Filter Events:
Barclays Global TMT Conference
Who: Rajesh Jha, EVP, Experiences + Devices
Event: Barclays Global TMT Conference
Date: December 7, 2023
Raimo Lenschow: Hey, welcome to our next session. I'm really happy, Rajesh. The one thing when you have a speaker from Microsoft, it's always like you're such a big organization. And it's a positive thing, like well done, and it's amazing, it's positive.
Rajesh Jha: It's a positive thing if big elephants can dance.
Raimo Lenschow: Yes. Maybe to get everyone on the same page, talk a little bit about yourself and your role in Microsoft.
Rajesh Jha: Yes, I lead the Experiences and Devices in Microsoft, so that would have Microsoft 365, which is Office and Teams, Windows and Surface, and then also our search and our browser efforts.
Raimo Lenschow: And so as you think about the last 12 months, it's been like a crazy journey in terms of new innovation, et cetera, like that. Can you talk a little bit, like how did you experience, how was this last 12 months for you guys.
Rajesh Jha: Yes, first of all, let me just say, as you all know, the last 12 months have been as exciting a time as any. And I've been at Microsoft a long time. I came when it was graphic user interfaces and we thought that was a big deal. Internet, mobile. This last 12 months, I actually feel generative AI is as big a technology shift, if not bigger than any or all of them put together. And so for us in the last 12 months, we got a lot done when I reflect back. So GPT-4 has now been out for introduction for about a year.
So we, in Jan we had APIs in Azure for that with all the safety and trusts that our enterprise customers expect. In February, we took the large model, we put that in our search in a browser because we did think this represented a new way for people to find information to reason over stuff. In March, we took a step back and we felt about it, "What about productivity? How would the generative AI, the follower of these things, affect or impact the way work is done?" So in June we brought a private preview of that for a lot of our enterprise customers. And it's been great to be with them on this journey. November, we launched the Microsoft 365 Copilot.
But when I reflect back, of course GitHub Copilot was the first one that we went after because it was very clear that we could help the developers be incredibly productive, so we had that.
In the last few months, we've also taken a look at how business processes and business workflows can be rethought in the world of generative AI, where you take systems of engagement and you take natural language and natural language interface and how you bring, whether it be Salesforce's workflows, a service provider's workflows, whether it be a doctor's workflow, so we built a Copilot for those things. We've thought about how to extend the Copilot to allow customers and developers to extend these Copilots with specific information.
So a lot's happened. And the thing I'll just tell you is a lot of this technology shift also intersected with the work that we had been doing anyway on Microsoft 365. Post-COVID, we had spent a lot of, we saw a lot of customer engagement intensity, and that populated what we call the Microsoft Graph, which represents the customer's most important database, I think, which is the projects, who knows what, then comes the knowledge, people's calendar, and email, with the right permissioning. So this is where the generative AI reasoning engine meets the Microsoft Graph to provide generative AI in the flow of work. So but still, it's very early, but very, very exciting to see the entire company rally to this.
Raimo Lenschow: And you guys must be thinking a little bit more ahead, and I just wanted to ask a slightly more bigger picture, like do you think now generative AI, what it can do, and you have more insight than us, like how would you think the way we work will change over time, going forward? Is there a glimpse that you could give us there?
Rajesh Jha: Yes. Look, I'm not a soothsayer on this stuff, but let me just share what we are already seeing with customers. So when we did the early adopter program in June with the Microsoft 365 Copilot, we had many hundreds of customers jump on the opportunity, because they see the same thing. And 100 to 200 of those, we've been really deep. They invited Microsoft to go deep with them and analyze the impact on productivity.
So we looked at three dimensions of productivity. One is efficiency, just how quickly can you get the job done, or how much more can you go do? But we balanced that by taking a look at both quality of work, so it's not just enough if AI assists you in doing something quickly. What is the quality of the output in terms of accuracy, in terms of up-leveling the skill level? And the third thing we looked at was effort. How much effort, how much grind did we take out?
And the results have been really impressive, and we are going to publish all of this in the Work Trend Index, I think, in the coming days. But about 68% or 72% of the people felt it made them more efficient, it improved the quality of work, and they felt they had to put less effort and so they had more time to do what is innately human, which is more creative, more reflective work, more collaboration. And 70% of them would say, "Of course the Copilot makes me more, generative AI makes me more productive." But what's also telling us that? More than 55% of the people feel that the AI makes them more creative.
So this is today, here and now. Now you ask the question, how do I see this going forward? I think workflows are going to have to be reimagined. So today what it does is the Copilot or AI helps you in your existing workflows, but when a lot of the grind--and let me just take a step back. In the Copilot, our vision of the Copilot is not an autopilot. It works on your agency, it works with your permissions, it works with your context. So today it's helping you in your existing workflows. But because it's helping everyone, how can an entire group's workflows change and be reimagined? And we are starting to see that, both in my team and with customers.
I want to leave you with one thought. So far AI, prior to generative AI, it was as if the human beings did all the work and the AI was the editor. It would show up in auto-complete, grammar check, spell check. But now it's flipped. AI is doing the work. They're doing the first draft, they're doing the summarization, and the human beings are now editors. So it's gone from AI being editors to now where it should be, which is the human beings are the editors and the AI is the assistant.
Raimo Lenschow: And Rajesh, do you have any--like on that reimagining workflows, et cetera, do you have examples to make it slightly more tangible for us? I hear you, but I don't hear you, if you know what I mean.
Rajesh Jha: Yes, yes. So let me give you just a couple of examples, maybe a couple of customer examples and maybe one from my own team. So one of the customers, what they did was, so when people started using the Copilot, the first thing they tried to figure out, "Hey, how should we go do this?" They asked us, "Should we distribute a few seats of this across different departments, or should we go all in, in one department?" And based on our work with them, we tell them, "Go in into one function," because we find taking a peer group and giving them all Copilot helps reimagine workflows. So this customer, what they did was they gave it to their service department, and they're an operational company. And so they had very complicated processes for long-running incidents because hundreds of--these incidents would be day-long. This was the core of what the company does. And then they had very complicated overhead of how, when people work a 4-hour shift on the incident and the next group came in, who was going to transfer the knowledge? How were they going to come up to speed and so on?
So they reimagined the entire workflow in the context of Copilot in the meeting. So when a new person would come in, the Copilot would just summarize. They could just be in, "Hey, can you summarize what happened in the last 6 months? Which ones apply to this specific function?" So they were able to get rid of a lot of the manual overhead, the human overhead of transferring knowledge between shifts and so on. That was an example of just reimagining the workflow.
We have another drug research company where there's a corpus of a lot of data around clinical trials and of research, but that's in a different system. But what they did was they extended the Copilot, so when the Copilot people could summarize quickly when they came in what were the latest findings from the last week from their colleagues in the context of all the emails that have come in around that, all the chats and documents plus what's in the system of record.
In my own team, we saw this super interesting thing. If a product manager wanted to go incubate an idea, the process used to be he or she would come up with an idea in the product management team. She'd go lobby for some resources in the engineering team and design team so they could go incubate. But now she just uses the Copilot. So this product manager, early in her career, she had this incubation idea.
She just used the Copilot to generate. She doesn't know coding. She's not a designer. She had an idea. She used the Copilot to create a perfectly workable mobile app to incubate a bunch of stuff, got to try a lot of different things. And then once the idea was gestated, she brought it up to the core teams to actually then go and implement it. And so this has saved that specific scenario months of just time of coordinating people. So I think what you will see is workflows are going to have to be reimagined in addition to augmenting the human ingenuity and creativity and taking the grind out.
Raimo Lenschow: Yes. And maybe I should have started, actually. If you think about the Copilot and the productivity gains and you started on get help on the Copilot side, and now we had the first guys using it on the Office side. What do you see in terms of productivity gains and changes to work behavior coming out of that one?
Rajesh Jha: It's the first thing, the GitHub Copilot in many ways has blazed a trail for us in the design language. So we spent a lot of time about how should the AI get initiated for teams and users and organizations? It's no secret that the AI can get some things wrong. Now, of course when you ground it in your information to a rack pattern or what have you, then if you're using it as a reasoning engine, not a fact engine. But what the GitHub Copilot taught us was the design language of it being a Copilot, not an autopilot, and working on your behalf of making sure the humans were in the loop to actually accept the changes or to commit the actions.
And then when we saw the GitHub Copilot land up to 50% to 55% efficiency for developers, the other thing that we did was we created the Copilot Stack. What I mean by the Copilot Stack is, so you take a look at Microsoft 365 Copilot, it's not just the largest model, the most capable model. It's the fact that that model is grounded with the Microsoft Graph, with all your permissions, and it's then brought into a Copilot-designed language and user experience that you use. And then there needs to be extensible. So this entire thing is what we call the Copilot Stack. So all the Copilots in Microsoft have been built with the same orchestration, same extensibility, same design language, and we've made this thing available as an Azure service, whether it be Azure AI Studio or for low-code/no-code with the Copilot Studio. So I think in many ways the GitHub Copilot has blazed a trail for us into this one design language.
Raimo Lenschow: And then what are you seeing in terms of the customers that have been working with the Copilots and adoption curves there? Telling you where I'm coming from is the productivity gains we're hearing is somewhere like 20%, 40%, like 50%. And the people that are working this are more developers, white-collar workers. They're expensive. So in theory, you think, "Hang on. Everyone should be rushing out getting this." What are things that on their adoption that we should consider there and that you experienced so far?
Rajesh Jha: Yes. So great question. So it's clear that for the end user, whether it's in the context of creation, summarization, first draft, all of this, the payoff. This is what I was just talking about and what are you going to see in the Work Trend Index? Benefits of 50%, 60%, 70%. In terms of meetings, summarization, follow-up tasks, about 80% to 85%. So it's clear value for the end user.
For now, let's talk the IT admin in enterprises. They're also responsible for security, regulations, compliance, and so on. So one of the things that we've done with the Microsoft 365 Copilot is actually built it to be enterprise grade. What I mean by that, it understands things like conditional access. If IT had a policy in place which is, "If Raimo's outside of this Zip Code and tries to access corporate data, do not allow that access without another challenge to the authentication," the Copilot understands those policies.
So things like all interactions with the Copilot are discoverable for regulatory reasons. It generates an audit lock. So we've done that work, but IT still has to go through the validation process for these things. So one of the pieces of feedback we got from IT early was, "We love the notion of Copilot helping in meetings, but we don't want to create a data retention policy where transcripts have to be retained to enable the Copilot." So we did the work so you could now get the Copilot to perform in a meeting without creating a transcript that lasts beyond the meeting. So there is a bunch of IT evaluation and adoption. But we have an adoption guide at Microsoft, adoptionmicrosoft.com for IT for those reasons.
The other one that we hear from customers is, "How do I know the ROI?" So we built a Copilot dashboard so they can actually take a look at the impact of the Copilot. They can join that with their customer data, so you can bring your HR's custom data, you can bring your sales performance custom data and join that with the Copilot data and see what the ROI is, so we've gone and built that, too.
And then for the end users, we are having to really teach them a new way to do computing. All of us have grown up with keyboard, mouse, swipe gestures, and so on. We're now using prompts. What is a good prompt as a salesperson? What's a good prompt in this organization in a manufacturing context? So we have that now in the Copilot labs.
In the end of the day, the Copilot is going to go through a bit of the adoption cycle of any enterprise because you've got to work the risks, you've got to work the compliance and procurement and make sure there is ROI. But the time to value for the Copilot, I feel unlike anything that I've done in the M365 space in the last 20 years, is much, much, much faster. The Microsoft Graph already exists. It's built to be enterprise grade, and it shows up in the flow of work.
Raimo Lenschow: And you mentioned the Microsoft Graph a lot, and actually I think on our side we paid not enough attention to that. Was that before and you just got lucky? Or how did it play out?
Rajesh Jha: It's super. Both, I would say. So here's what happened. If you go back to the Office business, the Office business when we were on premises was a bunch of silos, so we didn't understand the user well. We understood the jobs. So with Exchange and Outlook, we understood the mail and calendaring job. With SharePoint and Office, we understood what document management was. With Link at that time, we understood meetings and telephony. But they were all stuff pipes. When we moved to the cloud, we really got to be user centered. That's what the Microsoft Graph is. We then--and the reason we were pipes on premises was no customer could deploy all our products at the same time, so every silo had to be standalone and be able to exist, even if the other silos were different versions.
So in the cloud, now we had one data structure. We were doing the job of keeping it up to date. So that's what allowed the Graph to get created. Then what COVID did was it accelerated the use of Microsoft 365 with lots of intensity. The more you use Microsoft 365, the better your Graph is. Now after COVID, what we have now is tons of customers deeply immersed in Microsoft 365. The Graph has held the permission, and now you get generative AI, you take the Graph, you ground the generative AI into the Graph context, and you bring it to the user experiences that people are using today. And that's what the Microsoft Copilot is. And that's the differentiation of the Microsoft Copilot.
Raimo Lenschow: Yes. And then I wanted to shift gears a little bit, like what does it mean on the way how you do compute, like first fundamentally, but then also how do you do it in Microsoft at the moment, because you have so much to demand on your side, on the Copilot, et cetera. But then you have all the external demand on Azure. So who's getting the GPUs?
Rajesh Jha: Yes, first of all, this GPT-4 was built on Azure, and we've been optimizing all layers of the stack for over a year a now. That being said, to answer your question, there are a couple of ways. Of course a customer, when you come to Azure, you can use not just the frontier models from OpenAI, but you can use models from Huggingface, from Meta. We now have model-as-a-service that we have announced, where you can use Mistral's premium models, Coherus model, Jais model. So you can use lots of models.
But for the large model that we use in our first-party applications in Azure OpenAI API, we've done the work to leverage one implementation and one stack. So it's not that--so I can optimize the usage of the GPUs across Bing and Windows and Office and Dynamics and the Azure OpenAI because we all use the same APIs. And if the third party's using our APIs--you can come and bring your own model. You can use the Plethora model. But if you use the large model, we are getting the leverage of the fact that we are all using the same API, same endpoint. And then, of course, underneath the cover, we work clearly with close partners, with NVIDIA and of course AMD, and we have our own AI-first silicon that we are bringing to the picture.
So you should think about below the water line on the systems level work we've been optimizing for a long time, for a year, we have silicon-level work, and then above the water line, when you are hitting the large model, we get to optimize the traffic across all our first parties and anybody using the Azure OpenAI API.
Raimo Lenschow: Okay, okay, makes sense. And then I have Investor Relations getting a little bit nervous now, but I try to ask like not a numbers question, but like more a monetary question. If you think about you're creating with AI a lot of value for the client. But if you look at your pricing, it's still like, if you look at the $30 per user, it's still like a very classic model of, "Here is the price." I think it was for the GitHub Copilot, it was like 9 bucks. I tell you I talk to customers and they are like, "That's a steal, because I have 20% productivity gains and I'm paying 9 bucks for a guy that cost me $150 and I'm getting 20% more out of it. Give it to me." How do you think about that dynamic going forward? Do you have to rethink or reimagine that?
Rajesh Jha: You say early days yet. The thing that we believe is--take a look at Office. Before Office existed, Word was a business application, Excel was a business application. And what we did was we said, "We want to democratize this for all the users." And so Microsoft Copilot, GitHub Copilot, is a democratization of AI for all the users. So we basically said, "Hey, we want to take the core capabilities of what used to be productivity, and now AI enhanced productivity and want to take that to all the users, just like we've done with Office." In addition to the typical user, we have lots of other constructs that we have on top, whether it be E5 construct, whether it be Teams premium construct. So right now, really the way we are thinking about AI is, "Let's bring it to every information worker, every first-line worker, get to enhance the productivity," and then I'm sure there will be lots of opportunities for us to add value and capture value.
Raimo Lenschow: Yes. Since you're working on the experience side for Microsoft, how will I need to reimagine my UI going forward? Is there a classic way a good generative AI system should be more interactive? What's the work that you're doing there?
Rajesh Jha: Super good question. For 35 years, computing hasn't changed in terms of the way, you're on a phone, you swipe, on a PC you use a keyboard and mouse. And the operating system does a very basic job of abstracting away the hardware and allowing you to get to an application. That's going to change.
And so the way you'll see us integrate the Copilot, for example, in Windows, and we think that the lines between the shell, search, browser, Copilot is going to blur because you enable something that understands your intent at a higher level with natural language, it can take a lot of the grunge out for you. The application relationship to the operating system to the Copilot is going to evolve rapidly, and you will start to see us do this stuff by bringing the Copilot into Windows. It's going to be very exciting. But I do think it's a new paradigm for user experiences.
Raimo Lenschow: Yes. And that gets me to the other questions around devices. You guys were always very good of showing your partners what's possible. And actually, I remember when your Surface came out, I bought it, I loved it, and it was like, "Oh, this is cool," and in a way, you almost forced innovation on your partner side. In a way, like you need to almost reimagining devices, et cetera. Not that you'll announce something with me, but where are you in that thinking?
Rajesh Jha: I do think this naturally is this natural user interface, and these are multi-modal, by the way. This isn't just natural language, but speech and video and all of that stuff. I do think our existing hardware form factors are going to evolve to accommodate this, and there will be new form factors.
Raimo Lenschow: Yes, let's leave it like that.
Rajesh Jha: Let's leave it like that.
Raimo Lenschow: I tried. [Laughs] Yes, okay. And the other thing, shifting gear a little bit back to client adoption that we see on the Copilot side, what's your thinking in terms of the speed of adoption you can support with the given shortage of GPUs? How do you see that playing out? Do you give some to every client, a limited amount to every client? Is it like one client gets a lot so you learn a lot more? How do you think that will play out in the field?
Rajesh Jha: We are open for business now with customers on Copilot. It's really like the earlier conversation I was having where the customers are going through their evaluation of the security policies, governance, procurement and so on. I feel confident in our ability to optimize across all our first-party usage and third-party usage. We've been at this for a while. We know how to go optimize capacity.
Raimo Lenschow: Okay, yes. So it wouldn't be, like so if Barclays is, like I think we're a big Team shop for you guys. If we say we want Copilots for everyone, it's not going to be, "No." You could in theory, yes?
Rajesh Jha: I'm happy to sign you up right after this.
Raimo Lenschow: Yes, yes, it's on me, yes. [Laughter]
Rajesh Jha: I'll have to talk to you and tell you how we're enterprise grade. We are.
Raimo Lenschow: Yes, yes. And the last question for me is how do I think about you guys working with the partners, because one is like a very Microsoft-centric view. But on the other hand, it's not just you; there's a whole broader world out there. How do you see that evolving for you guys in terms of working with other people, bringing other people into the equation for going forward?
Rajesh Jha: A super good question. And again, back from the days that Bill used to run the company, I remember being in meetings with him. He would take incredible pride that for every dollar that Microsoft would create, we would create $3.00 for the ecosystem. So we are really a platform company and not just the Azure layer, but even at the M365 layer. And so already with Teams, we have more than 2,000 ISVs and countless other line of business applications that extend Teams. And you will see the same platform sensibility I talked about it and Build and Ignite, and in this coming year Build, we'll talk about how developers, both line of business developers inside of organizations as well as ISVs can augment the Copilot with their custom logic and their interactivity. Because not all--productivity is so human in all it ways and forms. We're not going to build all the most interesting applications.
And so we want the Copilot, Microsoft 365 Copilot is a platform, and that is we are working. I don't know if you've had the chance to take a look, Raimo, at Ignite from 3 weeks ago. We showed lots and lots of customers and partners actually extending the Microsoft Copilot.
Raimo Lenschow: Yes, the conference was really exciting, and it's really nice to see all the innovation coming out of there. So Rajesh, it's 58 seconds and I'm German, so I have to be on time.
Rajesh Jha: [Laughs] Nice job.
Raimo Lenschow: So thanks for joining.
Rajesh Jha: Thank you very much.
Raimo Lenschow: Thank you. I really enjoyed our conversation.
Rajesh Jha: Thank you very much. Thank you, everyone.