AI Solutions for Enterprise SaaS

Gen AI Summit 2023 - Open or Closed

Zoom: Smita Hashim, Chief Product Officer

Zoom: Mahesh Ram, Head of Digital Customer Experience

SUMMARY KEYWORDS

ai, customers, models, product, application, talk, users, zoom, collaboration, capabilities, data, people, understand, meeting, features, fine tuning, thinking, fast, meetings, optimization

SPEAKERS

Mahesh Ram, Smita Hashim

Smita Hashim  00:00

Well, hello, everyone. It's great to be here, one of one of the other sessions. So most recently I'm coming to today's that's what I saw as well now I'm coming from Microsoft Teams. I was reading various applications and all the video related products. Before that I was in Google for a long time actually I was meeting up and electorate so it was all Psalter scaling over there. I worked on modify the productivity that productivity for the boys calendar is I

00:36

had Chromebooks

00:41

if I would say so. What

00:44

was the type of big big big believer in Visual Communications and the father of communications? I think there are many unsolved problems here. There were many options on when it's going to be good for everyone. And zoom is really a mission based company. This moves incredibly fast. It has by far the best video quality and it's investing a lot in internal collaboration

Smita Hashim  01:10

as well as external collaboration. So so so so far, we are very excited to have you with us. So AI is everywhere.

Mahesh Ram  01:28

But how are we thinking a bunch of people position? It is our environment with all these advances, and zoom in doing with

01:39

absolute reality as

01:43

they are in the keynote. I really enjoyed it. I think it's incredibly it's a really exciting time. I've made a joke for a long time, but from from my vantage point, I just took an idea of AI and the possibilities if you don't like those are the most exciting times.

Smita Hashim  02:02

But from a job point of view, is actually is very interesting

02:07

on its surface area. If you think about it, spent a lot of time on that matters because we use that as our that's where they will use the tools. So in a way it's a great area to really get the users even more productive and have them plugged in better. And so we are as you know, as a product leader at this point, all of our product teams and we have, I believe, two meetings, but we have two team chat. We have four we have a whiteboard, we have our customer facing product to conduct centers in any particular order on the product team to activate them on building even more AI features. I mean everyone was putting some effort into his into the lot in the air quality, especially with reference regenitive Ai. We are moving very fast on these applications and bringing features rapidly to market couple of features should be in GA within the next couple of weeks. Maybe somebody's coming into my new chapter, boss is coming in chat on vacation is coming. So we are moving super fast on this because we learn by doing but apart from the front end, I think it's just equally important and we have heard a lot about that here today in the shifting landscape or the real elaborating on the message landscape of AI models. And you don't understand them most of them and figuring out like how do we evolve that? So we are taking what we call a federated AI that we are being very intentional about being able to use multiple models, right? So this combination of what we're doing on the application there to really our users and do that fast and then on the back end defensive approach. And you cannot really be in

03:47

the space without being responsible that customers can begin to that's preventing you from opening people up on extremities.

03:55

So I think my own observation that the founder of a unit is doing and the lesson to be learned from Eric is I've never met somebody who run the company is big listen to this much the customer Yes, unbelievable accurate, but you your personal days you've done a lot of that the refresh yourself and document. What are you hearing from our customers about what they want us to pay providers such as zoom in the area of AI and generative AI? And can you talk a little bit about how this led to the creation of our principles that we come up with? populated with?

04:29

Yeah, absolutely.

04:30

So as far as I think, I mean, I think zoom has a massive customer base has a really, really small and I talk a lot to customers. I talked a lot to some of our largest customers as you can imagine a lot better and frequently. And every customer wants to know what to do with AI. So you know, so first of all, the data sentiment to learn and more is pervasive. Everyone wants to know what they are trying to figure out what the AI strategy is, but as a solution to if they want to know what our strategy is regenerative AI all been different than they all want to know how they need to be handled, or you know, they want us to be thoughtful stewards of their data. So those are the two things which really strong the mind for the customers and then they're eager you know, when you talk to them about it. There's a little bit of I'm starting to wonder motivation issues and there's a lot of thought there to manage that we can actually play with. So they're also eager to get their hands on projects and start really using them. So I think expectation is good strategy, helping them understand data, you know, coming to understand navigate the landscape of models, and what does it mean for

05:44

getting things into Christ?

05:48

A great answer I would add to that one of the things I've noticed even in the last 100 days, is the level of sophistication of questions that customers are asking about AI has an increased almost as fast as we do. Which is increasing the complexity. And I think many founders that they're thinking about how you approach companies and just be ready. That may be within context. This is something I would give.

06:14

Obviously, she was an incredibly engaging area. I think so yeah. So I think all of us in this room are actually very lucky.

06:24

So diving further into the federated approach to AI that Eric mentioned this morning. How does how to balance the use of third party foundational models such as open AI and brought back with the models that we do and create and improve. How will you determine that weapon? How can this be determined?

06:45

So it's I mean, like you're talking about

06:51

the landscape is evolving. We have to look at multiple models

06:55

and you have to figure it out, unfortunately, or you know, you have to learn from that and figure out how to optimize from his own perspective. We have announced our partnerships with open AI the anthropic is really excited about those partnerships. We have a tunes own

07:10

proprietary models. As well and talk about that

07:13

too much especially our customers

Smita Hashim  07:14

want to use their own models that that they desire, we fully support it.

07:18

So it is a that is a federated approach. But how will we learn, optimize, apply, and then share some of the things that we are doing in context. of the application and maybe that'll help so we are about to launch our zoom meeting somebody feature and then also you can look at how many calories you know hopefully it makes people a lot more efficient. And we will launch it at scale to our to our customers. We are using soups or bottle for that, or bottles, understands the conversation data, it understands multi person conversations. We actually had an auto pharmaceutical group and is one of our products for sales, which is new market for sales and we are buying into meetings and predefined views on it.

08:04

A platform post which is

08:05

a more general purpose, conversational interface, if you will, this is a zoom IQ zoom zoom game chat. We are

08:10

using open app

08:12

for that right now. We just acquired announced a partnership and it's probably we're getting excited about that. And I thought atomic Wondershare Video anymore to apply to a contact center product and a small virtual agents have products very rule based, you know really excited about the Constitution AI approach. We want the answers to be

08:31

maybe like more definitive. So, you know, saw the answer virtually and the CEO saw me you gave your virtual agent has to be it's an energetic agent. So the quality has to be very high. So of course we have specific models

08:46

for that. So I share these examples to show that within the loop back over the use cases, how we apply these different models and we continue to evolve and learn from here, I think

08:59

yeah, I would just add, I think we already said it earlier this morning about you know, the real value is if you understand the context in which someone is using any technology, AI generative AI you can bring that context to the AI and I think that's how you differentiate it from from whatever, just generally public.

09:18

Absolutely. I didn't know. The application still matters, but

09:22

the application on decks still matters. The users are in the application. So as soon as you start doing this application work, you will realize how you want to tune your models. What are some of the challenges what's the sequence of data, you want to give it to how much context you provide. So if you look at for example, our meeting salary, which was trained on two IQ or sales needed to be fine tuned, again on general meetings, only look at these types of micro chat combos having done in terms of which user is to walk and what kind of tools we use that only emerges once you start applying it to the application today that application context and applying it to the model to optimize I think that is where the user ID is.

10:07

From a point of view of a business what you see in the unique value. So from a business perspective, yes for the user to describe the business.

10:22

perspective and again, as we talk to me this comes from our customers as well. Every business is looking for far more productivity, better outcomes, but they want to do this while not compromising on their

10:37

customer experience and user experience.

10:40

So that is the that is the service which you have to work with. So from an end user perspective, or from the Europe immediately they are hoping that their employees will become more productive with these tools. And then as you become more productive with the tools, you have more time for creativity, you have more time, more time for collaboration for relationship and negotiating

11:02

outcomes.

11:03

When you think about the customer service facing kind of applications in that case, also they want to optimize they want more and was attracted to to you know to AI but they want to also be very sure that the customer experience is not compromised. So I think that is where they are looking to see and of course they're curious about the cost of this product school. And how much would it cost them I would like dive in there's a lot of pre trials going on, but eventually all of us will have to pay for the finalization

11:35

Yeah, one thing I would just add that came from customer conversation that I had last week was a light bulb went on for the customer when they started to understand that we built for some of the core capabilities are applicable across business use cases. I'll give you an example of summarization on the external environment today, right you're gonna get some as a meeting is a very valuable thing to extract next steps are that next actions understand context and the meeting very valuable. But then you still attend, you're able to apply context and say, here's the conversation between the board and the humans. Let's summarize that for the agent and synthesize that with the agent that talked about a new one problem and synthesize that and summarize, it's a variant of summarization and fine tuning, but it's not right. It's sort of the light bulb came up and said, Oh, you're investing heavily in this so then you can bring it across maybe across science. I thought that was that was interesting. You have to get very specific on the use case,

12:28

but you may be able to participate in across the platform.

12:33

It's good to hear from them from from Africa. What are you most looking forward to in AI? As I said to somebody that probably last night, but here's the most exciting and most misunderstood. Technology is what happens in general. And I think most people would agree. But it's incredibly exciting, but what are you most looking forward to? And what are your own thoughts about how

13:00

the applications will evolve? So what I'm looking for is to go from the hype to reality and for the reality to get better and better over time. I think that's the journey which we all are trying to traverse over here. So we have, we have a very, I mean, we have a rich roadmap. We have 3040 features already on our roadmap, and I'm looking forward to, to Avid I love it but the features are being built as well and you can look at the quality and you can fine tune that and we'll continue to experience you can show them your customer. So I'm looking to these. These products continue to be bailed out, putting them in the hands of the users getting the feedback, learning from it, and then continuing to move as you were saying from single product flows to orchestration across multiple products getting into more and more assistant capabilities. Right now. I think the way we are playing with a lot of these use cases is users in the room they get some information and that you know the app on it. But can we move from also potentially being this more reactive or user prompted to walk the ladder? So I think there's a spectrum of capabilities which will go on and the power which we put in the hands of the users is going to

14:14

is going to continue to increase.

14:16

So I'm very excited about that. I'm also equally excited to see how the lair of wardens evolves and how the costs and the cost curves and the optimization of a product person. I look for a business I also want to figure out how to do the monetization and then how to do the cost optimization. And I will continue to to see that beyond investing a lot. We also have to get them

14:41

on these investments.

14:43

So one of the other things of course is tremendous excitement, because any customer call away this isn't the topic of one of the coffee cup of coffee with topics every every customer call them but there are also concerns right there is concern about data and privacy and security up a little bit about how what's our responsibility, you know, as an application providers that are leveraging these technologies and our own technologies, what are our responsibility, and how will we best do our jobs and be stewards for the customers

15:16

So transparency is, is existential, I think we all have to be very transparent and very careful about the customer's data. So any product that we build, we have to be I mean, it's always you know, they decide what happens to their data. So that part is inbuilt. And you know and we have to do the exercise that we have to give customers the choice and the flexibility. In this case, we're actually four. If you look at even that's why some of the customers deserve that transparency, that flexibility. If you're not customer you want to use your own models if you want to pay them in ways that is specific to your job and your vocabulary and your brand attributes. You have to find a way to let them do their fine tuning while still keeping control of their data. And you're getting to the point where they either to assign it as a time of like you're not getting the results from the module or you can be fine tuning in different ways in order to on the module, but even those

16:13

kinds of capabilities. The other thing which

16:16

we have to be careful about and this is something which I think a lot of us know over here, but because we are in the field. I think one of the dangers of these large models as we know is that they appear highly credible, the conversation is very cogent. It feels like a really complex one, but frankly, there are errors. So how do you responsibly help your customers who are trying to have their employees use the other day? I mean, are they employees, not people who understand that large language was used today. So we also have to provide the education aspect of it so that when they when their employees are using them, they can use them responsibly in ways that we actually I think we also have to highlight the limitations of these products. We should not be painting a rosy picture, which needs to give

17:09

an unexpected outcomes for our customers.

17:12

Yeah, I think even even our own experience of bringing GPP in your employee base before we moved in in our product release because we started to understand something I think a lot of week after we did it, I think he said the band it wasn't available in China. So you really started to think about it. What does that mean for inclusion in the company? Some people have some features and some don't. But how do you instrument the products as he would turn it off? Or people in that or what happened with that enjoy your travels from the brands? Are they allowed to use it or not? You know, these are very complex and different challenging things but if you don't think about them on behalf of your customers, then you know they are not the obligation. I think it's one of the learnings for us. When you do it ourselves. We were on

18:03

the project yesterday come on champion are you to

18:11

talk to maybe a lot of young company of your founders startup may want to partner with Zoom some of whom may be in a different line of work. What would you say to them about the ways to approach this exciting?

18:30

Yeah, I think I think

18:31

everyone is here because everyone is saying it is right so I think the first thing is I think it's important to I mean, it's important to be a practitioner. If you are an application provider. I hope you're building something. You're actually using AI when I help you accelerate your investment. you're activating it. You're getting to market that maybe you are looking at the quality carefully if you're thinking about how to communicate. So you are moving fast on it as an application provider. I hope you also everyone can realize it and I'm sure you do that. You know that you are standing on quicksand things are changing wave application. Or vendor picking up on your application time you're engaged to how fast in recognizing and you're keeping that flexibility in mind from from day one. And then you're also thinking about your costs. You're thinking about monitoring models, you're thinking about optimization, so I think you should be thinking about all those key dimensions, which is the user experience of having flexibility. As the backend flexibility you know, and user experience improved your customer data experience as well. And then you're thinking about your

19:40

costume organization and optimization. If you are a service

19:44

if you are a infrastructure provider or if you are actually I've been on the Bible before us talked about it quite a lot. I think it's such an incredible time for you in order to develop tools for our foundation models are ever sent. Do you continue to evolve them in order to pay for the tools and the services or the early solar arrays? And you know, and then I didn't watch the news last over here to to again, fill those kinds of products and hopefully we can only find on some of you when you find them with other applications and other projects.

20:22

So I have another 100 days what's in your first 100 days with deep immersion, understanding the culture and people will be looking forward to the next 100 days.

20:36

I mean, it's been such a busy 100 years it's been awesome like I really feel like you're not just where we are in terms of our roadmap and

20:45

or academia.

20:46

We have any teachers who aren't here. I think it's quite it's actually it's exciting. I think 100 days from now, he would be would be a lot further I think he would have learned so much more than that. Your customers will be using a lot more of these features. I've been able to a lot of us are putting on five features right now. I think they would have actual products which we can stand behind by that time and you know, and then from there, we would be ready to get to the next step. So from my side, I would love to see these kind of assistive capabilities. I think we will be in a very good shape for the first generation of it and for the next generation across both across collaboration as well as external collaboration. I think both surfaces are incredibly important, but think of internal collaboration as knowledge workers, and all of us over here, meetings such as for the eyeballs, many people have external. I think that sometimes everything was more excited when they started a rabid purely, you know, we already have such a model based

21:53

on Monday for responding to my weekend Fridays. But now as we begin the AI capabilities, and we can really create experiences, which I think would be absolutely transformational.

22:08

And one of the things that you talk about pretty exciting is this idea of going even one step beyond which is, which is all of us have nothing to workers. And you know, we'd love to have some time that you talked to us earlier about some of your ideas or, you know, freeing up time and thinking a little more of that but broadly coupled with the work what do you see the future?

22:37

Yeah, I

22:39

think we don't I don't know it's a big question. Because we are talking about applications of the Savior right and what is what is your application experience? And then how can AI come and actually take that application experience and then they need to bring user outcomes. So as an example, I think, again, that you know, we are the collaboration company and with that failure DNA when we think about collaboration you know, why do we need to attend so many meetings, we should be able to, to see what happened in that meeting very, very efficiently and be able to follow up on it. I mean, we are all we have Smart Recording, which already shows your chapters but even that's quite onerous. You have to go and look at all the chapters you have to read through it. So hopefully it will save a lot of times when this fall will go away because things will be fast and they'll be ready for you to digest. And I think that in itself is an added workforce. I think it changes the game for a lot of us but we have to do a very good job

23:40

with it. If you don't do your job, then it's not going to second. Absolutely. I'm excited about it. And if you're

23:48

really excited about that I see a lot of nodding heads. So I got a reputation for hiring great, great dialogue and thank you and