Lance Glinn:

Welcome to another episode of the Inside the ICE House podcast. Today's guest is Sam Liang. He is the founder and CEO of Otter.ai. Sam, thanks so much for joining us inside The ICE House. Happy to have you here.

Sam Liang:

Thank you for having me here.

Lance Glinn:

So, Otter.ai just made a big announcement evolving from what people know, obviously as an AI note taker, into what you're calling a conversational knowledge engine. So for those who may think of Otter as something that simply transcribes meetings, what does this announcement represent? And why does this moment sort of feel like an inflection point for the company?

Sam Liang:

Yeah, we just announced it, but we've been building this over the last 10 years. We created the AI meeting note taker, or AI meeting assistant category. We have processed billions of meetings, but the individual meeting notes are never the end goal. The end goal is to create a system that organize thousands or millions of meetings and created this longitudinal knowledge graph that connect thousands of meetings, thousands of people, connect the knowledge, and make that available for both human beings and now AI agents to execute workflows.

Lance Glinn:

So, you said you've been building this for the last 10 years. Obviously, when you started in, I guess 2016 now, because it's 2026, did you foresee or have a feeling that 10 years later we'd have sort of this AI, genAI boom that we've seen over the last two or three years? Did you foresee or at least say, "Hmm, maybe in a few years we could see this whole AI thing really start to take off more mainstream"? Or was it sort of not in your mind and you were just building to innovate?

Sam Liang:

Obviously, I wasn't that smart to predict that LOM had this huge breakthrough a few years back, but we always had the face that it will happen. We didn't know when it will happen, but we always believe that there is such a need to create a new system of record for conversational data. If you think about it, there's a system record for sales data, which is CRM, there's a system record for human resource system, there are ERPs for financial data, but there was never a system of record for conversational data. For most enterprise employees, people probably didn't realize most of them spend more than 50% of their time in meetings. So they're using voice. They're either speaking or they're listening. So, business are run on meetings. It starts with CEOs. They talk about their annual goals, their revenue goals. Then they talk about, "Okay, these are the milestones, these are the revenue objectives."

And then the VPs, the directors, the managers, at every level, there are tons of meetings. Both external meetings, sales team have meetings with customers, marketing team have meetings with marketing agencies, HR or recruiting team do a lot of interviews. And then there's a lot of internal meetings inside each department, and also across all the departments, right? Sales, marketing, product, engineering, design, people. There are tons of collaboration that happen between them. Where did all this data go? Most of them actually are all gone. Although people take manual notes either on a piece of paper in the past, or they type into a document system.

Number one is tedious. Someone has to do that. Number two is incomplete and subjective. Number three, even if people take submitting notes, it's usually fragmented and isolated. Usually, it's not shared with everyone. So, we see tons of opportunities there. We really want to capture the data first. That's why we create the AI meeting and note taker that capture the data. But also, we need to connect them. It's not just individual meeting notes. It already has value which gave you a summary, gave you action items. However, when you connect them, you will see compound value across hundreds, thousands of meetings. And also, over time, the knowledge evolve.

In the past, when people talk about knowledge base, they mostly only think about written documents, Google document, Microsoft document, Notion document, or CRM data. Most people forget that actually most knowledge is generated in meetings using voice. These days, actually, a lot of the written document, or purchasing soon, most of the written document will be generated by AI. Won't be written manually by human anymore, most of them. But people won't stop talking. People will still talk, even though they will probably... Most people will stop writing, because writing is so hard.

Lance Glinn:

Yeah. And I think to that point too, you mentioned those three things that taking those manual notes do. I also think a fourth one is it sort of takes attention away. If I'm talking and having a conversation with you, whether it be face-to-face or over a computer, whatever it might be and you're telling me one thing and I have my pen right here with my paper and I'm writing it down or on my computer typing, I'm hearing you, but I'm not necessarily consuming the information you're saying. I'm sort of just hearing you putting it down without actually taking the time to understand what you're doing. So to add that fourth layer is that it really does sort of take the tension away from the actual conversation and more towards just focusing on, "Okay, well, what did Sam just say?" Or, "Well, what did Lance just say?"

And I know Otter estimates that this opportunity could exceed $100 billion. How did you and your team come to see this sort of massive potential? And what changes when you connect thousands, millions, billions of conversations across people, teams and time instead of just analyzing one meeting at a time? When you bring all that data together, what does it really change?

Sam Liang:

I would say the $100 billion, it sounds big. I would consider that a extremely conservative estimation. There are many reasons. Studies show that their CRM market is over $100 billion. However, CRM only handles sales data. If you look at enterprises, yeah, there are a lot of meetings generated by the sales team. However, there are 10X or, I don't know, 50X or more meetings generated by other departments as well, both external meetings and internal meetings. Well, sales team talk to customers, marketing team talk to marketing agencies, people team and the recruiting team interview candidates. But as I mentioned earlier, there are tons of meetings needs to happen within those departments inside the company, and also across multiple departments, because they have tons of collaboration to do. In addition, not just online meetings, there are tons of in person meetings that happen, not just happening on Zoom, Microsoft Team or Google Meet. They're in person meetings inside conference rooms. There are in person meetings that happen at Starbucks at a restaurant.

In addition, 80% of the world, global workforce actually don't even work at a desk. They work in the field, on the construction sites; in a hotel, at a retail store, or in a restaurant. Those people are having tons of conversations. Those should be captured as well.

Lance Glinn:

How do you do that? How do you capture those? If you're talking to two people on a construction site, for example, how do you capture that conversation? If I come up to you on a construction site, how do you capture that?

Sam Liang:

Yeah. We really have customers actually who work in construction business who use Otter. We have a mobile app on iPhone and Android. You can easily turn it on when you are having a conversation. Eventually, we think, AI should be always on eventually, because you never know when interesting conversation happens. You may run into someone in a hallway, you talk to someone at the coffee machine, right? You want to capture that knowledge.

Lance Glinn:

Absolutely. And Otter's evolution is backed by some new product capabilities, including AI chat connectors, deeper MCP support, and Otters for four desktop capturing conversations. From your perspective, how does this evolution and this innovation illustrate Otter's shift from sort of passive documentation to driving real automated action across an organization?

Sam Liang:

Number one, we need to capture the knowledge. Number two, we need to organize it and share the knowledge between people. Most of the meeting notes today, although there are dozens of other meeting note taker apps, they mostly copy us. They're still staying in this first generation. They're mostly focused on individual meeting notes. And most of them are actually not shared with other people. They're mostly operate in a private mode. If you think about it, if you have a meeting with 10 people, and one person take the notes privately, other people cannot benefit from that. So the best mode should be using AI to take notes and share that with everyone in real time. Otter offers that. Not just sharing one meeting notes, but a lot of meetings are recurring. You can have weekly recruiting meetings with your team. There could be daily stand-ups. And also, with your customers, they're usually biweekly, or monthly, or quarterly meetings. Over time, the knowledge evolve as well.

So you need to capture that and share that. So, when you organize it and make it available for everyone, then you create the opportunities for people to get business intelligence out of this. Not just the meetings they went to themself, but also meetings that they didn't go to. A lot of time for you to do your job, obviously you cannot go to all the meetings. You'd only have so many hours, right? But your job, actually, have dependency on the job of tons of other people. They have their meetings, they talk about how things change. A lot of plans made three months ago, maybe 50% of the assumption have already been changed. You don't necessarily know all of them. So you need AI to help you.

Someone changed the product launch date just two days ago. You are still operating based on the old assumption. So, you may not be notified until three weeks later. Then during that three-week period, you're operating on-

Lance Glinn:

Under one assumption that's not right.

Sam Liang:

Yeah, the wrong assumption. So you could be wasting some of your time. So ideally, AI knows that you need that information. And then just push that information to you. So, that requires the meeting notes to be shared. Of course, they're always confidential information. We do create a number of mechanisms to control that. You can use either private channels or public channel, depends on the-

Lance Glinn:

Sure. You have to have that trust in it, of course, too.

Sam Liang:

Yeah. So this is similar to the Slack channel system, or Microsoft Team chat channel system. They create that system to organize text-based chat messages, but there was not such a workspace channel system for verbal communication. So we created that system for voice and verbal communication. You can share meeting notes in either private channels or public channels. And then you can use Otter.ai to either search across all the meeting notes where you can also aggregate insights with AI to summarize across hundreds of meetings and say, "Hey, anything I need to know to work on my projects."

Lance Glinn:

And so Otter has already years of accumulated conversation history from a lot of the Fortune 500 and the majority of the Forbes Cloud 100. How powerful is that longitudinal data as you roll out this conversational knowledge engine? Having all that previous data and all that history and these thousands, millions, billions of meetings already transcribed and stored, how powerful is that data?

Sam Liang:

It's actually really hard to estimate how powerful it is, because as we mentioned earlier, traditionally almost all that data is lost. Human being by taking manual notes may capture 10%, 20% of that knowledge. Even after its capture is mostly isolated, not everyone have access to it. But with our new conversational knowledge engine system, it makes the data as available and accessible as broadly as possible. Is that going to improve productivity 10 X, 20 X? I think it's very possible. In addition to making human workflow better, it also creates this opportunity for AI agents to execute the workflows.

Today, the way most people use AI is to type a prompt into ChatGPT or cloud. Most people are actually not very good at creating a good prompt. They usually don't give enough context to AI to help you. However, if you use Otter and you might audit to all the meetings, you talk with your colleague for hours in possibly weeks or months. That's tremendous amount of context and data. And if you make that available to AI, AI actually get all the context by just listening to you. So your voice is the prompt. You don't need to work hard to create a prompt anymore, because over months or years, AI will know exactly what you need. AI really understands who you are. And not just you, it understands who your colleagues are as well. And what's the opinion of each person on your team.

So that when AI try to execute certain things, it has a full context. You don't have to give it instructions anymore, because when you talk with your colleague, that's enough instruction because Otter knows what your objectives are, it knows what's your action items; it knows the schedule, it knows who does what. So again, that's the longitudinal knowledge graph. It consists of many entities, consists of people. What's the expertise of each person? What's the role of every person? What are their strengths and weaknesses? Actually, AI already know by using the data over many years. And what are the projects? What are the clients you are working with? Those are the entities. And then the knowledge graph creates the connections between them. It actually even has a voice social graph in your company. Who do you talk to? Who do you talk to the most often? What topics do you discuss with everyone? So we create that graph and make that available to both the humans and the agents.

Lance Glinn:

Yeah. So you're teaching these agents just as humans are having a conversation like you and I are right now?

Sam Liang:

Yeah. You make that knowledge available to both humans and agents. And then agents can become super powerful. So you ask me, how powerful is it going to be? It's actually really hard to estimate today. We haven't seen the full capacity, or the full power yet, but I can only say it's going to be huge.

Lance Glinn:

So today, AI feels like the most crowded platform in tech. There are startups starting up all the time for lack of redundancy there. They're obviously big platforms integrating features similar to one another all the time as well. How do you think about competition now compared to when you started Otter? Because obviously it's changed drastically just over the last few years, but Otter's been around for a long time. So, when you could compare the two, what does competition look like?

Sam Liang:

It's definitely more crowded. They're more competitors. But the way we look at it is this: the more competitors actually help us validate the market, help us educate the market as well. When we created the Otter, a lot of VCs ask us, "Who are your competitors?" I show them a paper notebook and a pen and say, "These are my competitors, because that's the old human behavior." It's very hard to change people's behavior. A lot of people are still using paper and pen.

Lance Glinn:

I have a pen literally with me right now. Not that I'm going to use it to write down all our notes, but it's still prevalence.

Sam Liang:

Or the keyboard. It's our competitor as well, because people are using keyboard to take manual notes. We say, "Hey, eventually, why do you need to do that?" Because AI can use your voice to take notes for you. So all these competitors actually help us educate the market and help the enterprises and the workers to change behavior. So, it's going to be a culture change. It's going to require mindset change.

On the other hand, I see this is still the very early inning for AI revolution. When internet emerged, there was internet bubbles. There's a lot of competitors. Well, a lot of internet-

Lance Glinn:

The dot com boom, is something that everyone always refers to.

Sam Liang:

But now if you look at it 20 years later, there are actually even more internet companies. So, the same thing will happen. You may say, "Oh, there are already so many AI companies." Who knows? A few years later there could be 10 X more.

Lance Glinn:

And you look at what there was compared to what there is now, how do you think about designing a product, designing a company, a brand? How do you think about designing something for the long term versus just designing something for a particular problem in the moment?

Sam Liang:

Yeah, that's something I have been learning. I think we need to create a bridge. We need to target the current workflows first. We need to solve people's current problems first. In the meantime, we need to create something new and lead people to adopt a new habit, and lead people to change their existing workflow. Because when you have AI, when you have agents, a lot of things will change, but it won't happen overnight. So that's one thing we learned, is that, hey, we do need to appeal to people's current habit, but we do need to educate them and teach them that, "Hey, there's a better way. You can do it the way you operate today, but hey, there's something new you can adopt."

Lance Glinn:

Yeah. We talked a little bit about this earlier in the conversation. I had mentioned trust and making sure that there's some level of privacy. You talked about all the mechanisms in place with Otter to make sure that that happens. But as Otter becomes and has become more deeply woven into workflows, how do you think about trust and assistance versus overstepping or taking too much control? Because I think that is a, or it could be a tight line to cross. So, how do you think about having that trust, but making sure that there's nothing overstepping?

Sam Liang:

Trust is super critical. Security is super critical. We absolutely take that very seriously. The security model, if you think about it, is actually no different than the written document, and no different than the chat messages in Slack or Microsoft Team or Zoom. We use similar security models. We encrypt everything. We create a permission system. Google Document has permission system, Microsoft had permission system. You can specify who has access. I mentioned the Slack channels. They're private channels. You can control the membership of each channel and who can access the information in this channel. Of course, for general information, you can put it in a public channel so everyone get access to, so that you can share the knowledge more broadly. In the meantime, we heard a lot from our enterprise customers. By the way, actually, financial institution is our fastest growing vertical. It's sort of ironic because financial institutions have the highest security requirement, the highest compliance requirement, but it turns out that they seems to have the biggest pain and biggest need to use a system like this.

So we work very closely with many financial institutions, and hear and listen to them, and hear what they need in terms of security and confidentiality. So we take a lot of product requirements from them and build that into order. One type of system is a data retention mechanism. So for them, they tell us, "Hey, we need to have a system to specify how long do we need to keep the voice data? How long do we need to keep the transcript? How long do we need to keep the summary?" So everything can be configured. Some companies say, "Hey, we never want to capture the voice data." So we actually, once we transcribe it, we can remove it immediately, so all the voice data can be erased instantly, so that nothing is left behind.

Or for transcript, they say, "Hey, we only want to keep the transcript for two weeks. After that, automatically erase everything." And most of them actually want to keep the summary for longer time, because that's the knowledge that they can search, they can enable AI to execute workflows. So, there are a lot of security system we built into the conversational knowledge engine.

Lance Glinn:

And tell me this. From someone who's been a founder and been working in this space for a long time, what's more important, not only to you, but just to the general customer too, what's more important? Is it being first with a product or a company? Or is it not being first, but then having a better product?

Sam Liang:

I would say both.

Lance Glinn:

It would be naive to say speed doesn't matter, but you also want to make sure that you don't put something out that has problems. So it's sort of walking again that fine line of wanting to be first, but also wanting to make sure you have the best product out there.

Sam Liang:

We want to be the front-runner, we want to be the pioneer. We want to push the limit. So, we want to be the first. As I said, we anticipate 10 years ago that we will need to build a knowledge base, or knowledge engine. So designing system, what has been preparing for that day, of course, the knowledge engine couldn't be built overnight. You have been designing the whole architecture over many years.

And then in terms of being the best product, Otter, when we first came out, we were the first one, and also we won a lot of awards. We won the Apple Editor Choice Award on iPhone, we won Google's Best App award. Of course, we have to continue to improve, we have to continue to evolve the system. For both organic users, we have tons of organic users who just pick up Otter on their own, or they heard about Auto from their friends. But also, we start to see that the enterprises, there's a new top down motion that starts to happen. And we didn't see that two years ago, because most enterprises, they worry about security, they're worried about compliance, but now they are more eager to look for solutions to use AI to improve their collaboration and productivity. So we have to improve the product to appeal to both the organic individual users, and to enterprise customers.

Their CTOs, or chief AI officer, who start to contact us, and discuss how our knowledge engine can help them. So they have very different requirements. For the chief AI officer, they think about the entire company, not just individual person. They think about how my sales team, marketing team, product team, engineering team can all use this product. Absolutely, is this secure. A They need more control. Every company is different. Every company have a different requirement. How much data do you want to share? Some companies want to share more, some companies want to share less. Some companies say, "I can keep the data for a longer period of time," others say, "I only want to keep it for two months." Every company is different. You need to give them the control.

For individuals, it needs to be super intuitive, super easy to learn. But then for individuals, you also need to have etiquette way to show them that if you share the meeting notes with your colleague, you're going to help them. And when they share meeting notes with you, it will also help you. So you need to create that sharing. You need to form a new habit of sharing meeting notes with each other. So, there's definitely so many things for us to improve.

Lance Glinn:

So Sam, as we come to the end of our conversation, you mentioned that this conversational knowledge engine has been in the works for the last decade plus. It's now obviously been released, there's been announcement about it, but now looking several years down the road, what does success look like moving forward for Otter in your eyes?

Sam Liang:

We are building a enterprise engine on top of the Product Lake growth system. We have over 35 million users. These 35 million users are based in, as you mentioned earlier, 80% of the Fortune 500 companies, but now most of them don't have an enterprise contract with us yet. I mean, this are mostly organic users. So we want to put them in enterprise contract, so that those companies can control the data security. And also, once they connect all those meeting notes with each other, and as I mentioned earlier, enabled a lot more people to become more productive, and then create the opportunity for AI agents to execute a lot of workflows for humans. So what we see is, I hope more and more US enterprise start to adopt this, and then also globally.

I was just in London two days ago running my marathon, and so the world record was broken, the two-hour barrier was broken. And before that on Friday, I talked to CNBC and BBC, and we're also expanding to Europe as well where we have added more languages in addition to English, such as French, Spanish and other languages. So, it's going to be a global market. So this is just the beginning. Next three, five years will be even more exciting.

Lance Glinn:

And well, Sam, thank you so much for joining us Inside the ICE House.

Sam Liang:

Thank you. Thank you, Lance, for having me here.