Arvind Jain, Glean CEO: Revolutionizing the Future of AI-Powered Work

Thank you for having me.

Yeah, well, good question. I mean, there are so many lessons, you know, that I've been fortunate to actually learn from, you know, like really good people in the industry, especially like my time at Google where I spent over a decade, you know, starting with my early parts of my career to growing as a leader there. There are so many things that I learned at Google which I've actually then implemented in practice, you know, at Rubrik and now at Glean. Some of them, like so one thing that I learned was, and these are obvious by the way, there are no surprises, but you know, so first like, you know, it's about like, you know, the team. Like, you know, your company is actually nothing more than its people. And like making sure that, you know, you are finding, you know, really talented people who are very aligned, you know, with the mission, with your company mission. You know, that's the team that is going to build the company together with you. And so we focus a lot on that, like, you know, our team building, finding folks who are passionate about the problems just like, you know, how I am, we are. That's a big part of it. Second thing that I learned at, again, at Google was, you know, technology and innovation, like they have to be at the forefront of your company. And like as long as you make sure that you innovate really, really well, and then there's a supporting cast of folks who are actually helping your engineers build great products, you can do a lot, like, you know, with building a good product. A good product sells in some ways itself. Of course, you need a great sales team and marketing team, but that's like, you know, the basic of what you need. And like often times, you know, entrepreneurs underestimate the difficulty of building a good product. And so that's one of the things that's been, you know, core part of all our companies. At Rubrik, at Glean, we were always first focused on building a great product before anything else.

Well, I mean, I think first, you know, I don't actually really claim that, you know, we have succeeded in a big way or more than others. Like, I think we're working hard trying to actually build good products and good companies. And I think like for me, like again, these learnings, you know, that we were just talking about, like, you know, focusing on like building the great team. So that's one of the things that I do, you know, is really like to spend, like I'm the number one recruiter, you know, of talent at our company. Like, you know, love to be in that recruiting job. I love to spend a very, very large fraction of my time actually doing that. And I think that is what has helped us, you know, build these really incredible founding teams, you know, that then go and actually build the company for us. So I think just having, you know, that mindset of like, you know, the company is basically, you know, the people and not much more beyond that, I think, you know, that's sort of helpful to us. You know, that's one thing that's helped us, you know, build great teams. And like, you know, once you build a great initial team, you know, they are the ones who are able to attract, you know, the rest, you know, like the next set of people who are going to come in the company. And, you know, the cycle sort of perpetuates. So Glean today, for example, is known for having very, very strong talent, you know, great people. And it's because of like, you know, the early team, you know, that actually had those credentials and had accomplished so much in their own careers that people wanted to follow them.

Yeah, so the problem that Glean solves is helping employees quickly find information that they need to do their work. So when we started out, like, you know, that was our mission. Like we wanted to be the Google for people in their work lives. And our product was not focused on helping, you know, one user population. Like, you know, we felt that, you know, this problem of finding information, finding documents, you know, conversations and getting answers, you know, to questions that you have, this is a very fundamental problem that every knowledge worker has. So in a company, whether you are an engineer or a customer support person or, you know, in sales or in marketing or HR or IT or finance, like it doesn't matter. Like, you know, all of us are constantly working with data and knowledge and information. All of us are interacting with our colleagues, you know, and asking them questions all the time. And that's how work happens. And so all of us need that product that we were trying to build. So we've been horizontal from day one. We want to build a product that actually is used by every single person inside the company. Glean typically gets deployed wall-to-wall. And, you know, you have to think of it as like, you know, it's your personal assistant. Like, you know, all of us have day-to-day work, you know, where we constantly need help. And think of Glean as an AI-powered companion that's sitting next to you and it helps you with whatever things you have. So for example, as an engineer, like I need to actually, when I'm learning about some new technology, I want to actually know who are the people who built it, like who are the ones who I can actually go and get a quick overview from, or maybe not the people, but like, you know, just have AI read all the documents and give me like a tutorial, give me like an onboarding sort of course, you know, on that technology. Like, you know, these are the kind of things, you know, I would like to use Glean for. Or if I'm trying to troubleshoot an issue, again, I want to sort of get some, you know, hints and guidance on like a way to start. And so Glean helps, you know, a lot of these engineering use cases. But then switch over to customer support. Customer support agents are actually constantly resolving, you know, cases that they have and answering questions for their customers. And again, like, you know, Glean will help them, you know, auto-answer those cases, you know, resolve those cases or answer questions that their customers have. As a sales person, like you want to actually write a great message to a prospect, and Glean will help you sort of utilize all of the company's context and information on that customer to help you, you know, create, you know, great prospecting emails, things like that. So you think about work, like, you know, AI is here, you know, to actually make an impact on all of us. And it can help us, you know, with our day-to-day work. And that's how you have to think about it. It's that personal companion that sits next to you. It doesn't matter what role you play in the company, we're going to be helpful to you.

Yeah, yeah. So first, just to make sure that, you know, our listeners understand what Glean does, we answer questions that your employees have. And we answer those questions using both the world's knowledge as well as all the knowledge that sits within your company, inside your company, your private company knowledge. So we're able to make use of both to actually help people with answers or with their tasks. Like, you know, we can actually do work for them. So actually just functions like ChatGPT or Google inside Enterprise. Now, what are the difficult parts though? Like if you think about AI technology, all the AI models, these great models like Gemini or GPT-4 or Claude, they've all been trained, you know, on the public world's knowledge. And then these models don't know anything about a specific company. So inside a company, like, you know, if you have a simple question like for example, how many days do we get off, you know, as part of our plan, or, you know, can I reimburse my cell phone bill, you know, these are questions that these models won't know the answer for because all of this information is private and inside your company. So what Glean does is it does a really good job, you know, bringing all of that internal company knowledge in one place. This knowledge actually sits in many, many different systems. So we have these deep integrations built into most of our enterprise systems, you know, systems like Workday or Salesforce or ServiceNow, you know, Google Drive, OneDrive, Office 365, you know, like that's where, you know, your company's knowledge and data lives. So we have hundreds of these really, really deep integrations into those systems to bring the knowledge, the data, the information from all of those systems in one place. And that's what you need because ultimately you have to connect your enterprise data with the power of these public, you know, language models to make magic happen inside your company. So we're really good at that. You know, that's one of the places where we distinguish ourselves from any other enterprise AI tool, including ChatGPT, you know, which again, like, you know, is really, when you think about ChatGPT Enterprise, you know, it actually still using the world's knowledge to answer questions for you. Second thing is if you think about enterprise knowledge, it's actually very different in nature than the knowledge on the internet. It's private, it's private in nature. Most information is actually privileged. Like, you know, for example, you may write a document and, you know, and you've shared it only with two people inside the company. So only the three of you then can actually use that document for anything. The rest of the company is not allowed to because that's how you structured this. So most company knowledge is like that. So understanding, you know, the governance, the permissions models, who can access what, these are very important considerations. Like you can't train a model on all of enterprise knowledge and then use that to deliver information to everybody inside the company because then you're leaking information. So like doing this in a safe and secure way, you know, properly understanding like the governance of your knowledge inside the company, that's another big focus. We spend a lot of time on that part. And then the last thing I would say is, you know, if you think about companies, you know, large enterprises, they've been there in business for like, you know, maybe 100 years, 200 years, a long, long time, right? And there's a lot of knowledge, a lot of information, you know, that's produced over many, many decades. And so first, there's so much of that, so you have to build scalable systems to process all that knowledge. But second, you know, a lot of that knowledge is obsolete today. Like, you know, it's actually no longer accurate. And so when you think about any enterprise, you know, there's tons of information but of varying quality. Some of it is fresh and up to date. Some of it is inaccurate and stale. Some of it is actually written by, you know, subject matter experts and therefore it's of high quality. Some of it is low quality. So you have to sort of really understand the knowledge deeply inside the company and always use the best knowledge and bring it to the AI model so that they can use the best knowledge to actually answer questions for people. And a lot of work that we do actually goes into that part of like, you know, really deeply understanding an enterprise and its people and how they interact with knowledge and use all of that information to figure out like, you know, the best pieces of knowledge for any given individual and any given question and then make AI work on it. So those are the three different vectors, you know, where Glean distinguishes itself from other players.

Yeah, that's actually the magical power of language models and AI technology of today is that AI is actually really, really good at reading long-form text documents, unstructured information and make sense of it. So our approach is that, you know, first we connect with all these enterprise systems, you know, we bring all this unstructured knowledge in one place. And now, you know, we are actually using this technology to build a RAG-style system. You know, we index all of this content, we embed, you know, we build custom models inside your enterprise which are trained on your own, you know, your company's corpus. And then use those models to embed that content, build the semantic search layer, you know, which sort of understands everything about your company and people using these custom-trained models. And so that's the first part of like, you know, the architecture of the Glean product is building semantic search over all of your enterprise knowledge. Now when you come and ask a question, like the way it works is that first we will take that question and we'll try to retrieve the right pieces of information. And this is all unstructured, but we're going to actually retrieve, you know, information that we think is relevant to that question that you're asking. And then after that, all you have to do is, you know, you ask an AI model to actually process all of that unstructured information that we're providing to it, which we think is relevant to the given question. And AI is going to do a great job at actually making sense of that unstructured information and actually producing an answer. Like whether you're trying to synthesize, summarize, analyze information, AI is really good at that. So just to be clear, like, you know, handling unstructured information effectively, like a lot of that magic actually comes from the AI models that we are able to actually use.

And that's, by the way, that's true for every company or the world in general. Like on any given topic, you know, you have lots of information, you know, contributed by a lot of different people. And they're all sort of different. Like, you know, we all think differently, we all have different sort of view of the world and information and knowledge, and we actually share it. So you always have the situation of in a company you have information and knowledge that's produced by humans. Some of it is, you know, inaccurate because, you know, it's basically now obsolete, you know, things have changed since the knowledge was created. Some of it is, you know, incorrect because, you know, it was incorrect from the beginning because whoever wrote it, you know, didn't know the right facts. And so you have to deal with this. This is natural. Like you will have different levels of quality of information. And so the way to solve for that is you have to sort of go deep inside the enterprise and build an understanding of the company and the people. And building understanding of people is actually very important here. So in general, like, you know, it's true that the newer a document is, the more likely it is going to be correct because, you know, it's reflecting, you know, the changing world and how things happen today. So basically just understanding that part from all the company knowledge, what content was produced recently, typically, you know, is of higher quality and we'll actually promote that in our retrieval. But then like, you know, as you said, like sometimes, you know, that's not necessarily true. You may have a person who's not a subject matter expert on a given topic and they write something new, but that's maybe not as good as somebody who's a real deep authority on that topic and they wrote something else like a week before. So you have to sort of take that also into account. Like somebody may be more knowledgeable on a topic and so therefore you have to give information that they write a bit more weight. And then finally, you know, there are a lot of signals in the enterprise. Like, you know, if you think about a document that actually a lot of people inside the company are consuming, or recently, then that actually gives you a good signal that that's probably important. Like, you know, there's a reason why there's so many people interacting with that content. It gives you a signal that both it is sort of maybe up to date and fresh because otherwise somebody would have complained that, you know, hey, this is not giving me the right information. And so these are the, like, so when you start to think about an enterprise, there are a lot of signals available if you actually focus on understanding the people and the knowledge, look at interactions, look at how people are creating knowledge, how people are using knowledge, then you'll be able to accumulate all these different signals which sort of tell you like for a given topic, for a given person, what's the most important information for them. And that's, of course, the core technology for Glean. You know, we need that to build a good search experience, we need that to build a good chat experience.

Yeah, yeah. So that's a great question. The org chart actually, and in fact I think the smaller companies probably are even worse because in a small company you may not even have an org chart or like, you know, people may have some title but like, you know, you do so many different things and like, you know, one title won't capture everything that you do. But see, the way Glean works is that we connect with your HRIS system. So we look at the organizational hierarchy, we look at who you are, you know, who are the members of your team, who do you work for, what is your official title. All of that is useful information, but it's not sufficient. More importantly, what we do is we're actually building a knowledge graph for your enterprise. So we're actually also looking at what every individual is actually doing, right? What content are they actually writing? Like are you authoring a lot of documents on a given topic? Are you actually answering questions? Like if you think about systems like Slack or email, you know, when somebody asks questions, are you the one who actually answers questions on a given topic a lot? Right? And so those are the things that you have available to you. The way the Glean model works is that, you know, our language models will first actually understand your enterprise. It'll look at all the corpus, all the documents, all the information, and it'll figure out like, you know, what are the key subject matters and what are the topics, what are the concepts that are relevant to this enterprise. Then we will look at like, okay, who are the ones who are actually contributing to those topics? Like for example, who's actually writing about language models or who's writing about AI or who's actually writing about our backend system inside the company. So you look at that contribution that people are making and that actually allows you to then go and build an understanding that, hey, this person, it seems like, you know, they are very active on talking about these four different topics. There are a lot of documents, they answered a lot of questions for other people, and therefore they're an expert on that topic. So that's how we build an understanding. It's not based on just reading your job title, it's actually based on seeing the work that you've done in the company.

Yeah, well, first, so like everything that we talked about, like, you know, about like when Glean gets deployed in an enterprise, it connects with your different systems, it understands, you know, the people, knowledge, subject matter experts, all of that stuff. Everything is actually automated largely in the sense that, you know, the software does all of this work. You know, our integrations are also very turnkey in nature. So when you deploy Glean in an enterprise, you actually can be running quickly from a technology perspective. Like most of our time, you know, like deployment, you know, why it takes long is because of like the actual processes that you have to go through, you know, which are not necessarily technical in nature. Like, you know, the surround security and business value and like a lot of other things that we have to work on. But the actual product is not, we're not building a different product for different companies. It's the same software that actually goes and it does work well, you know, for different personas within a company. Like, you know, the same product does a good job for engineers, for support people. But similarly, you know, the same product works well in a high-tech company or in a finserve company or a healthtech company or retail or manufacturing. Because, you know, we are actually solving a very basic, very fundamental problem, which is that, well, like, you know, if you have people who have questions, who work with knowledge and information, we're going to be helpful to them. So like, of course, like in all these companies, you know, you have lots of knowledge workers and so therefore the product generally works across different verticals. But then we do go and like there are some differences, you know, vertical by vertical. For example, you know, for financial companies, they have certain requirements, you know, when it comes to like how do they deploy AI. You know, they have restrictions around being compliant with laws and regulations. So there is a lot of work that actually then happens vertical by vertical where we have specific teams focused on those verticals and make sure that our solution, our product is actually working well within those companies. That we are actually working on making sure that we are bringing in all the compliance-related work that we need to, for example, to actually serve healthcare companies properly, like we need to have HIPAA compliance. So then we go and work on that. But overall, the idea is that, you know, it's a product that we ultimately make it work for different verticals, for different companies, largely in an automated manner without requiring long implementation cycles at our customers.

Yeah, yeah. I don't know if we went through like that real solid process of figuring out our ICP. So it's one of the things that in hindsight, like I can say that well, we should have done a better job. But overall, and that was one of the difficulties of our product is that our product is actually so basic, so fundamental in terms of what it does for people. Like, you know, it's like Google, it's like ChatGPT inside a company. So every single person, no matter what company you work at, you know...

Whether it's a large company, small company, what vertical it is, everybody has a need for this product. And so therefore, when every business in the world, large or small, tech or healthcare, whatever vertical you're in, if everybody's a customer, who do you go and focus on? So we initially, it was sort of easy. We knew that the largest enterprises have more requirements, and also they won't feel as comfortable working with an early-stage startup. So we said, let's pick a set of companies which can actually move fast, which are more amenable to experimenting with the product that a startup is just building. And so we picked somewhat smaller, but we didn't pick a company with less than 100 people because we felt that a less-than-100-people company, maybe they're not as motivated, maybe the pain of not being able to find things is not that high, because generally these companies will be younger in nature. They're only a few years old as a company, and people typically will know who to go and ask for help or where to go and look for things. So the need might not be as super strong. And so that's what made us choose midsize companies, which are 500 people to 2,000 people, as our sweet spot, where we knew the pain would be felt. It would be large enough. It was a pain that I felt like Rubrik sort of fit that same segment. We were 1,500 people and we saw there's a big pain there. That's how the company got started.

So we started with that as our ICP segment. And then the second thing that we did was, well, we were familiar with the fact that the pain existed within tech companies. So then we said, okay, let's go and solve the problem first for them. Keep it simple. Even though we're building a product in a very general way which will apply to all the different industries, start with what you're familiar with, who you're comfortable with, where you may also find it easy to actually identify the right people to go and do an outreach to and see if they're interested. So that's sort of what our thought process was when we started. So it was 500-to-2,000-people companies. It was actually mostly tech companies in the beginning that we exclusively focused on. So our first 20 or 30 customers probably were all tech companies.

And then in terms of how do you actually find those first 20 or 50 customers, so I was the salesperson, the solo salesperson for a while. Then I got one or two more people to help me, and we were all doing it as a part-time job, where we would do outreach and we used LinkedIn as our primary mechanism for connecting with people and sharing with them the problem that we're solving. And sure enough, you send messages to a lot of people and there are some who are going through that pain and they will be eager to learn more. And we're not selling. Initially when we did this outreach, we didn't tell anybody that, hey, would you be interested in buying a product? In fact, we were actually just looking for somebody to do a discussion with and make sure that we understood their requirement so that we can build a good product. So that's how we did the initial outreach. And there are a lot of people who are excited to talk with startup founders. They want to look at what innovation is happening and they have a lot of great experience that they can share with that particular startup and help them. So we found folks who were super helpful to us in our journey in the beginning. They gave us requirements, they told us how to actually go, what are the key pain points that they had, which allowed us to build the product.

Yeah, so our target persona was the CIO and also the CTO at these companies. Sometimes also the head of support. These were the three primary personas that we chose initially to go and talk to and learn from. And the reason why the CIO is the number one persona for Glean is because they have the broadest purview of bringing the right technology to the company across all the different functions within their organization. So we started with that.

I think if you are solving a problem that matters to people, then if you're doing an outreach in the right way, if you're actually making it clear to somebody that there is something for them to benefit from in that conversation, then you will get that interest. You will actually get some people to respond. I mean, everybody's busy. So many people have a lot of work that they have to do. I'm also very busy, for example, and I get a lot of outreach from other startup founders who are building great products and trying to actually sell something to us. And I take those calls, not all of them because I can't, but there are some which actually match a problem that I have today in my company and I want to learn more.

So I think it's not so much about, hey, you need to have a profile or a background or a status, because everybody has to start from zero. We also started from zero, right? We had no customers, no product. Just imagine, I'm actually trying to get somebody to give me time even though I don't even have a product to offer to them. But I'm promising a good technical conversation. I'm promising that I will share something with them that they might find useful. And of course, I'm actually hoping to learn a lot from them. So I think when you do an outreach the right way, you'll get the attention of the folks you want to get attention from.