you can tackle with relatively low risk. And I think that's what's top of mind for a lot of execs right now because there's a lot of uncertainty in how you take this tech and put it into production, and there's all kinds of discussions. And, I mean, when you talk to tech companies, they're typically like you guys.
Sunil Rao:Like, you're typically at the forefront of taking this tech and making it part of your day to day processes. But, of course, as we know, insurance companies and other large verticals that it takes longer for this tech to diffuse into, they tend to be a little bit more methodical about how to incorporate this into a core business process. Like, you don't want the core business shutting down because you've taken a super risky bet. So in the context of this discussion, it was just, hey. You're you probably have salespeople.
Sunil Rao:They're probably spending a disproportionate amount of time juggling Word documents or Google Docs that are heavy in prose, understanding things and creating these output formats like proposals? Can we drive some efficiency there so they can spend more time selling? And then that becomes the okay. That looks like a pretty good area to start exploring this technology and seeing where it goes. So that that's kind of one of the many conversations, Michelle, I'd say.
Sunil Rao:But squarely in insurance, I think it's a very high leverage use case. At least that's what we're hearing from our customers.
Christopher Wells:That's exciting. So I I wanna I wanna hear you talk a little bit, Sonny, about sort of state of the union when it comes to generative generative AI, especially with these big enterprise companies. It sounds like you, in a similar fashion to myself, I see a lot of excitement and people asking questions and everyone's like, hey, how do you do RAG? And how should we do RAG? And all of these kinds of these types of things.
Christopher Wells:What are you seeing in turn you talked about that diffusion through the organization. How how far has Gen AI penetrated? Or is it still we're it's sort of like we're dilly dallying with it. We're not really doing anything. What are you seeing?
Sunil Rao:You know, it's it's, it's something that came up. I I was I was in San Francisco yesterday, and and I had a coffee with the VC, and we were just chatting about the space that we're in. And, he'd mentioned that they put together some sort of a survey where they highlighted the top 100 use cases that are being used actively today across industries for generative AI. And, he has yet to still send me that report. I'll share it with you once I get it.
Sunil Rao:But one of the interesting things he mentioned to me was like, actually in the top ten creative use cases didn't pop up, which was very counterintuitive to me because I figured everyone's using Midjourney in like, not production, but, like, you know, some variant of DALL E or Midjourney to generate. I can tell you right now, all of the decks that we present, you can almost, like, immediately tell. It's, like, generated imagery. It's just such such a low value
Christopher Wells:weird fingers in all of your decks.
Sunil Rao:Exactly. Well, yeah. Not I think those were gang signs, but whatever whatever that might be. Yes. Weird figures, indeed.
Sunil Rao:And and it was interesting to me that that that didn't make its way to the top. And I I do have to dig into the data because it might be that it's, like, companies that offer that as a service aren't as prevalent within production in these orgs. Maybe that's the lens with which, this was looked at because it's b to b. But the net of it is, like, sales tech is is permeating very quickly. It sounds like people a lot of people are taking risks on implementing software.
Sunil Rao:Like, I don't know how I don't know about you, but my LinkedIn InMail is it is in it is in a in a situation right now. So it's like I look at that, and I look at the amount of email that we're getting that's becoming, you know it it is a little bit more personalized than it was a year or 2 years ago, but it's very clear that it's like picking random things from my work history and cobbling it together with some other nonsensical things. And so that's happening at scale, and and you're seeing a lot of smaller companies, like, build tech like this and scale up to, like, 1,000,000, 2,000,000 in RR very ARR very quickly. So that tells me that this sales tech is being adopted and showing up in the enterprise, faster than some of the other areas. But that's that's at least what I'm seeing preliminarily.
Sunil Rao:I think in our own experience, it's it's a little bit more involved to validate whether or not this is something they're willing to put into the business process. There's a lot of questions around security, a lot of CISO conversations. Where is this data going? Is it being sent to some third party API? How are you thinking about obfuscation?
Sunil Rao:So top level, those are some of the considerations I'm seeing.
Michelle Gouveia:So a question for you on that because I also ignore some of my LinkedIn and Mail now because I just know it's just I just need to. And then also a ton of things coming into email that you could very clearly tell the same thing as what you're experiencing, right? They they they swabbed your LinkedIn. It's it's one reference to something or on your website and they, you know, looks like you do work in this area, etcetera, etcetera. But as as you identified and as we're identifying just as as consumers of that, it's not very high quality AI.
Michelle Gouveia:Right? And so I can't imagine that even if a company is is getting to that 1 to 2,000,000 AR very quickly that they become very sticky from a performance standpoint. And so, what what are the steps you think that a company that's providing some type of AI solution really needs to go through in terms of validation testing, like to really become credible, both from with existing customers and then being able to tell the right story when they're in that room pitching their solution against the broad swath of other AI solutions that that are just coming up seemingly every day.
Christopher Wells:I'm gonna take notes on this answer, by the way.
Sunil Rao:Look, and I gave the use case that I think is most prevalent because once again, it comes back to the situation of risk analysis and like, how quickly can I get something stood up and am I willing to take a risk here? So I think with this email and, you know, this this outreach kind of use case, everyone's trying to generate leads. Everyone's trying to generate pipe. Outbound, right now, the baseline, you know, a lot of CROs will say, like, the worst BDRs are not doing the best job. So this maybe even just lifts that a couple of basis points.
Sunil Rao:So it's definitely an improvement, but it's not a good improvement because now there's a deluge of of generated emails coming to people's inboxes and to people's LinkedIn emails. So I think from that perspective, it's definitely not the best use case from a durable business standpoint, and I think that's where you're coming from, Michelle. It's like, how do you how do you create something that's meaningful and generates value over time as opposed to, like, a spike of, like, people experimenting with it, and then it dissipates? In our experience, and this is the reason why we're not building, in that space specifically, We've we've specifically looked at this sales engineering role and and thought about the deeper workflow that we can address for that individual in that company in that role. And and for us, the value comes from how much can you take off of that person's plate consistently with quality.
Sunil Rao:Right? And I think I think if we as we double click and triple click into that role, like, one of the things we've come across early days is, hey, like, we can take a fairly complex RFP and we can process it and we can come back with answers. We can only answer about 50% of it. Right? So then the question is, what does the person do in that situation?
Sunil Rao:Well, at Indico, maybe it's they hit up Chris, and they're like, Chris, for these five questions, I need your help. Right? So we thought about how does that fit into the workflow of the individual? Is it a company that uses Slack? Do they use Teams?
Sunil Rao:That's probably how they usually reach out to the other person. Can Trivble just do that? So we we try to graft onto the process of answering this as a team. Tribble joins the team, and Tribble ends up becoming a team member that goes out and fetches the answers from different people. So we think that allows it to become a team member in a continued way, which is a little bit stickier than, like, this one shot experiment that you go out and try and, you know, you can turn off.
Sunil Rao:So the question I always ask myself is like, hey. If someone turned off the switch for Tribble, like, what's the immediate pain you would feel?
Michelle Gouveia:Mhmm. Yep.
Christopher Wells:2 thoughts springing out of that. 1, it gives me great comfort that I'm no longer in ignoring an email written by a human. Makes me feel better. And then 2, I the the word copilot is, like, way overused today, but it sounds like you're taking advantage of one of the most exciting aspects of these models, which is if you give them tools, and in this case, the human beings are kind of the tool. Right?
Christopher Wells:If you give them access to the tools they need, they can their their capabilities expand exponentially. And and I think that approach is is part of the recipe here.
Sunil Rao:Yeah. It it absolutely is, Chris. And and, yes, copilots like, there's there's no shortage to the number of copilots available today. I think for us and and just when we talk about agents, and I know that's another word that will be hyped and, you know, everyone will claim to have agents and all that good stuff. But for us, it really is, hey.
Sunil Rao:The the thing that we are building, does it have agency to select from a set of tools that it has access to and dynamically use them in order to complete the task at hand? And for us, it's when we talk about agents, it's very much the agent that will facilitate this business process. So we look at it as an employee that may not be as capable as other employees but can get this one thing done consistently well, given the tools that it has access to. And when we when we when we constrain the domain of the problem to that, it becomes both tractable and also delivers consistently with quality what is expected of it, which which is very much the way we're thinking about shaping
Christopher Wells:this. It's also nice because, I mean, you have that Slack or Teams conversation. Right? So you have an audit trail. Where did the answers come from?
Christopher Wells:Why did this go into this RFP? Right? I I think that's super powerful. And especially, you know, with Michelle here talking about investing in tech for regulated industries, like, that's core. That has to be there for adoption to take place, I think.
Sunil Rao:Yeah. And the one if you go to our website, it says kiss your apps goodbye. And the reason for that is we have SaaS fatigue. Right? We've got way too many apps to log into.
Sunil Rao:And when you hire a new person on your team, what are you doing? You're telling them, hey, here's a Google Drive folder. There's like 500 bajillion documents you have to read over the course of the next 30, 60, 90 days. Oh, and by the way, here are the licenses that you need to the 17 systems that we use, and you can swivel chair from system to system and you can figure out how to use them after you've ingested all these docs. I was like, that looks like a really good pattern for some sort of RAG based workflow with an agent that can go to those systems and do stuff.
Sunil Rao:It's like, okay. Let's do that.
Christopher Wells:Yeah. I gotta ask Michelle, like, as you think about underwriting. Right? Like, this sounds like a slam dunk.
Michelle Gouveia:Yeah. I I literally as he was giving that answer, I was like, okay, so all the business lines that have very straightforward standardized like question sets or data that comes in, you know, how do you start just having one of those agents be going to triage, raise the main points that's coming out of that that main application or submission and then push that through. Or even the ones that are non standard just going through and raising like this is missing, this is here, like there's nothing to extract from this and then pushing it forward. I I my mind also went somewhere else as you were talking about. We had on a on a previous episode someone talking about agents and this idea of building, or having multiple agents available to connect workflows that were previously, somewhat disparate because they were part of coming workflows that were part of, like, silent systems or things like that.
Michelle Gouveia:And so as you're talking about building an agent, this very specific, capability that you know, whether Triple's doing or just generally thinking about an agent that's very good at this. How do you think about connecting or expanding that that agent to to be able to do like 2 steps, is pretty standard workflow. Like, what are what are the challenges there, like, of, you know, human in the loop and keeping someone apprised of that? Or what are the rules and things that may have to change as part of a workflow to do that?
Sunil Rao:Yeah. It's, it's also the namesake of the company. I think we've got many tribbles that can you know, there there are multiple copies of of each other and they do the same thing. Or do we have different Tribbles themselves that all do and specialize different in different things? And I think we've we've wrestled with this as well.
Sunil Rao:And I think right now with the focus of where we're at, it seems to us that the right thing to do it seems like the right thing to do is is to give the Tribble skills and the skills can vary over time. So, right now, it just happens to really be good at retrieving information from specific classes of documents and specific systems. It it is really good at producing specific shapes of output, when it comes to documents and proposals. But in the future, it might also be able to cobble together some videos and create some demos that an SE would do. But that becomes a skill you add to Tribble, the entity, which then has multiple copies of itself.
Sunil Rao:So I think there's discussion of, like, do you need to partition parts of the process into different work flows and then have agents specialize in each thing, or do you need to just have an agent? I look at it as a very capable employee. Like, you can choose to have 2 people that are specialized in 2 different domains and have them as 2 people serving those functions, or you can have one person and a copy of both person, but, like, you know, give them different access to different tools and have them do different things. So I think it really comes down to how how you're thinking about expanding the core product and whether or not there are any inherent benefits from having, like, some sort of domain segmentation across the agents. I I'm not convinced either way, actually.
Sunil Rao:I feel like it's an area of research where, I haven't seen anything definitive. I'm I'm curious to get Chris' thoughts on this one.
Christopher Wells:No. I I, I think, I have so many thoughts about this. One, I think the possibilities are endless if you think about the tools you could arm a triple with or an agent with. As you were talking about this, I was thinking, why isn't there one that just watches Slack conversations and looks for common questions answered and, like, auto builds your FAQ for the organization. Right?
Sunil Rao:Funny. You should bring that up. That is actually one of the things we ingest. So we spray it
Michelle Gouveia:on. Nurses.
Christopher Wells:No. Not at all. We're just great minds and all that. That's it. But, you know, that that almost functions as like a meta agent.
Christopher Wells:Right? And so what does it look like? I talked about, you know, endless specialization is possible. But then what what happens if you start breeding the Tribbles? Right?
Christopher Wells:And you you combine the capabilities. Right? Can you really build a generalist from these things? Getting some Tamagotchi memories from decades ago. Yeah.
Christopher Wells:I was just thinking of the quote from the great doctor Leonard McCoy. As far as I can tell, they're born pregnant. But but, like, what what what actually happens? Right? Like, can can you go from specialist to generalist?
Christopher Wells:The same way that happens in an organization. Right? Like, someone who's been underwriting one line of business for years and then switches lines. Eventually, they just kinda know everything about underwriting your organization. I think we have a few years before I can really answer that question definitively, Sunny.
Christopher Wells:Hey.
Sunil Rao:It is Go ahead, Michelle.
Michelle Gouveia:No. I, if you have something on on that point, Chris, please, Robert. I was gonna go back to to, something else that you you said previously.
Sunil Rao:Yeah. I was just gonna unpack that a little bit more because I think the general purpose nature of the the agent's like, ultimately, you're conversing with this thing that's oh, I said this yesterday in this meeting as well. It's like everyone is chasing the omniscient thing. Right?
Michelle Gouveia:Yeah.
Sunil Rao:Basically, it's like plug into a bunch of things, it learns, and it's able to do whatever task you throw at it. And those tasks become ever more complex with the more capable foundational models that are gonna come out in the future. So let's assume that that's like at some point in the future, you know, delta t that's going to happen. So it's like, okay, what's tractable now and how do you think about use cases today? Well, we still have this world with humans and UXs, and we like to use these tools in different systems.
Sunil Rao:So it's like, how can we build a scaffold or a UX to interface with the agent that is very grafted to the workflow? But it's like conversational plus plus it's not just pure play conversation because most people suck at asking questions or knowing what to ask. Right? Which then is the problem because you come like it's like when we go into a restaurant, you see a menu with 2,000 items. It's like, ah, I don't know
Christopher Wells:what I want. Right?
Sunil Rao:And then you just don't do anything. So I feel that that's kind of what happens when you just take this super capable thing and dump it into a conversational interface unless you give some guide rails on like how do you interact with it, what kind of value can you get. So the opportunity for us has been, hey, can you build the scaffold around this RFP use case to guide them through what their business process looks like? So that's kind of 1. So when we talk about skills, like, the skill comes with this kind of grafting.
Sunil Rao:So you probably don't wanna shove every possible UX and make this an omniscient thing. Right? So like, there's probably this intermediary phase where you have like these scaffolds around the capability that get these business processes is done. But then the question over time is like, do you even need that? Right?
Sunil Rao:And this is where I think it gets a little bit tricky.
Michelle Gouveia:I'll go back to something else you you were talking about, you're talking I I my mind went to, Chris, to broker use cases also or for agents and something like this where so when you're thinking about from a carrier perspective, right, they're getting all of these, submissions and things like that. But then there's the responses that come from the carriers back to the brokers and they're having to create, put these packages together of, you know, here's all of the policy information and offering from each of the carriers that we've gone out to here, you know, customer might now pick like the being able to automate that. And I don't I don't know what what's happening in that space today. This is a call up to any broker that may be listening like to come join and have a conversation with us. But I think that I've spent a lot of time thinking about AI and use cases from the carrier perspective.
Michelle Gouveia:But for sure, there's there's a large number of them that would be helpful and add efficiencies and automations in in the broker world as well. I don't know, Chris, if you have any thoughts on on that also, but that's where my mind went when suddenly you were talking about, just pulling in all these like various sets of information and packaging it up.
Christopher Wells:No. I I'm right there with you. I think the paradigm that Sunny is describing, which is treat it like it's a coworker and give it access to the things it needs and clear instructions for how to get back the right thing. I think that's a great paradigm and abstracts across a lot of different industries and roles.
Sunil Rao:Yeah. And and I'll jump in, Michelle, just because brokers did come up in the conversations with a couple of insurance companies. It's just that You and I did not rehearse this. Yeah. But but but it is it is it is so relevant.
Sunil Rao:Right? Because it's like essentially these representatives for your company. Right? And you want them to you want them to be able to take this vast amount of information and represent it consistently and accurately while they're serving the channel. And if there are any back and forths with the end customer, then then you wanna equip them with everything possible to do their job well.
Sunil Rao:Right? And I think I think this is where, you know, know, I was actually talking to to a leader at one one of these big software companies just a couple hours ago, actually, and we were talking about the consistency of responses across their global team varies greatly as a function of available capacity. So it's like, hey. Like, if people are busy, they're gonna very quickly go through this stuff and fill it out. Quality takes a hit.
Sunil Rao:But if you're optimizing for just getting everything done, your consistency and quality varies. Right? And it's a function of what capacity is available on the team. So it's like even normalizing that curve has inherent benefit because your win rates or your potential you know, your end to end experience with a customer goes goes up, and that has a positive impact in the business. So I think the broker model is very much an area ripe from at least from what I'm hearing from customers, ripe for disruption given the document heavy process that's in place.
Christopher Wells:I think we're gonna see it on the consumer side as well. Like, you wanna buy insurance. Right? The progressive has their tool, whatever it's called, that finds the right insurance for you. Like, why do I have to fill in anything in a website?
Christopher Wells:Why don't I just have an agent that goes out and shops for me? Like, same with travel. Like, here's where I wanna go, rough dates, like, you know, meet the criteria, find this stuff for me. Like, I I think that's coming. Eventually, it's gonna be a world where it's just our agents talking to your agents and, you know, we're all on an island somewhere maybe, hopefully.
Christopher Wells:I
Michelle Gouveia:I know. Does a good job booking your trip.
Christopher Wells:That's right. Yeah.
Sunil Rao:And by the way, I'm not sure, like, I know the GPTs are out there and what OpenAI released, like, lead assistance like that. It can already do a set of things and actually, when you and you guys know this. Like, when you configure the assistance, you can provide APIs and you can ask it to access specific systems. What we've noticed in our testing is and this is like the best foundational models. I think where things tend to fall up fall apart is the complexity of the task, especially in a B2B setting.
Sunil Rao:You can have any APIs that are exposed as tools to the model. The model's ability to select the right tool, like the combinatorial explosion of adding 15 systems and the complexity of the task and the depth of complexity of the task within that system all play a role in its ability to act. So it's almost like there's like this you know, I think back to this curve, I think it was, I don't remember who it was, but it was like this curve of like intelligence. Right? And it's like compute versus intelligence.
Sunil Rao:And it maps like, different organisms in terms of, like, what their compute capacity is. Like, I think where the models are right now is still, you know, there there's this is what the whole concept of AGI. They're not at the point yet where you can throw 15 systems at them and it'll figure it out every single time. Right? But not too far off, I think, is a time where the foundational model, you can now throw more and more tools at it and just have it figure it out.
Christopher Wells:Absolutely. I wanna circle back. You were talking about some of the obstacles, security, obviously, throughput, things like that. One of the, in my experience, largest obstacles to adoption of new technology is just the people who have to use that new technology. So, like, as you're out there deploying your troubles, are you seeing, like, that organizational immune response to new technologies, or is it feeling like it's seamless for them and and they're adopting it well?
Sunil Rao:You know, when when it was in earlier days for the company when we were exploring kind of like, hey, how do we how do this is like magical. Like, you can put this in the hands of people and they can get so much value out of it. But if you don't precisely tell them how to use it it doesn't get adopted. Right? And I think that's been that's been like the first learning, which is conversational interfaces are great.
Sunil Rao:It's really hard to build consistent habit or workflow around it in a business setting unless there's a repeated thing you're going to it for. So that's why this focus on on the RFP use case made a lot of sense for us because it's an established business process. There are people in the company that are responsible for doing it, and we can kind of look at the existing workflow and interject into that. And that becomes the place where you reinforce the way to retrieve information. So that's that's been a good way to, like, attach or graft yourself to an existing business process.
Michelle Gouveia:Question for you. I'll reverse a little bit of what you you just said. Right? So it's not it sounds like that insurer that that you were working with had a very had what I'll call at least a a, fairly good understanding of what's available in the in the Gen AI landscape and had identified already a business case that they wanted to pursue testing that technology and those capabilities with. In your experience in talking to either insurance carriers or larger enterprises, how often are you finding that to be the room you're walking into versus a room where it's like we've got to do something with Gen AI.
Michelle Gouveia:What can we do? And just kind of leaving it like open book because I I I feel like it's probably split where there's what I think we're talking about this earlier where there's there's something like we we we don't want to be behind. We have to find something to do gen AI. What are all the big things we can do? They don't come in with a plan or something to really test versus others that have maybe maybe an innovation group or someone that is very focused on, really identifying like viable AI solutions to bring in that will be longer term, part of a longer term strategy.
Sunil Rao:I think it's a fair point, Michelle. Like we're seeing both. And, you know, I think back to my my Salesforce days where, the company evolved fairly fast and the platform, core Salesforce platform was built in such a way that you could build any use case out on it. So a lot of the times, you'd have these executives who are like, hey. I made this big investment.
Sunil Rao:Like, what other things can I build on this thing? Like, let's mediate. Let's come up with all the different ways in which we could use this wonderful thing. I I kind of feel a little bit of that now when when we go into conversations where it's like, hey. Like, help us identify the highest leverage or the highest value use cases.
Sunil Rao:And a few conversations do end up like that. It's very exploratory, almost consultative, if you will, about what makes sense. I I think for us, it's been really important to, like, identify where there is acute pain in an existing process and walk in with a point of view on that as opposed to making it a very broad use case exploration kind of discussion. And those tend to be the better, you know, better discussions to have when you're when you're early stage company and, you know, adding customers. Right?
Sunil Rao:I think to your question about just in terms of what we're seeing in that vertical, I think, you know, there are people that don't know what an LLM is. Right? Like, which is, like, one end of the spectrum and then the other end of the spectrum, it's like, hey. We have a mandate to figure something out in the next 3 months for these domain areas. And and and there's, like, a full spectrum there of of, like, education.
Sunil Rao:So I think it really matters at the company and, you know, I once again, there's a bias working a lot with tech companies for us. We tend to walk into meetings, like, assuming a level of knowledge in this space. But there are people out there that just are not ingrained in this day to day like us. Right? So I think you have to appreciate the fact that there's some education needed in some conversations.
Michelle Gouveia:Do you find in your experience and I know it's it's early days and every company does things differently, but do you find that the the stages from that initial meeting to getting something approved to go through and then launching whatever is a pilot or full engagement, that those steps are impacted by by which room you're walking in? Too many is it harder to get things approved if they're very specific and what they want and they kind of have buy in? Or is it they're so open to it that like they'll still push it through? Like like what's the dynamic there? Like who's the decision maker?
Michelle Gouveia:Do they are they already identified for you based on the room that you're in or are they, like, kind of, staying their wheels a bit as you go through?
Sunil Rao:I think with the focus on the use case, the buyer is pretty clear. And this is why, you know, for for me and what my guidance to the team is always we don't talk about AI. It's a business process that we're helping you make better, and we think that there's at least a 10 x improvement in this specific process. It just happens to be using AI as a technology underneath the hood and all the great stuff that's coming out. And then, of course, there are some questions around, like, we've never seen anything move this fast.
Sunil Rao:So are you as a company thinking about that? Because we don't want to place our bet with you and then, you know, 6 months later, I have to change to something else because the stuff is moving at a blistering pace. The answer to that is, like, yes. We're thinking about that. It keeps us up every night.
Sunil Rao:And in fact, we build the company in a way where you can swap out the engine as the more capable things come out. We don't think you can survive if you don't build a company in that way. Right? And then and then that typically alleviates that conversation. But I think the the way to do it, Michelle, is most people are looking either for, you know, if you're a hammer looking for a nail, that's kind of like the, hey, AI, let's go find a solution or space to find a solution, which is probably not a good discussion to be in as an early stage company.
Sunil Rao:But if you're feeling acute pain somewhere and we're coming in and we're talking about pain medicine for that pain, AI can be the way to solve it or not. And I think this is just the the better way to look at it always. At least that's what, you know, is working for us. Are I'm curious. Are you seeing IT trying to insert itself
Christopher Wells:in these conversations saying things like, oh, well, we'll just build it. You don't need to buy that from those guys.
Sunil Rao:I literally had this conversation 2 days ago. Oh, we're investigating if we can just build this on chat gpt. And, you know, it's it's like once again, it's like deja vu. It's like, oh, wow. I remember this in like, you know, 10 years ago, AWS has this beautiful portfolio of services that you can cobble together and build software software on.
Sunil Rao:You got an IT department that, you know, used to manage servers. They've got more time now because everything's on cloud so they can build this stuff. Right? It's like, great. I had and once again, I'll go back to my previous employer, Salesforce.
Sunil Rao:It's like, oh, we can just buy the platform license and rebuild the the core sales cloud ourselves. It's like you could do that, but then, you know, you're not going to have the team behind managing it as a product and and and you're now splitting your company's focus. So it's like this age old build versus buy, but I think it's more happening more acutely now because it's a much higher percentage of where you how far you can get out of the box with some of these. Right? It's pretty convincing.
Sunil Rao:The demos are really good. And and I think that the tech allows you to create these POCs that are much more complete than they would have been 5 years ago had you tried to cobble something yourself, which I think is actually really dangerous. Right? Because it's like the illusion of it working end to end, but then what happens is like, yeah, it's like when you when you have a consulting project and a POC build, it's not a product. It's it's something that was built for that engagement.
Sunil Rao:Right? And then it's like who's maintaining it? Who's keeping it up to date? How do you manage it? The whole product life cycle is gone.
Michelle Gouveia:Well, I think the oh, go ahead, Chris.
Christopher Wells:No. I was just gonna say it. It takes me back to the days of, like, doing proofs of concept as a data scientist where it's like, look at this awesome Jupyter notebook I created and it's like, yeah. What the hell am I looking at? There it made a chart.
Christopher Wells:I don't get it. Right? But now your middleware developer, your back end developer, they can put in a little bit of code and get off the JavaScript so it has a front end and obviously much more convincing. So I'm right there with you. Go ahead, Michelle.
Michelle Gouveia:Oh, I I was gonna say, I like the way you talked about that. Like, what really goes into the build versus buy, like, like, the the longer term impacts of that. And the one thing I'd add to that too is, you know, once once you do split that focus, right? And then let let's say you do successfully internally build what you are looking to buy, You're then stuck, I imagine, supporting that single thing that you built. Any innovation that's happening on that that platform or that technology, that company that was selling it to you, that they're they're moving forward.
Michelle Gouveia:They're adding enhancements. They're expanding capabilities. Your built product isn't doing that because it only knows what's happening within the four walls of your company. To then expand capabilities there, that's a whole different initiative, most likely in a whole different project than just having something naturally grow, and just have some resources there to support it. So I, my completely unbiased opinion is obviously on the buy versus build side.
Michelle Gouveia:But I think especially with with new technology, the hype hype cycle is happening and it's hot and fast and people are just like, well, maybe the differentiator is that we do it. I think people really just look part of that and they need to understand what they want it for and how long that that builds.
Sunil Rao:I I think the irrelevant. Michelle, I think it's exactly right. Like, the builders is by the allure of wanting to use this technology to build. It just it looks so easy to do. So people are infatuated with the idea of very quickly standing things up.
Sunil Rao:I mean, look, even even for us, like, we're a fairly small company. We've been around for a year. We're growing really fast, but it's not lost on me that if we've got we've got a whole bunch of RFPs and infosecs coming in. Well, we, of course, use Trivble to answer those things because that's the core product. If we see a feature is lacking, own if I need something for social media posting, I'm not going to put my engineering capacity to build that out using AI.
Sunil Rao:It doesn't make sense. It's away from the core product. So it's like buy a product for that unless it's part of the court. And I think sometimes you lose focus of that when you have highly capable folks that are in these different functions that can also experiment with technology. And the question I think most people have to ask themselves is like, is this what my customers are asking for from me?
Sunil Rao:Or am I building this as a pet project that I then have to dedicate engineering capacity and cost to? What's the strategic benefit of that?
Christopher Wells:Yeah. Why why do you think there's an arbitrage there? Right? Like, unless unless you have some inside information, there's some secret sauce in it's naive at best, and I think dangerous at worst.
Sunil Rao:Yeah. Focus. Right? That's that's the discussion. Yep.
Christopher Wells:You're you're talking about selling and talking to tech companies a lot. I can see, you know, the stuff that you're you're building out, that you're selling being complementary to a lot of other solutions. So how do you with all of this Gen AI going on and companies kind of looking the same, how is the collaborative versus competitive landscape looking among Gen AI startups these days?
Sunil Rao:I think there's going to be a mass rationalization of copilots. I feel like it's like every if if there are copilots for every single thing, we've created this ridiculously massive fragmentation of of functionality. Right. So we're now moving away from like having SaaS apps to the copilots for everything. And it's like
Christopher Wells:Kiss your copilots goof up.
Sunil Rao:Exactly. So, I mean, I I don't know how this plays out long term. I feel I feel as though, you know, when you look at the big software companies, they're all always there. They've always been about bringing everything to our ecosystem. We offer you a functionality or capability for everything you need.
Sunil Rao:But if you talk to most CIOs, like, the landscape is there there are many systems in place. Right? So I think aggregators of functionality across systems who make it easier for you to operate across existing processes and systems, there's benefit there. And there's probably some some platform leverage for some of the big guys to say, hey, we can we can do all these flows together and that'll make buying easier. I do from a competitive standpoint, you know, I think if you're in the space of we are a sales copilot, it is a massive ocean and differentiation is near impossible.
Sunil Rao:And for us, like, if you're very much in the space of RFP response software, you know, and and you're using this approach to solve for that, it narrows down the field. And and we just last week, you know, on on g two, we're now rated the number 3 software from an RFP response stand standpoint based on user reviews and customer reviews and feedback. Right? And that's that's clearly been a function of the the step change improvement that we can offer our customers given the tech. So narrowing the focus and reducing the playing field, if you will, I think makes it a little bit more tractable and less competitive.
Sunil Rao:But there is just a lot of companies out there that are trying to do stuff in this space. You sort
Christopher Wells:of interest yeah. Go ahead.
Michelle Gouveia:No. I have a question. So if you got a comment, you go first.
Christopher Wells:No. No. No. I was gonna I was gonna ask a question. You do it.
Michelle Gouveia:In that lens, if I'm if I'm the insurance carrier of a large enterprise, what are the right questions that I should be asking these startups in this landscape to really get to identifying the right capabilities and the right solution that for for what I need, that's who I need to partner with.
Sunil Rao:It's a great question. One of the early lessons that we learned, like within the 2nd month of operation because we wanted to sell to enterprise from day 1, SOT 2 compliance, just immediate certification and building on infrastructure that scales. There are so many open source libraries that you can pick up and hack together a solution in a weekend that's very convincing from a demo standpoint. And that's scary because you can dress that up fairly quickly from a UX standpoint, show it to people, and then get them excited. And then the question becomes like, is this built on scalable, secure infrastructure?
Sunil Rao:I think the kinds of questions I would ask as a buyer is like, what are you built on? How are you handling my data? What can it scale to? What's your maximum throughput? How many users can you handle concurrently?
Sunil Rao:How's your infrastructure built to scale? All questions that we've come across and had to come up with answers for over the course of the last year. Actually, fun fact, we determined our our theoretical maximum throughput thanks to, something that happened. I won't I won't name the specific provider, but basically, it exposed, like, the total maximum throughput of all the model deployments we have in Azure of the tokens per second we can get to a theoretical max. It was great because we can essentially handle, like I think something in the order of like 25 GPT-four 32 ks calls per second consistently for hours.
Sunil Rao:Which is
Michelle Gouveia:I don't know what you just said, but it sounded a million.
Sunil Rao:Yeah. I'm just looking at Chris's face. So it's like in order to do that, that that was a lot of infrastructure work from the engineering team over the course of the last 6, 7 months. Now, nowhere do we we never hit that kind of load at that frequency, but it's good to know because you know that you've built it in a scalable way, right? And I think, Michelle, to your question for enterprise, what's important?
Sunil Rao:Well, it's consistency of service, it's security, scalability. And I think the experience of the team that comes to the table and knowing to work with enterprise is crucial as well because the way you work with an enterprise and work with those teams fundamentally different from PLG sales where you sell it and kind of move on. You just have time. Right? And that's been that's been a differentiator for us as well.
Christopher Wells:We can you kinda touched on this earlier with the sort of the lower grade BDR maybe coming up a little bit. I I'm not convinced of that. I think the the the lower quality might just get to doing bad work faster with some tools. Not yours. Tribble won't allow that, but some of these tools out there
Sunil Rao:Amplifying the the not so good.
Christopher Wells:Exactly. Yeah. I think the distribution in a lot of cases becomes very bimodal. And I think we're seeing that already and, like, people are forgetting best practices with software because it's so easy to get software out. Right?
Christopher Wells:Yeah. They're forgetting best practices with model rigor because, well, it just gave me an answer and it was so quick and easy. Like, why should I test it? Why should I have a suite of tests? Well, maybe because the model checkpoint is gonna change in 3 months.
Christopher Wells:Right? Like, it's it's madness. So all of that leading up to this, my favorite question, you already touched a little bit on it with, the great copilot reckoning that you've predicted. We're gonna time stamp that one. Okay.
Christopher Wells:Looking into the crystal ball, and you tell me what the time horizon is, what do you think in terms of the impact of generative AI in the enterprise? What do you think the next big wave is? And on what timeline?
Sunil Rao:I say this to my team all the time. I think the speed with which like, I even just use as a proxy, like this 1,000,000 token context window development that's popped up. Right? I think Chris, like, we talked about a year ago when it was like 32 k context. Holy crap.
Sunil Rao:That's insane. So imagine all the things you can do. Right? Well, now guess what? You can take an entire 200 page government RFP, throw it into either cloud 3 or into Gemini 1.5 and extract from it every entity that makes sense to understand and and get a very good understanding of that doc.
Sunil Rao:Right? So that's a significant step change than what was available a year ago. Right? And I feel as though in 2 years time, 3 years time, it becomes very hard to predict how fast things are moving. If the trend holds that, you know, if you just increase compute the model scale, which seems to be true, right?
Sunil Rao:And that's what everyone's doing. And that's why NVIDIA is worth over $2,000,000,000,000 Then the speed with which things change is also going to stay constant, at least, you know, whatever you want to call this rate of change. For us, you know, I think I think when you think about the next wave, it comes back to the conversation about the ability to reason, the ability to use more tools. I think it's agents. I think I think it's ever more complex workflows and tasks, and we'll likely hit an inflection point of the percentage of work being done by an individual being done completely autonomously and the need for the same distribution of individuals in those roles.
Sunil Rao:And I think what's not happened yet is, once again, it comes back to, like, this the the great copilot reckoning will come when copilots are no longer needed because agents will be able to do a large percentage of these work. They don't need to be attached to a human. They can be independent of the human. And now your ratio between human and agent changes, and that distribution changes over time. So I think we're gonna eat into the opex for the workforce over time, and that to me is the next point in time.
Sunil Rao:And I don't know if that happens in 3 years, 5 years, but best guess, I can't see past 3. That's that's the way I look at it.
Christopher Wells:I'm with you.
Michelle Gouveia:That was great. I think that sounds like a great place to end since none of us know what's happening after 3 years from now. So we're
Sunil Rao:in a good time. We can help you with your RFP problems over here in trouble. So yeah.
Michelle Gouveia:There you have it. Thank you everyone for listening to the latest episode of Instructure Unlocked. We've had Sunil Rao, the founder and CEO of Tribble join us today. Thanks so much
Christopher Wells:for for chatting
Michelle Gouveia:with us.
Sunil Rao:Good night.
Christopher Wells:Thank you for joining us for this episode of Unstructured Unlocked. You can find all of our episodes wherever you listen to podcasts today. Be sure to write a review if you like what you hear.