Madrona Managing Director Soma dives into the world of intelligent applications and generative AI with Common Room Co-founder and CTO Viraj Mody. Madrona first invested in Common Room in 2020 — and we had the pleasure of having the founders join us on Founded & Funded the following year.
Common Room is an intelligent community growth platform that combines engagement data from platforms like LinkedIn, Slack, Twitter, Reddit, GitHub, and others with product usage and CRM data to surface insights from across an organization’s entire user community. Customers like Figma, OpenAI, and Grafana Labs use Common Room to better understand their users and quickly identify the people, problems, and conversations that should matter most to those organizations.
Soma and Viraj dive into the importance of deeply understanding the problem you’re trying to solve as a startup — and how that will feed into your product iterations — why organizations need a 360-degree profile of their user base, how Common Room has utilized foundation models to build intelligence — not just generative intelligence — into its platform — and so much more. So I’ll go ahead and hand it over to Soma to dive in.
This transcript was automatically generated and edited for clarity.
Soma: Hi, everyone. My name is Soma and I’m a managing director at Madrona Ventures. Today, I’m excited to have Common Room co-founder and CTO Viraj Mody, here with me. I’ve been fortunate enough to have been a part of the Common Room journey right from day one when we were co-leads in the series seed round the Common Room did a couple of years ago. So it’s been fantastic to see the company come from start to where they are today. Viraj, welcome to the show.
Viraj: Thank you for having me, Soma, and thanks for partnership from the early days.
Soma: Absolutely. So Viraj, why don’t we start with you giving us a quick overview of the genesis of the idea and where you guys decided this is the problem space you’re going to go after?
Viraj: So one of my co-founders and our CEO, Linda Lian, led product marketing at AWS and did a lot things by hand. So AWS has a phenomenal champion development program called AWS Heroes, and Linda was involved in that and that planted the seed for her in terms of the power of unlocking community and champions out there to help. And then independently myself, my previous experience was at Dropbox, which was pretty early in the product-led growth journey. And so we spent a bunch of time at Dropbox building internal tools that essentially helped unlock a lot of the insights like Common Room.
So Tom, one of our other co-founders, him and I worked at Dropbox, and then our fourth co-founder, Francis, was one of the early designers for Facebook groups, which was a very community-led, powered by the people-type surface within Facebook. So all of us had various perspectives on the same problem, and so it was a very natural fit when we all started chatting and exploring how to convert all of our various experiences into a product that we can help other companies leverage their community and customer base.
Soma: People say that hindsight is always 2020. Today, you’ve got a product out in the market. You’ve got a lot of great logos as your customers. S o you can say like, you know, “Hey, based on the traction, I can sort of project what the future could look like.” But when you started at Common Room, how did you decide at that time that this is a bet that was worth taking, and why you decided to spend the next chunk of your life working on this company, building this set of products, and going and doing something fantastic in the process.
Viraj: I think it comes from really understanding the problem space, and I think that’s why having co-founders with complementary skills is really important. Each of us brought a unique perspective but had a pretty unifying vision for where we want to see this product and the company go. We were pretty early in terms of seeing some of the motions that were being unlocked by community, both based on our experiences, but more importantly, talking to customers who are already doing this as part of their product journey. Some of our early customers, like Figma and Coda, were great partners in helping us think through how they would like to shape their business growth engine. And then that spurred a bunch of ideas for us.
One thing we did a lot early on that I would suggest everybody spend time if they’re in an early team, is talk to customers, not with the idea of helping them give you solutions, but really deeply understanding their problem. And then using your unique perspectives and experiences, plus your context of what’s going on in the ecosystem. We’re pretty well connected in terms of not just our networks but also a lot of peer companies. And so connecting the dots of the problems customers face, especially progressive customers who kind of see where the world is going, and then partnering with them in order to build that vision of the future. I’d say that’s been a key ingredient in the early days. And then, for me personally, I think I have a lot of experience and confidence in my ability to build this.
Back when Common Room started, there were a few fads going on. There was crypto and there was FinTech, and those were all exciting marketplaces, very exciting. But for me personally, my strength and experience and scale lined up really nicely with building business enterprise-scale SaaS software. And that also beautifully coincided with some of these leading customers we were talking to who had real problems that we thought we could uniquely solve that no one else was paying attention to.
Soma: Love to hear that confidence, Viraj. Both you having in yourself as well as your co-founders and what signals you’re seeing in the market. I don’t know if you remember this, but we had you and Linda on this show a little while ago. And at the time, we were talking a bunch about what goes into making a great founding team and how do you find people that are aligned or bound by a common mission and vision and there was a great story that you guys talked about what you were looking for, what were some of the mishaps you had along the way, and how you ended up with the founding team that you have now and, and, and all that fun stuff. If you fast forward two years from then, how do you think that journey is going? Do you still feel the same level of excitement and energy around the founding team, or do you wish you had done anything differently?
Viraj: There are plenty of things I could have done differently, but I feel like, all in all, I feel fortunate having the founding partners, but also, the rest of our team has just been phenomenal working with all of the Roomies. I wouldn’t change the team for anything. I think we have a great crew. One thing that I think summarizes how we’ve operated over the last few years is just relentlessness and focus on executing with velocity. I feel like these two have been pretty consistent. As with every step of scale, we’ve had to change our approach, but we haven’t changed both of these just focusing relentlessly on customers and their problems and then internally focusing relentlessly on speed of execution. And I think both of those together have been really impactful in helping us get to where we are. So I definitely wouldn’t change those. Anyone who says they wouldn’t change anything about the past is generally not seeing the whole picture. So obviously, there are things, but all in all, I feel like we’re positioned to do well as long as we continue focusing on the things that matter.
Soma: I always say, Viraj, that having a great founding team is sort of half the battle won. And I have a variety of companies that I work with and I look at like the forwarding team that you guys are and have put together — I feel really good about what you guys have done. People talk about ICP, or Ideal Customer Profile, but I want to take a step back and ask you what kinds of companies do you think need engagement with their communities. And how do you think it impacts the business? Is it all around customer satisfaction, or it does go beyond and say like, “Hey, I can help you with the top line, I can help you with the bottom line, I can help you with product adoption. I can help you with this, I can help you with that.” How do you think about that?
Viraj: Broadly speaking, this is important for every company out there because it all starts with the definition of community. I think it’s very easy to try and paint a very narrow picture of what a community is, but really you think about your community as existing users, future users, people who engage with your brand, people who have heard about your company, but not really used it. You can sort of encompass all of these and then build a bottom-up community strategy. And I think the answer goes way beyond, you know, just the customer satisfaction part of it. That’s kind of the bottom of the funnel in many ways. I think every company in the world really needs to think about how they can accelerate their own growth with data and insights unlocked from their broad community of users, not just like a very narrow definition of social media community or forum community, or Slack community. Companies that do this right unlock all sorts of superpowers, not just in terms of growing their top line and bottom line and absolute dollar numbers, but also getting really high-quality signal from people out there about problems that need to be solved that they may not have on their radar. Or identifying some of your most active champions in different parts of the world, who exhibit behaviors that are not easily spotted using conventional tools.
Soma: I’ll tell you from my vantage point, one of the things that I thought you guys did a great job, even in the last 12 or 18 months, is with the kinds of customers you’ve been able to sign up to start using Common Room and to see the benefits of Common Room community engagement. You go down that list, it’s literally the who’s who of technology customers and logos. And I would say that’s a fantastic place to be in because it’s one of these things where you get the leading companies then others follow fast. Was that a strategic imperative that you guys took or how did you guys end up with literally a phenomenal set of logos and customers, even in as early stage of a company as you are?
Viraj: That’s been a key focus for us from the early days, making sure that we have some of the best thought leaders in our space. And I think that’s been key. When partnering with early customers, it’s important to identify companies that are seeing the future the same way you do. So many of the logos you see on our website and using Common Room embody that — they are at the bleeding edge of how to engage with their community, how to leverage their community, and how to grow and cultivate champions. So it was a very deliberate decision on our end to identify who we think these companies are. And then it’s been almost a “practice what you preach,” right? Once you work with people who are really aligned with your vision, they then help sort of promote your product and your vision to their peers. Who, by definition, are other leading logos. So it’s been really helpful for us to sort of use what we’re helping our customers do on our own to grow our customer base. So between that and the networks we’ve been able to unlock, obviously from the team here, but also from our investors and partners, it is been very deliberate and I think it’s been paying off, at least so far.
Soma: Can we now go one step further and talk about a couple of specific examples? I know for example that Figma and OpenAI are a couple of your customers. Can you specifically talk about what they do with Common Room?
Viraj: We’ve been fortunate to work with some of the best companies out there and some incredibly well-known logos. Each one obviously has a different focus, but there’s a pretty common overlap of use cases across all of these. Since you mentioned Figma and OpenAI, I can chat a bit more about those companies. Both of these are very community-first in terms of not just a community of users but also a community of practice, where, you know, “Hey, we are building a product and obviously we want users to use and champion our product. But independently of that, we also want to build a very robust community of designers for Figma who speak with each other, bounce ideas, and help each other grow and nurture. And then for OpenAI, you know, a bunch of researchers and people on the forefront of AI practice. And then tying that back out to business outcomes.
So when Figma launches a new feature, how do they go and reach out to their champions who’ve been requesting that feature to help them spread the word and generate content? How do they host the best events geographically across the world, bringing together in real life or online people who are their biggest champions, people who’ve been generating content independently and talking about Figma features?
Similarly for OpenAI, as they’ve gotten to where they are, they’ve had several versions — GPT-2, GPT-3. Along the way, they’ve had communities they’ve developed on Discord and forums, where they discuss best practices about how they take this really nascent technology and then help unlock powerful use cases amongst each other. But then also having the company collaborate with them. So how do you make sure that people at the company are paying attention to these conversations going on?
Both of these companies have the fortunate problem of just having so much engagement that one thing that’s been powerful for them is being able to unlock signal from the noise, right? When you have something as powerful as GPT-3 and ChatGPT unleashed in the world, everybody’s talking about it and we have some of the most significant community growth we’re seeing amongst any of our customers in these companies. But from a company’s perspective, how do you take all of this great activity going on everywhere and extract the things that actually matter to you, which is where we use a lot of machine learning models of our own, but also foundational models from OpenAI itself to help.
Soma: One of the things that I’ve heard you and Linda talk a lot about recently is the intelligence layer that you guys are building as part of Common Room. Tell us a little bit about why do you think that is critical to what you’re building. And, more importantly, how do you go about building it into your platform?
Viraj: Once you start collecting information about all of the conversations happening across various digital channels. For some of the best companies out there, the volume of conversations happening is pretty overwhelming. The typical company will have social media presence with Twitter and Facebook and so on, conversations on Reddit. Then they’ll have closed forums, for example, where people are asking usage questions or questions about product, or open and or closed conversational communities like a Slack or a Discord server. Then you have technical conversations in GitHub and Stack Overflow. Plus, you have a lot of your internal systems with your CRM and your product usage data.
When you start thinking about each of these merging with the other, the amount of data you have starts to exponentially grow. And then being able to convert that into meaningful signal is where some of the most impactful outcomes happen. So where we’ve invested a lot in the intelligence aspects of Common Room are on the axis of community members. Once you have members in your community, really understanding who they are and building a 360 profile for them across all of these various channels, that’s one layer. The other one is around the activity that they represent going on. So someone publishes a YouTube video, someone else has a post on a forum, and someone else has a GitHub pull request or a GitHub issue — how do you take all of these as part of that 360 profile and then help paint a very clear picture of what sentiment is this member expressing? What are the key frustrations? What topics are they talking about? What are the different categories of conversations they’re having on all of these platforms?
So, intelligence about the members, intelligence about their activities. And then on a third axis, you can think about intelligence about businesses who are likely to buy your product or the propensity of businesses who are either existing customers or future customers — and how that interacts with the previous two. So, a business entity is made up of members who are having conversations. How do you build a model that helps you see the propensity of conversion or propensity of churn or propensity of upsell? And all of this can be derived from various signals for each channel there are different signals. So from day one, our focus has been not just on collecting data, because collecting data is a starting point. But how do you unlock key insights and outcomes from that data and then drive actions based on those insights and outcomes?
Soma: I really like how you framed it, Viraj. Talking about it as intelligence on people, intelligence on activities, and intelligence on outcomes. Now, switching to how did you decide what approach you are going to take with your models or with the models that you’re going to use? How do you approach training, you know, tuning of these models, are you using any of the currently really popular foundation models or are you thinking of building your own foundation models? How, how, how are you thinking about all of this?
Viraj: Yeah, it’s a combination of both. We have certain layers of intelligence that leverage custom ML models built with a pretty standard tech stack — you know, XG boost on SageMaker and feature engineering in-house. And then we also go leverage some cutting-edge foundation models, for example, OpenAI’s Da Vinci model but then help fine-tune it to perform ideally for our use cases and then also help us scale across our production data.
So from a custom ML capability perspective, we have a bunch of features we’ve built around the ability to auto merge members and organizations across various signals that we have about them. So, Common Room integrates with Slack and Twitter, Discord, GitHub, Stack Overflow, or for LinkedIn, Meetup, and dozens more. And the same person may have different profiles across all of these. The same person may have different conversations on each of these, plus internal systems like Salesforce and HubSpot. So we’ve built custom ML models that use signals from all of these different sources that we’ve trained in-house, using some of those technologies I mentioned earlier. Then we’ve built out a scoring that allows us to go say, “Hey, look, with a high degree of confidence, we think that this person is the same regardless of having a different name here or a different avatar image there, or a different email address”. So there’s an aspect to custom machine learning models that we’ve built for the use cases of merging members or merging organizations.
Then there is another use for custom models around propensity. Once you see a community-qualified lead of some sort, either through your CRM or either through community activity — how do you go build a model that helps predict a propensity of certain outcomes? Like, “Hey, you know, this organization is ripe to adopt your technology based on their champion behavior”. Or here’s one that’s likely to churn, so please go invest some time in making sure they don’t. So these two are examples where we’ve built a bunch of in-house models. But where the world is really exciting now with some of these foundation models is NLP and LLMs, providing a capability that just didn’t exist until recently, where you can go and quickly extract sentiment or extract conversational topics that are not necessarily keyword search. Or even categorize conversations as, “Hey, here’s a conversation about a feature request, or here’s somebody asking for support, or here’s somebody complimenting your product, you know, maybe you want to use that in marketing material”. So this is where we use foundation models by companies like Amazon or OpenAI. But in order to scale them for production use cases, we have to be able to fine-tune them. So, you know, OpenAI has fine-tuning capabilities, so we’ve been able to take the Da Vinci foundation model and fine-tune it for our use case, both as a performance optimization for better performance for our specific customer base, but also from a cost optimization so that we can actually go apply these models in a scalable way across our entire user base. Because without these, it can get really costly. It’s very easy to put on an exciting demo that leverages the hot new foundational model, which is great for a weekend project or with like toy data. But the minute you want to scale it to the kind of customer base that we have, or beyond that, you have to start worrying about the practicalities of, you know, downtime. If these hosted models have downtime, you don’t want to have downtime yourself. Or cost, if you are going to just simply pass through all of that cost, it’s going to become really expensive for you or for your customers.
The other one obviously around is precision and recall, right? A lot of the foundational models are built for general-purpose use cases, and they do a phenomenal job for them. At the end of the day, your specific use cases are going to be slightly more nuanced, and so how do you tune those so that your precision and recall are both even better for your customers? That’s where we’ve spent a bunch of time investing. I know generative AI is sort of the buzzword of the day, and we have some like, pretty clever ideas. But even before you go there, there are so many powerful things you can do with just unlocking capabilities that don’t need generative AI capabilities is just extracting signal from noise in interesting and meaningful ways. There’s some huge opportunity there as well.
Soma: Whenever you talk about Open AI today, most people immediately jumped either thinking about GPT-3 or ChatGPT kind of thing. And the fact that you are sort of not necessarily using that, but you’re fine-tuning a model to make it work for what you are looking for and do it in a cost-effective way that’s great to hear.
It’ll be interesting to hear how is OpenAI helping startups like yourselves. There are people who use ChatGPT. There are people who use GPT-3, and that’s sort of one set of people. And then, for people like you, has OpenAI been helpful?
Viraj: Yeah, absolutely. We’ve been partnering with them since the early days. We’re fortunate enough to have worked with some of the early OpenAI team, and it’s been really interesting to explore how to take some of these research and exploratory models and help commercialize them. OpenAI has different tiers of models internally. There’s Ada, Da Vinci, and several others. Each of them have a different cost, they have different performance characteristics. They have different use cases that they’re optimal for. And we’ve had a pretty open channel with them just in terms of trying new things before they are available to the general public or giving feedback both ways on what’s working, what’s not working. On pricing models, etc. So it’s been, it’s been extremely collaborative for us since the early days. And part of it also is walking the walk. We are OpenAI’s community, along with every other developer out there who’s dabbling in their technology. So making sure that they have the ability to get feedback at scale is pretty important to them. And so I’m, I’m glad we’re able to make that happen
Soma: You did mention a little bit about cost, and in today’s economic climate, managing the burn rate is super critical for every startup of every company for that matter. There is so much hype and buzz and excitement and craze around this generative AI and everybody’s experimenting with that in one way, shape, or form. And the cost could add up pretty quickly before you realize what’s going on. Do you guys feel like you are encountering that, or do you feel like your approach has enabled you to sort of stay ahead of the curve?
Viraj: One example of one of the models we use is it’s 10 x cheaper than off-the-shelf models that we could just pass through to. And a lot of that is the result of fine-tuning that I mentioned earlier. It helps us not just get higher quality results than just basic prompt design, but it helps us train the models so that it can optimize for our use cases without all of the extra cost. And then, from a deployability perspective, this just helps us deploy it in a way that takes a lot of the critical dependencies in our control as well. So monitoring cost, I think, is super important, especially as you are making some of these foundational capabilities available to customers. Because for some of the problems we solve, activity can change wildly. So, if a customer has a conference, you’ll get a week where there is a huge spike in activity, which will obviously drive a whole bunch of additional cost if it’s not built in a way to sort of foresee that event happening already. So if you build a company and then actually deploy it to production where you’re simply passing everything down to some foundational model, be it OpenAI or whatever else, you are likely to be in for a surprise if there’s a variance in volume that you’re driving, which is where some of the lessons to focus on our, like, “Hey, how can we keep our costs under control while still making sure you leverage some of the most exciting capabilities out there”.
Soma: Before we wrap up, Viraj, are there some hurdles you’ve run into when getting the company off the ground into where it is today? And what did you do to get over those hurdles, that might be helpful lessons for other people who are coming from behind?
Viraj: I think there is a level of paralysis that can happen if you try and game theory out every potential outcome. Even in your product — you could hypothesize till the end of the world around what customers actually want and what they’re saying, what they’re not saying, but nothing beats shipping product and watching customers use it or not use it. I think being comfortable shipping things at extremely high velocity with high quality, and that’s, that’s a hard one to balance. So my advice would be to have strong conviction within the team, not just the founding team but the broader team, around your expectations for what it means to ship. What it means to ship quality software, right? You don’t always want to throw stuff over the fence and say, we ship a lot of code. But also, you have to have some ability to ship an MVP. And so develop a consistent understanding internally within your team of what is and isn’t acceptable for who you are as a company. And then live that day in, day out. Many companies will say, “Oh, we should embrace failure,” but then they don’t actually embrace failure. Or many companies will say, “Hey, we should like ship MVPs.” But then when you ship an MVP, they point out all the a hundred things that are broken. And so, clearly defining how you want to operate as a company and then backing it up with the actual execution of how you work, I think, is important. There’s no right answer. There’s no single answer that works for every company. But I feel like each company needs to have a well-understood definition of what, how they ship, and what they ship.
Soma: That’s awesome. That’s, that’s a great answer as, as people think about getting off the ground kind of thing and, and sort of going through their execution environment and, more importantly, the culture. Because I think sometimes these things all come together, and you really need to think about these different pieces of the puzzle and how they come together as you build and scale a team. So with that, Viraj, I do want to say thank you for being with us, it’s fun talking to you. As much as I’ve been part of the Common Room journey from day one, just hearing it, and some of it is rehearing, it just gives me a lot of energy and excitement for what you guys are and what you guys are doing. Thank you again for being here.
Viraj: Absolutely. It’s been great so far. I’m looking forward to more fun times ahead.
Coral: Thank you for listening to this week’s episode of Founded & Funded. If you’re interested in learning more about Common Room, please visit commonroom.io. Thank you again for listening, and tune in in a couple of weeks for an IA40 Spotlight Episode of Founded & Funded