Episode 202: Avish Bhama on Sonia, an artificial intelligence tool that makes meetings more productive
September 16, 2019
Our guest today is Avish Bhama, the CEO of Sonia, where the mission is to build Artificial Intelligence that interprets human intent, and use it to democratize access to information-flow at work. Sonia builds tools that empower you to focus on the interesting and creative nature of your work rather than the repetitive and mundane parts.
When you sign up for a plan with Sonia, you get an AI tool that can dial into your phone calls, take notes, identify the action items that are agreed to in the meeting, and then send an email afterward.
While this sounds a bit sci-fi, it seems that it won’t be long before this sort of AI involvement in the work flow of knowledge workers will be the norm.
Learn more about Sonia at Sonia.ai
Will Bachman: Hello Avish, welcome to the show.
Avish Bhama: Thanks for having me, Will.
Will Bachman: Let’s talk about Sonia. Give me the overview.
Avish Bhama: Sure. Yeah, Sonia is a company that we started about two years ago and it was actually born out of a previous startup that I was working on where we had hired some Bridgewater folks that joined our team and told us that we were the most disorganized company that they had ever worked at. They got us in the habit of recording meetings and they’re kind of notorious for recording meetings over at Bridgewater. We got into the habit of recording meetings and one of our engineers actually created a way to search the audio content of these meetings, and that’s kind of where Sonia started.
What we do is we’ve built a way to dial- in to conference calls and extract things like the audio content. Now we’re working on some pretty interesting machine-learning that’s able to predict human intent during a conversation and auto-complete tasks. If you were to be on a conference call and say something like, “Hey, as a follow up, I will email you the marketing materials and I’ll introduce you to my friend Bill.” What Sonia does is it understands that there’s keywords and key phrases that are said to express intent, and what we do then is predict action items during a meeting and then we can also auto-complete tasks after a meeting to kind of save users time as they are going through meetings.
Will Bachman: That’s very cool, so this actually came out of something that you built for yourself.
Avish Bhama: Yeah, totally. We actually were building it and using it for one of our prior companies that I was working on, and it was used amongst both our engineering team and our sales team at that company.
Will Bachman: That’s very cool, so a few questions around this. First, just the very practical thing like dialing in Sonia. Let’s say that I want to call a contact, let’s call it, let’s say Sarah. I’m going to call Sarah and have a conversation and [inaudible 00:02:09] Sarah as a client. Now does it have to be a kind of a conference call, like a dial-in or do I, can I like call Sonia and then use my iPhone to kind of conference in somebody? How does a practical thing work of getting Sonia on the phone?
Avish Bhama: Yeah, so we actually sit on top of conference calling providers, so whether it’s Zoom or Webex or GoToMeeting or UberConference, or any sort of a third party bridge or dial-in with a pin that you might have on your calendar. What we do is we parse the events on your calendar and can dial-in to the calls that you already automatically have on your calendar. Then we also offer a bridge number on our web app, which is Sonia.com and so we sit on top of the kind of the conference call provider layer and any call that you have, we are able to extract, dial-in to it and extract the audio of it. We start a recording for you and then we’ll also be able to summarize the meeting where we use a machine transcription service to transcribe the meeting using voice-to-text technology.
Then we also can extract the action items using keywords and key phrases that are said during a conference call. It gets … It’s most accurate where there are kind of two callers on a call talking directly with one another and it gets less and less accurate where there’s multiple speakers or you know if people are on speaker phone or there’s ambient noise in the background or people are talking over one another. The general idea is for us to kind of create the ability to sync with their calendar and automatically dial-in to meetings and then kind of email you the meeting notes after the meeting is done, for you to be able to kind of review them at, edit them and circulate them amongst your team.
Will Bachman: Wow, that’s very cool. Just in terms of the practical piece of dialing in, once you get it all kind of set up, let’s say that you regularly use Zoom, then you wouldn’t even have to do anything in particular. It just sort of by default, Sonia is gonna dial-in to any Zoom meeting you have on your calendar.
Avish Bhama: That’s right, and you can edit your preferences on which calls you want Sonia to dial-in to and which calls you don’t want it to dial-in to.
Will Bachman: That’s very cool. Okay, and then after the meeting you’re … You know, I would get an email that says, you know, “Boom, here’s the transcript of the whole discussion,” and up top or something it says, “And here’s some action items that you know, for you, action items for the other person,” so you get that. Then how does the … Let me just first … Is that kind of what happens?
Avish Bhama: Yeah, so after the meeting is done, what we do is we actually use a machine transcription service that goes in and transcribes the meeting. I mean that’s all kind of using a speech-to-text recognition engine that is able to take that speech and convert it to text in real time. That’s usually available between five and 10 minutes after the meeting is done. What we’re now doing is based off of certain keywords and key phrases that are said during that meeting, we not only have the recording and the machine transcript of the meeting available, but we also can predict action items. After a meeting, what happens is you get a recording, a transcription, and then kind of the predicted action items of that specific meeting where we’re able to say, “Hey Will, you just had this call, here’s, you know, 12 action items that came out of this meeting.” What we’re kind of working on is getting more and more accurate at understanding what the human intent is during the meeting to help auto-complete tasks after a meeting is done.
Will Bachman: Yeah, I can see how this would be very helpful. I mean, even just the transcription would be pretty handy if you’re a consultant doing interviews let’s say of some experts, so that would be handy by itself. Then the action items, super helpful if you’re kind of just reminding you what each person agreed to. How does the kind of auto-complete part work? If you say, “Okay, I’ll send you my one-pager on that,” or the the other person says, “Okay, well I’ll send you a contractor,” does it, you know, at this point, kind of draft an email or something that has the one-pager attached and it says like, “Hey, as we agreed on the call, here’s the one-pager,” does it give me a chance to review and approve it before it sends it off? How does that all work?
Avish Bhama: What we’re working on right now is kind of a set of actions that users might do after a meeting is done. A set of actions might include things like sending an email or sending out a calendar invite or updating a CRM or sharing something with your team on Slack in regards to the meeting that just happened. The types of integrations that are required to do these things are things like Gmail, Outlook for sending an email, maybe Slack or Microsoft Teams for sharing a meeting content or meeting notes with your team and maybe Salesforce or Asana if you wanted to update a CRM. If you were to be on a call and say something like, “Hey, as a follow-up, I will send over the draft of the update that we’re working on,” or whatever it is.
What we do is we are able to classify that as the user wants to send an email. Obviously, we can’t go in and predict the type of email that you want to send, but what we are able to do is we’re able to kind of serve up a draft-templated email that you had intended to send, and after the meeting we might say, “Hey Will, you just had this call with Avish, here is the recording, here’s the transcription and here’s the seven action items that came out of this meeting and here’s the four draft emails that you should take a look at to review. Review them here and click this to send.” You can kind of review those types of drafts and send them out.
It’s similar to the way that, I don’t know if you use Gmail, but Gmail is kind of able to auto-complete your sentences now based off of your speech and text patterns that you’ve developed over time. That’s kind of what we’re getting better and better at, is just predicting what you intend to do after a meeting, and rather than you dropping the ball or forgetting a follow-up or not able to kind of communicate something with your team right away, we’re able to kind of alleviate that cognitive bandwidth that you would spend on kind of figuring those things out and just kind of serve them out for you.
Will Bachman: Yeah. That’s amazing. Tell me a little bit about the kinds of professionals that are using your tool today and how they’re using it?
Avish Bhama: Yeah, so most of our users right now, we have about 2000 users, about 900 companies on our platform. Most of the users that we have on our platform are management consultants. We have a number of users that are in the enterprise sales industry where they’re kind of selling enterprise software of some degree. Then we have a number of users that are in other industries, other remote industries. Usually teams that are using Sonia are collaborating in some way where they want to share meeting content with one another at an org, so as you can imagine, if you’re at a company and you are interested in going to a meeting where let’s say you’re double-booked, you might want to read the meeting notes of this meeting after a meeting if you weren’t able to attend.
Conversely, let’s say your entire schedule is just filled with meetings and then you have meetings about other meetings, you know, this is easy for you to be able to kind of upload and download information and kind of transfer it to another team member seamlessly. We work with a lot of remote teams, a lot of consulting groups and a lot of sales professionals.
Will Bachman: Amazing. I can certainly see how this would be a useful tool for particularly an independent consultant where you know, you maybe not don’t have an executive assistant or a more junior team member who can sit on the call and take notes. When I started consulting, it was sort of the typical thing where you’d have two people on a call, one to lead the call and one to take notes and this to some degree can can a substitute for that second person. Tell me a little bit about the kind of pricing model that you have for people that would be interested to try the tool.
Avish Bhama: Yeah, so right now we’re kind of in beta testing mode and we’ve given this out for a select group of users that are giving us regular feedback on the product every day. Now we have a three-month free trial where after three months is up, we are pricing the product in tiered way anywhere from 50 to $150 per seat per month. That’s based off of the types of things that Sonia is able to do for you. You know, for obviously the upper tier, if we’re auto-completing tasks, it’s almost akin to having a personal assistant that is able to kind of save you eight, 10 hours a week of follow-up time. Then if we are just used for people that are reading meeting content, we have a free version available. We kind of start everyone off at a free trial and it’s a premium business the way that our software works.
Will Bachman: Cool, and where would people go to to sign up for that free trial?
Avish Bhama: It’s at Sonia.com, S-O-N-I-A .com.
Will Bachman: Fantastic, and what … Tell me a little bit, I mean, one thing that a lot of people might ask is around security, so tell me how you address that.
Avish Bhama: Yeah, so we work primarily with AWS and that’s where we store all of our meeting data and it’s all encrypted and it’s all fully controlled by the end user. If the end user wants to delete their data, it’s fully in control of them being able to log in and do so. It’s similar to the way that we would kind of handle emails, you know, it’s all kind of fully managed by AWS and in control of the end user being able to do so.
Will Bachman: What about getting the sort of permission of the other person? Some states you can record a phone call, I believe without … With only one-person consent. Some states require two-person consent, so does Sonia get like announced on the call like, “This call is being recorded,” or how do you typically address that?
Avish Bhama: Yeah, good question. There’s, you know, one-party states and two-party consent states. In two-party consent states, both parties are required to consent for the call to be recorded. What we’ve done is we’ve designed Sonia to announce that at the beginning of each call, it is able to dial-in to that meeting and it announces that it’s going to be on the call recording that meeting and storing a record of the meeting separately. The user is kind of in control of whether that announcement is on or off. If That announcement is off, the onus of responsibility for the user to let the meeting participants know that the call is being recorded is actually on the user. We kind of … The way we deal with it in the product is we have built-in announcement. It’s a quick three or four-second announcement at the beginning of each meeting and it’s kind of easily controlled by the user.
Will Bachman: Fantastic. I’d love to hear a little bit about your story, Avish. Kind of about some of the previous startups that you were involved in before Sonia and how you got to the point where where you’re leading this company.
Avish Bhama: Sure. Yeah, so you know, my background is predominantly in startups. I’ve been working on startups for more than a decade now. My first company was actually in the crypto currency space. We were amongst the very first people to start a Bitcoin exchange way back in 2011 when Bitcoin was very, very difficult to trade. During that time, we had actually recruited a few Bridgewater folks, one of which that invested in the company, and then one of which actually joined our team. When he joined our team, he had kind of got us into the habit of recording all of our meetings as I mentioned, an exceptionally smart guy. We had kind of passively started building habit of just recording meetings and being able to kind of have a central system of record for what is happening at our company, so that if anybody has any questions, we could share that content with folks internally.
As an example, if somebody is having a conversation with a client on the sales team and engineering is curious to understand what the pain points are the client is bringing up with the product, that’s all kind of centrally stored. What we ended up working on was a way to not only store the meeting, but to also search the content of the meeting through searching the audio and then summarize that meeting so that somebody doesn’t have to go in and reading the entire transcript or listening to an entire hour-long meeting. That’s where a lot of our machine-learning came in. We had started working on this years ago and it had kind of become the most useful thing that we had built internally. As we were kind of thinking about this pain point that we were solving for ourselves, we kind of thought, “All right, this is pretty compelling, and this is going to be a company in and of itself.” That’s kind of when we started working on Sonia back in 2017.
Will Bachman: I’m curious to hear, you know, beyond Sonia, what other kind of personal productivity kind of habits do you have that have helped you get things done? It seems like someone who would be involved in building this company and this tool would have figured out some smart ways to more efficient.
Avish Bhama: I’m actually the opposite. I was a very disorganized manager and you know, I was a first-time entrepreneur when I had started that first company and I was not good at organizing notes or taking minutes of meetings or, you know, I stored a lot of information in my head. What happened was, people kept wanting to meet with me to get a download of what I was thinking, and I had to kind of work on my communication skills and work on kind of extracting what was in my brain onto a central Wiki for the entire team to benefit from it. This had kind of been born out of a pain that I had myself where the pain wasn’t necessarily for me directly, but it was for other people around me that didn’t have the same knowledge base.
Conversely, similarly when they were in meetings I wasn’t privy to, I also didn’t have the same knowledge base that they had. In terms of tools that I use today, I obviously use Sonia, I use Asana, we use HubSpot to kind of manage our sales process. I’m kind of an avi- note taker now on my phone where I’m kind of jotting things down regularly when I have an idea of pop up in my head and, and you know, or something that I want to talk about in a conversation with somebody.
Will Bachman: What do you use for that? You use Evernote or some other tool?
Avish Bhama: I currently just email things to myself. I use the Notes app on iPhone and then I’ll kind of screenshot it and email it to myself.
Will Bachman: Okay. Wow. All right. That’s one where I love Evernote. for that kind of purpose. It’s amazing as you probably know. I’m curious, just to backtrack a little bit, what led you to create a Bitcoin exchange in 2011? That’s pretty early days. Someone who bought some Bitcoins then and held them till today, we’d probably be doing pretty well, so that was pretty early on. What got you kind of aware of the whole crypto currency field and wanting to dedicate yourself to building a startup in that space?
Avish Bhama: Sure, yeah. I had actually joined Apple on their foreign exchange team in 2010 and in 2011, I got interested in in not only foreign exchange but crypto currency at that time. I had actually met Brian Armstrong who was working on what became Coinbase, but he was working on an app called Bitbank at the time, and I had met him a few times and he actually is the first person to tell me about Bitcoin and explain it to me in depth. When he did, I got really interested in trading it and he was working on a wallet app, but you know, back in 2011 it was so illiquid, it was very difficult to trade. During that time, the trading in the exchange landscape was so fragmented and illiquid, that it made sense to actually build something to connect liquidity and pools of liquidity through an exchange, and so we had started working on that in 2011 and it was relatively early back then but obviously, things have come a long way since then.
Will Bachman: Sure have. On Sonia, what do you see kind of as the developments over the next couple of years for tools like yours? I mean you can now do kind of the speech recognition and figure out what the action items are. What do you see on the horizon over the next couple of years of things that we can expect tools like Sonia to be able to do for us?
Avish Bhama: The big kind of predominant theme that we are actually working on is workflow automation. What that means is, right now every day you have tools that you use, you have people that you interact with, you are constantly perhaps in meetings and the output of the meetings require you to do many follow-ups. In workflow automation, what we’re working on is for the human to be able to expend less energy doing things like repetitive, manual, tedious tasks and for them to be able to create a way for those types of things to be automated so that you can spend more of your day and more of your time working on the more creative work that you do on a regular basis. You kind of … If you look at like my workday, I spend a lot of time in meetings, we have to update things like a CRM so that I can kind of share meeting notes across organizations and across kind of teams and functions at my company. Then updating things like Salesforce or HubSpot or doing follow-up emails. A Lot of those will will get dropped if you’re not kind of on the ball.
What we’re working on is ways to kind of automate that over a period of time where you show up to work and it’ll say, “Hey Will, yesterday you just had eight calls. Here are the 12 follow-ups that came out of these eight calls and here is five draft emails that you have to send out today. Please click here to review and click here to send.” It’s able to kind of be a copilot for you at work where it’s able to automate the things that you tediously have to do right now, so that you can kind of spend more of your time doing more creative things at work.
Will Bachman: Well, so you’re taking off my plate, all the stuff that I’m comfortable doing and I know how to do. It sounds like you’re going to be freeing up more of my calendar for the hard stuff. That’s a good thing.
Avish Bhama: Yeah. This is going to be a major wave kind of over the coming decade and a lot of machine-learning and a lot of AI that’s being built is kind of being built in a way that helps alleviate human beings from needing to do tedious things at work. That’s kind of the things that we’re working on and it’s going to be a long road, but that’s the output of the work.
Will Bachman: Fantastic. Well, Avish, it sounds like a really cool tool for … And I encourage listeners to check it out. Sonia.com, S-O-N-I-A .com. Get a three-month free trial. Avish, thanks so much for being on the show.
Avish Bhama: Thanks for having me, Will. I appreciate it.