Tom Arbuthnot: Welcome back to the Teams Insider Podcast. It is a pleasure this week to have Adam Jacobs on the podcast. It's amazing — I haven't had him on yet. He's a good friend. Him and I established the Microsoft UC User Group London back when he lived in the UK, which has been going for almost 14 years now. I think I said 10 on the pod, but an amazing amount of time. Adam now works in the IC3 team at Microsoft, which is the backend infrastructure behind services like Teams. We talk about his role, some of the features he's been working on, AI's impact and AI Cogs, and all the exciting features that AI enables in comms and communication. And also a little bit about his history previously working in financial services and then working for Poly as a Devices OEM. Many thanks to Adam for jumping on the podcast — really interesting conversation. And also many thanks to Pure IP who are the sponsor of this podcast. Really appreciate all their support of everything we do. On with the show.
Hi everybody. Welcome back to the podcast. First time we've had Adam on the show, which is ridiculous cause we go back literally the longest, as we'll talk about. Adam and I co-established one of the User Groups in London that's still going strong, well over a decade later — actually quite a lot over a decade now. But Adam, let's start off with a bit of an intro — who you are — and then we'll go back through the history.
Adam Jacobs: Sure. So I've been at Microsoft now coming up on — well, it'll be four years in April. And I work in a part of the org which is not often known about, because you usually interview and hear from people that work directly on Teams — people like Ilya and whatnot. They're sort of focused on devices and client-side experiences. I work in a group called IC3. I can't remember what the acronym stands for right now, but basically we are responsible for the platform behind Teams and other communications components within Microsoft. I'm specifically in the real-time communications part of IC3. It's interesting — we work very closely with the people on the client side, because obviously when they want something from a client perspective to deliver an experience, more often than not they need backend and platform changes to light up the end-to-end experience. So we work quite hand in hand. But yeah, that's kind of my area of expertise.
Tom Arbuthnot: Yeah, so it's a really interesting area. Obviously on the podcast we often have people on the front end. We've had Mahendra on the podcast a few times talking about the wider platform and the scale. But it is an incredible scale that obviously your team are supporting in terms of concurrency of minutes and sessions and video and everything else.
Adam Jacobs: It is, and it's amazing. I have to think about things that possibly aren't always thought through by other product managers. For example, when we release something — and it might just be adding a new internal partner to an existing feature — we have to look at what's the cost impact gonna be. And sometimes it can be quite extreme. So I have to make sure that those things add up. I'm a regular on our — we have a Cogs council internally where our CVP reviews where we're at with various things, and AI features are very much in the upper end of costs. I work on a number of AI features, and as a consequence I've become quite the regular at our Cogs council.
Tom Arbuthnot: It's interesting — when you multiply any feature by 320 million monthly active users, then it adds up.
Adam Jacobs: Yeah. And actually, I'm part of Mahendra's group, so if you've interviewed Mahendra then I'm sure he's already given you a good grounding in my area and what we do.
Tom Arbuthnot: So I want to get into some of the AI stuff. I know you can't talk about all of that right now, but you were also responsible — I think you're still responsible — for Cloud Video Interop. It'd be great to talk about that and what that does, because that's partner integration as well, isn't it?
Adam Jacobs: Yeah. So I am still the PM for Cloud Video Interop, and it's an area that I think we didn't think would exist at this point in time a long time ago. We always kind of thought about Cloud Video Interop as, well, it's a bandaid — it will be around for two, maybe three years. But yeah, we still have a number of customers that rely heavily on it. And believe it or not, we're still even building features. I mean, they're not groundbreaking features — for example, we literally just rolled out support for being able to see PowerPoint live content. And there's other features we are looking at because we do have some very large financial services customers that still rely heavily on Cloud Video Interop. We actually haven't seen much of a decline in that area from a traffic standpoint. So yeah, it keeps me busy when I'm not working on AI stuff, but it's not the sort of sexy, shiny new things that we're trying to get out to customers right now.
Tom Arbuthnot: So I'm really interested from your perspective — obviously you've seen over the four years the kind of focus on AI at Microsoft. They were doing stuff before then, but it's really come into focus in the last four years. What's been your perspective on that in IC3, in your role?
Adam Jacobs: So for me, in my role, it's really been about meeting the scale and the needs of our customers now that AI really has become a big deal and a primetime feature for us. We've had to basically go back, look at our platform — and I think the same could be said for any technology company. When AI became a thing, a lot of people went fast and quick, and sometimes when that happens you don't always build things for efficiency. And when I say efficiency, I mean things like efficiency around costs and compute and whatnot. I think now that we're at a point where this has become a really serious offering and something that we're investing in extremely heavily — as I'm sure you're aware — we've had to go back and really take our platforms to the next level. And try and look at, well, where are we duplicating things? Where have we got multiple teams doing the same thing, when in actual fact we should be doing it once and doing it in the right way at scale.
One example for you is I've been working quite heavily on moving transcripts to a new platform which can scale more significantly and also offer more insights beyond just transcripts. In meetings, transcripts are one of the biggest pieces of grounding data. So that's been a big thing for me, wrapping up a project associated with that. And as part of that project, we're bringing transcripts to other clouds where they weren't previously available. It's one of those things where it's like a big housekeeping exercise — it doesn't sound really exciting, but it plays a big part into AI and the M365 Copilot strategy.
Tom Arbuthnot: No, it's huge. We had on the pod — I think the PM on the Teams side for some of the recording and transcription. And obviously it's hugely become important with AI, and suddenly everybody really cares about transcriptions and transcription quality. And I presume suddenly usage has gone through the roof, because I'm now transcribing everything I can all the time because I want to go back with Copilot later and possibly use it.
Adam Jacobs: Yeah, a hundred percent. It'll come as no surprise that I think it's become — it's up there as one of our big IC3 costs, because yeah, now everyone's using it. And it's not just transcripts — there's technologies like facial recognition and voice recognition, which we have, and it's all layered on top of our transcription services. Those things have also seen much more adoption, especially now that we have things like Cloud IntelliFrame, which will not only do what it did in the past — framing people when they're in Teams Rooms when they don't have intelligent cameras — but now it will also recognise who's in the room and update the roster. All those things are kind of linked around our transcription story, and it just means we're sending additional metadata. We are trying to also roll that out at a larger scale as well, but again, there are some other efficiencies that we're looking at so that we can roll this out more broadly to more and more customers.
Tom Arbuthnot: Is it the right thing to think about it in kind of two camps — as in, there's post-call processing and that's one set of costs, and then there's real-time things like Facilitator and face recognition? Because when I talk to the contact centres, they talk about it as being very different — trying to be real-time versus post-session.
Adam Jacobs: Yeah, this is an interesting area. Largely speaking, we treat everything the same right now, and what I mean by that is that we try and do things as real-time as possible. So for example, if you are transcribing or recording a meeting, at the end of the meeting or shortly thereafter you'll get the recording presented, the recaps, the transcription. Customers don't always need things straight away — we currently focus on doing things as and when, in real time. That might change as we try and look at various efficiencies. But clearly when it comes to things like AI skills, we need to be as quick as possible and do things — in fact, we even need to do things quicker than we do them today.
For example, in the case of a transcript, quite often a transcript will capture an utterance and then it will write that utterance over in one piece so you have a complete sentence of what someone said, and that's contextual and that's important. But for things like other intelligent bots where you want to speak in real time, you need to be even faster. So there are other transcription technologies that we're looking at to see whether we can do things quicker than we're doing things now.
Tom Arbuthnot: The real-time voice models are super interesting to me. Some of the stuff I've been playing with — things like the idea of a virtual receptionist, virtual attendant type stuff — and there's various platforms that have different voice models today. But again, that is real, real-time, because now you are interacting with people, so you've got to take the voice, process it, and respond in a humanlike pace for it to be the best experience.
Adam Jacobs: Yeah. And unfortunately that comes with tradeoffs insofar as accuracy is concerned, because the quicker you process stuff, the less context you get and therefore you can make mistakes. So sometimes you have to do a bit of both and try and do things in one pass, and then when you can, go back and correct things. You'll often see that when you do captions in meetings — you'll see things get corrected in real time as and when the context comes in.
Tom Arbuthnot: Yeah, it's interesting how constrained or not constrained the use cases are as well. So I guess a virtual receptionist — there's a finite amount of variables there realistically — but a Teams meeting could literally be about anything and everything, in multiple languages, all over the world.
Adam Jacobs: That's right. Yeah, exactly.
Tom Arbuthnot: And — can you just give us a little bit — just to kind of wrap the conversation up, I want to talk back to the olden days. So you were customer-side and then you were OEM partner-side before you came to Microsoft. Take us a little bit through that journey.
Adam Jacobs: Yeah. So I guess a lot of my grounding came through the fact that I was originally in the IT industry. I worked mainly in financial services, and I eventually became an IT Director at a large financial services organisation in the UK. I just got more and more interested in, at the time, unified communications. We were one of the first sites to integrate Office Communications Server with a Nortel PBX using things like remote call control at the time — things that don't exist anymore, but at the time this was like really cutting edge, or maybe bleeding edge, stuff.
I started talking more about it — I was blogging and speaking at a few, actually, Microsoft events at the time. And then I landed at Polycom, and I started working on the devices directly, which I was previously purchasing and deploying. I was at Polycom for the best part of 10 years, and that's when I moved over to the US. I worked on a number of different devices at Polycom, which I'm sure you'll remember well — we had the Lync Rooms System and a number of phones. I did a lot of work on VVX phones and HDXs at the time, which had that sort of semi-native integration with Lync. And yeah, it gave me a good understanding about video and real-time communications, which is ultimately why I've landed at Microsoft.
Tom Arbuthnot: It's really interesting that you've gone from the customer buyer to the producer of the solutions, working with Microsoft as a partner, to being on the Microsoft journey. I feel like you've seen all stages of the service, if that makes sense.
Adam Jacobs: Yeah, and it's interesting — I've been on the other side when I was a partner working on building devices, as I mentioned. And now, with the Cloud Video Interop program, I was at one of the partners — actually Polycom was one of the first partners to start with Cloud Video Interop, certainly when it came to Skype for Business Online. And now I get to work with the partners directly. As strange as that was, it's kind of worked out well because I understand what they're building, I understand their challenges, and we have a great rapport. I work very closely with the folks at Pexip and Cisco. And yeah, it makes things a lot easier.
Tom Arbuthnot: I feel like you've got an understanding of what they're dealing with as well as what you are dealing with internally, which is always nice. Microsoft's such a big org — I feel like people can kind of get wrapped up in the inside of it. So it's nice to have someone who feels the pressures of being in the ecosystem, or in the end customer for that matter.
Adam Jacobs: Yeah. I mean, I've always been quite customer-focused, but sometimes I probably focus a little too much on the customer. But I know what it's like to be the customer. I know what a customer expects, and I know how hard it is for large tech companies to meet customers' expectations. But we should always try, and I do.
Tom Arbuthnot: Awesome. Well, Adam, thanks so much for giving us a bit of a behind-the-scenes look at what's going on. Really interesting perspective on the AI stuff too. I will see you virtually at the next user group and hopefully in person soon. And I'm sure we'll have you back on the pod again when you can talk more about the stuff that you can't quite talk about yet.
Adam Jacobs: Sounds good. Tom, thank you so much for having me. I really appreciate it.