Your AI Meeting Agents Aren’t Enough: Otter.ai's Sam Liang on Enterprise Knowledge

0:07Hello, everyone, and welcome, humans, to the Neuron podcast. I'm Corey Knowles, editor of the Neuron, and we're joined, as always, by Grant Harvey, writer of the Neuron Daily AI Newsletter. How's it going, Grant? Going well, going well. Today, we've got Sam Liang, the co-founder and CEO of Otter AI. Before Otter, Sam built the blue dot for Google Maps, you know, that thing that tells you where you are. Now he's transcribed over a billion meetings. And last I heard crossed a hundred million in annual revenue, which is pretty awesome with less than 200 people at the time. That's pretty awesome. Welcome to the show, Sam.

0:42Sam Liang:Thank you. Thank you for having me here. Yeah. So you recently announced that Otter is moving beyond being just a meeting notetaker app and building into an enterprise knowledge base with agentic workflows, MCPs in there, kind of trying to take the insights that people get from meetings and then expand that and connect all of your enterprise tools to it. That is a big, awesome strategic shift. Could you explain the thought process there and what all you're doing in the enterprise space? I think it's really awesome. Yeah, of course. I wouldn't use the word shift. I would say it's an evolution. We created the AI meeting notetaker space.

1:25Sam Liang:Basically, we started back in 2016. We launched our first product 2018. Then we built the other AI meeting notetaker that can join your Zoom meeting, Google Meet, Microsoft Team, WebEx. No matter what tool you use, Otter can help you. Can work for online meetings, for offline meetings. If we meet in person at Starbucks or restaurant, we can use other mobile as well. We also recently released other MacBooks. We're working on a Windows version as well. So I would say there are two stages. The stage one is the meeting note taker. The idea is that all of us spend so much time in meetings. There are a lot of data show that enterprise knowledge workers spend at least 30 % of their time in meetings.

2:15Sam Liang:And if you're a manager, If you're VP, you spend maybe 50, 70, 80 % of your time in meetings. Traditionally, all this data is lost. People use a paper notebook. I still have paper notebook in front of me. Or Google Doc or Notion to manually take notes. So we create the AI meeting note taker to automate all of that. Then we see that more and more people are using this, even in Fortune 500 companies, there are tons of people using the order, but most people are still using it as an individual tool. You know, they record something, they keep it to themselves, but the value is way bigger if you aggregate all these meeting notes as a team.

3:14Sam Liang:For ourselves, for example, we have just over 200 people now. We record almost all our meetings in the last eight years. Wow. Sales meetings with customers, marketing meetings, product, project management, design, recruiting, both external meetings and internal meetings. That allows us to operate really efficiently. We organize the meetings in either public channels or private channels, very similar to the way people organize their workspace on Slack. We actually build other workspace using the same model as Slack because we see the similarity between Slack and other. because basically Slack you communicate using text messages, but on order, you know, we capture all the meeting contents where you communicate using voice.

4:17Sam Liang:The similarity is really strong because between Slack and order, you basically talk to the same group of people on the same set of topics. So that's why we build other workspace in a very similar way to Slack. So other workspace allows you to organize and manage all your meeting contents. So effectively it creates a meeting-centric knowledge base. The reason I use a meeting-centric, the reason is interesting is that traditionally when people think about knowledge base, they only think about written documents, like documents in Google Doc, Notion, emails for Slack message, or some data in MySQL CRM.

5:09Sam Liang:People really think about voice data because traditionally all the voice data is all lost. But now with Otter, we help people capture that voice and meeting data. And we create this workspace to help you organize it so that you can find the content in the right relevant channels, or you can search globally. In our company, you can almost find anything in any team, no matter where you work. So most of or content is actually public, the idea is to reduce information silos. Again, in the past, almost no meeting is captured, but now some meetings are captured, but most people still keep it to themselves, keep the meeting notes to themselves so that each team, there's still a war between each team.

6:10Sam Liang:It really slows down information dissemination or propagation that slow down the operation of your team. They create a lot of inefficiency. Yeah. So we see this is why it's important to create a meeting-centric knowledge base. Another reason is this. If you think about it, number one, for most enterprises, actually meetings is the most expensive activity. Most people didn't realize that. Just think about how much time all the team members spend in meetings. Yeah, it's a lot. It is a lot. If you're a manager, if you're a VP, you spend more than 50 % or 70 % of your time in meetings. Effectively, the employer spends most of their money paying people to go to meetings.

7:08Sam Liang:Yeah. Right. You can count up all of the salaries of the people involved, and you can say how expensive the meeting is. Exactly. If you spend 50 % of your time in meetings, effectively, half of your salary is spent in meetings. Traditionally, there's not even a method to evaluate, are these meetings even effective? What's the return on that investment? it. What are the contents that people talk about? Again, people can just, brain can remember a small fraction of the contents of meeting and they forget really fast within the week, probably 90 % of the things they already forgot. So, you know, the facts are pretty clear right now.

8:00AIOT is really reshaping industrial efficiency, security, and decision-making in a very real way. We've all heard the buzz around artificial intelligence of things, but what's interesting is that it's finally moved from promise to performance. Companies are actually seeing measurable improvements now in efficiency, in resilience, in safety, even in decision speed. And the organizations that are leaning in the hardest, they're pulling ahead already. A new global study from SAS digs into this shift looking at how leading manufacturers and energy organizations are combining AI and IoT to run smarter operations, accelerate their company innovation, and build a genuine competitive edge.

8:44And if you're wondering whether now is the time to act, the takeaway is pretty direct. It really is, right now. So if you want to understand what the top performers are doing and how you can accelerate value from AIOT in your own organization, go check out the full findings from SAS and their new study. It's packed with insights you can actually put to work right now. Learn more and find out how to accelerate your AIOT value today with SAS. We've got a full link in the description ready to go for you. Another issue is like, for instance, say you're having a meeting and you're using the built-in Google transcription.

9:19And then, you know, that's nice that it's transcribing everything for you, but it's sitting in a Google Doc. And unless you go back and check that Google Doc, that data is just still as lost as if you had had the meeting and nobody transcribed it.

9:32Sam Liang:It's there, but nobody's doing anything with it. Yeah. That happens to me a lot. Transcribe a meeting and it's just lost. Yeah, unless you share the meeting nodes, you know, with the team, with even cross-functionally, the value of that node is very limited. So it's really important to create that network effect of meeting data. The definition of network effect is that, you know, more content is created, more people have access to it, the higher value it generates. So this is, you know, again, why a meeting-centric knowledge base is so important. Another point I want to make is that I already saw some data that claims, I don't know how true it is, that it claims that 50 % of the written document on the internet, new content, is already generated by AI, rather than written by human manually.

10:30Sam Liang:If you think about it, maybe in a few years, 90, 95 % of the written document will be generated by AI. So people will rarely write themselves. I can see writing being a lost art, yeah, especially as voice transcription gets better. Unless you are a writer by design, I think that is probably the case. Most people are not writers by design. Most people hate writing. Yes. However, everyone talks. Everyone talks. And no matter how advanced AI is, people won't stop talking. That's right. that's true no one no one's gonna make you stop talking yeah yeah right we're talking on this podcast and this podcast will continue that's right right it won't it won't go away um yeah and and then my point is that in the future voice may become the biggest data generator Oh, I can see that.

11:34Sam Liang:Yeah. So this is why a system like a meeting knowledge base is so important. It helps you organize all this authentic data content, authentic voice content. Once you've done that, I know you guys are putting into place some agentic workflows and such. What kinds of applications could a company do once they have that data stored together and accessible? There are tons of applications. The way we look at it is just think about the workflows of every role. What's the workflow of a sales rep or a sales manager? What's the workflow of a recruiter? What's the workflow of a product manager? a user researcher.

12:27Sam Liang:So we think about their workflow. And then once we have this meeting centric knowledge base, we say, hey, how do we leverage this knowledge base to optimize their workflow? Right. For a sales rep, sales rep do a lot of talking. They talk to customers. One of the tedious job of the sales rep is actually after they talk to the customers, is to enter the data into CRM, for the HubSpot. And a lot of them skip it. A lot of them are not very diligent. They either completely skip it or they just write something random. Then they move on to the next meeting because they're busy. They oftentimes have back-to-back meetings with customers.

13:20Sam Liang:But that workflow can already be automated. Other actually, after your sales call, we actually extract the important data or important insights and push that into Salesforce for you. And whatever data you want to extract, you can actually create a template. We have this customized template system allow you to say, after each meeting, there are several methodologies, a MacPick or some other methodology to attract data, like authorities, customer budget, customer use cases, objections that customers have, competitors they mention, and you can configure their template, then we can attract things as a push-in to Salesforce.

14:10Sam Liang:Recruiter is similar after an interview. They need to evaluate. Maybe they have a scorecard. There's five categories, right? You can do that. And those are the external facing meetings, but for internal meetings, we look at our own internal meetings, like product team meeting, engineering team meeting, the project management, there's cross-functional weekly, bi-weekly sync up. So all this data, we capture that, but then we say, hey, okay, after the meeting, what do you need to do next? Do you need to generate document? Do you need to write the email? Do you need to schedule a meeting? Do you need to follow up on the bug?

14:57So this is where Azure can do next is to call the right agents to execute some action

15:07Sam Liang:items for you. And you're doing that agentic connecting in Otter, right? So you're using MCPs or whatever tools on the backend, but you're doing all of the management of the workflows inside Otter? Totally. Totally. That's why we create a public API, create an MCP server. And we're also building MCP clients so we can pull data from other data sources like Google Drive, Gmail, Slack, Jira, Asana, CRM, then, you know, and, you know, correlate that with the meeting data. Right. Again, you know, the volume meeting gets so big. And another advantage of meeting data is actually compared to a written document is usually the most up-to-date information.

16:01Sam Liang:for a startup like us or for any enterprise teams, things change really fast. Anything you wrote a month ago, maybe 30%, 50 % of that already obsolete because things change, the plan change, the customer request change. But the meetings, every day you discuss the most up-to-date information. Yeah. So this is why voice data, you know, oftentimes has higher weight and priority and is fresher. And if anything change, it's really important to actually, if you discuss, oh, customers just told us this, we have to do something different. And then if that meeting note is public or shared with other teams, the marketing team, product team, engineering team, they can all be notified really fast.

17:03Sam Liang:In the past, when things change, that information is not propagated fast enough. So other teams still operate based on the old plan. So that's where a lot of inefficiency happens. So the people who are in the meeting where that decision was made, they know what's happening. They're operating accordingly. But if they miss someone in the shame command, especially in a remote first company, not everyone is going to know that that's the new plan. And that data kind of dies with them. Yeah. This is another thing we're building. We call it a leader dashboard and leader workflow. We think about what leaders do.

17:44Sam Liang:Earlier I talked about sales, recruiter, product manager, but now there's another big role that's the leader. You could be a manager, a director managing three managers, or you can be a VP managing several directors. Leaders tend to be even busier. They manage more projects. They have bigger responsibility. They also manage many people as well. So there's a project management part and there's also people management part. And there's no good tool to help leaders. Actually, and the leaders, again, they spend even bigger percentage of time in meetings. They have even more meetings to go to. And they're often time double booked, triple booked.

18:39Sam Liang:one thing we can actually help them on that problem is actually you can send your other meeting note taker to another meeting you couldn't take where you still get your notes and gain the information from it still so now other can take notes for you but the next step which we are also working on is to build an avatar for everyone so that you can send your avatar to another meeting which can do two things. You can tell your avatar, okay, ask the VP product or ask the marketing guy these three questions for me. And if they have questions, answer those questions on my behalf. So your avatar in the future can actually join meetings, ask questions on your behalf and answer question on our behalf.

19:37I'm just imagining a digital boss just showing up at every meeting. What's the ROI? What's the ROI? How could we lower expenses?

19:44Sam Liang:Yeah, because if you think about it, actually, you know, for some meetings, I usually have a fixed set of questions. But then my avatar can ask a question. So back to the leader dashboard. So because the leader manage more projects and things are always updated, in real time in meetings. So we're planning to build a dashboard for the leaders so that they can see the real time status of each project or the problems, the blockers, or anything change would notify the leaders. And then the leaders can also keep track of, you know, for the people they manage, you know, what are the key things they work out based on the meetings they talk, they join or the topics they discuss.

20:38Sam Liang:So it can give them a lot of data and other visibility on how their team is performing. Oh, wow. That's awesome. I really like the idea of that. Same. What would be interesting, and I don't know how you're going to handle this, is what if somebody says something in a meeting that's incorrect? Is there a way to fact-check stats and data to the actual data that's connected in the dashboard? Potentially, yes. We don't do that yet. But potentially, yes. You can do fact check. If they quote a number, they say, hey, we grow 20 % last quarter, but the order could go into their database and put the data.

21:26Sam Liang:And it could even surface that in real time. Say, hey, maybe, are you sure? Because, you know, I just checked my SQL. It's at 15%, now 20%. One blast in the middle of the meeting. Yeah, that's where I see potentially in the future, the AI note taker actually will become active, not just passively listening. It can talk as well. Wow. Correct me if I'm wrong. Your current meeting agent can talk in meetings, right? We actually already built that for business and enterprise customers. They can ask Otter a question. They can say, hey, Otter, why did we lose that deal? Then it will put data from Salesforce and give you the answer in real time.

22:20Sam Liang:So Otter can already answer questions on the mat. it doesn't talk proactively yet. We see that in the future, Otter can talk proactively, just like another teammate. So basically, we see Otter will become your AI teammate. Not only can take notes, but it can also contribute information in real time as well. Oh, wow. Kind of like C3PO, but for the business, right? Where he's always giving you information that you maybe didn't ask for, but could use in a moment and then, you know, just surface it at the right time. Yeah. Something I'm curious about as well, with permissions, for example, knowing that sometimes like your leadership may have a meeting that they would rather the rest of the people not be privy to.

23:09Is there a way, some kind of a hierarchy in place that? Yeah.

23:14Sam Liang:Yeah. There's something we're building as well. You're totally right. Not all meetings should be shared with everyone. Although there are two things. One is that we have to respect the confidentiality requirement. So we already have a system allow you to configure that, you know, for this type of meeting, you can share it into this public channel. But for this type of meeting, only share it to the private channel or for certain meeting you never share. We already configure that. And the hierarchical system you just mentioned is also important. We're building that system so that we can ingest the organizational chart.

23:56Sam Liang:You know, say, hey, the CEO, the VPs, the directors. So we can incorporate that into the system. So there's certain default sharing hierarchy there. Cool. On the other hand, this is where, you know, culture we see will change. In the past, you know, meeting notes are rarely shared with, even the meeting notes are not necessarily shared with all the meeting attendees. That needs to be fixed first. But then meeting notes are usually useful beyond just the meeting guest. Oftentimes, the content is useful for a broader audience. Yeah, you want to avoid those data silos. Yeah, to avoid that data silos.

24:55Sam Liang:We're building that system, and also that requires corporate culture change. In general, I see that the world is moving into a more transparent, more open system. If you think about in the last 10, 20 years, we're sharing more and more in both our social life and our professional life. For social network, right, people share tons of things on the Internet. Compared to 30 years ago, people are way more open. We see that will happen in enterprise, in professional world as well, because the more you share, the less boundaries have between what... You cut back on the silos, and at the same time, you probably increase alignment across a company to ensure that everyone is zeroed in on the same goals.

25:59and casks ahead of them.

26:03Sam Liang:Absolutely. Like what you just said, increase alignment, yes. Excellent. Well, you know, with your background in engineering at Stanford and having built the blue dot at Google Maps, I feel like you have a good understanding of like large scale systems. What are the current technical bottlenecks in voice AI that maybe people don't realize exist yet? or do you see anything big coming around the corner? I just want to say, you know, I did contribute to the Google Blue Dot. I wouldn't say I'm the creator. I built the backend location platform, you know, tons of other people built the application.

26:49Sam Liang:So... That's fair. That's a good call. I'm sorry for that. That's fair. You know, a little priority, but not so much. Fair enough. The challenges, there's still a lot of challenges. Even, you know, to understand, well, the biggest challenge is actually to fully understand what people are saying. It's actually hard for humans to understand each other. It's true. Then, you know, the first problem or the first goal for us is actually how much can AI understand you? We think that AI does have advantage because of the infinite memory it can have. It's actually, if I'm meeting a stranger, I don't know too much about that person.

27:50Sam Liang:It will be harder for me to understand that person. but um it was so usually you know when i before i meet a stranger you know i i need to do some background some check right to you know i look at this person's linking profile i look at um you know in any articles about him and the podcast about him right so all that take time um but you know i i hope order can help me do that in ms before i meet a stranger the order will go out on the internet and search for any information about this person and find the LinkedIn profile. So before the meeting, it gave me a brief in advance and said, hey, Grant, all the major podcasts you did in the past, any blogs you wrote, any LinkedIn posts you did, it gave me a summary.

28:46Sam Liang:Yeah. Yeah. So so so I can really help people to understand each other better. How do you one thing that seems like it would be particularly challenging to me with voice in a meeting setting is that you might have someone from the northeast. You might have someone from the south, someone in the Pacific Northwest, maybe a couple of people who English is their second language. You're dealing with a lot of dialects sometimes in one call. And that seems like a thing that could be really difficult for an AI to keep moving and keep straight. Or tone, right? Tone, yeah. How do you interpret tone when someone says something?

29:28It's like, are they saying this in a way that is misinterpreted because the words don't match the way they were saying it?

29:37Sam Liang:Exactly. I'm an immigrant myself. My accent is still difficult for some people or even for AI to fully understand myself. So that's on the surface, the superficial part. But then, as Grant you just mentioned, a deeper issue is the meaning, the semantics behind the words. yeah right what kinds of people use the same word to to mean different things uh it depending on your background um totally and you know where you come from so um it usually more contacts can help me understand how people understand each other better um so where do you get all the other uh contacts uh i think ai can really help you do that um this is where the history that the the corporate knowledge base is useful, even between each department, right?