We have a great panel up here today. We have Aaliyah's design. She's head of digital at Hyatt. We have Kip Bonner, CMO at HubSpot. Alex Bob Danko, he's head of data science at MassMutual. And Dr. Bloom off.
He is CTO at Akamai. We have a great panel here today. We're going to dive into innovation, investing, innovation. We're going to specifically look at A.I.. So with that, let's kick it off and let's talk about A.I.
in the hospitality and leisure area. Arley What are the air use cases that customers are seeing today, and what should we expect around the corner? Thank you. Yeah, exciting to be here. We do have a lot live today. We've been actively experimenting in the U.S. for some time.
I think the generative bug got all of us about a year and a half ago, and since then we've been able to release some really exciting consumer facing items. I think when you first land on our sites, which is our front door to how we welcome guests to our hotels and properties globally, you will see Hyatt Hotels.com absolutely has some search activity happening. So if you feel so obliged to go out, my team would enjoy a little spike in testing for sure. So whether or not you are now opening up the context of how you can find a hotel or property globally, whether it's weather based and not just location based, whether it's interest based and not just location based, or maybe you really, really have a strong preference for a pool or a gym. We're absolutely getting smarter in that
arena and something that you can experience live today. Another fun part of what we're doing is really making sure that we have 1 to 1 communications. You know, the AI machine learning environment absolutely allows us to excel in creating 1 to 1 and not one too many. And I came up in the publishing industry at Condé Nast for a minute, and that was definitely one too many in terms of how we were able to produce content. Now we're able to customize offers, content that you see on our pages and other interesting tidbits as we go through the booking experience to make sure that we're really elevating everything that we do. And those are just two with many, many
more to come on the horizon. And your point about customizing the personalization experience across your customer spectrum, how do you differentiate versus other peers? Is it or are the industry players kind of understanding the value proposition of AI in hospitality, or is there a uniqueness to Hyatt's approach? I think when we start with our company's purpose, which is we care for people to be at their best, we absolutely take that into the center of how we put together our overall outcomes for what we want to achieve with every project. So I do think depending on your company strategy and how you take that in consideration will absolutely lead you to what your specific touchpoints are and also where you start. Everyone has incredibly similar, I would say, data in terms of what we're able to collect in our industry. But however, how we choose to specifically take that in the projects that we decide to take on and fund are all rooted in our antithesis of care and how we have that come out. And that goes from everything to our associates all the way up through our guests and customers that are staying with us. It's interesting points.
And you said a magic word data and good data equals good air equals good enrichment or value. Head of Data Science. How much data can you possibly manage for your AI platforms and how do you distinguish? You have a good set of data or less so good type of data set? Okay. It's a great question, and I think that the answer to that question can be measured in a similar way to many other important questions where it is important for you to give the right answers in the right context to the question we have on Q8. How do you know if you have good data, bad data, and the primary lens we look at this is innovate to unlock business opportunities. And so the isolated metric on the
quality of data is is interesting but becomes important in the context of specific opportunities. So in particular, we're really excited at MassMutual to be the first U.S. insurer to offer our policyholders free access to Weiser Assure Weiser Assure is a chat based AI powered mental health and wellbeing app so eligible policyholders are now able to access mental wellbeing. Chat 24 seven when they when they need it most and. Yeah. I'll stop there for now. Keeps coming.
And you raise a good point, Alex, in terms of the data we see specifically as a consumer of life insurance policies. How are customers realizing this vast amount of data that you're harnessing in terms of features for an insurance policy, let's say yes. Or one way that a policyholder would see it. If they call up their financial advisor today, they're advisors and able to log into our advisor, virtual assistant Ava and have at their fingertips the knowledge of the most knowledgeable person at MassMutual that enables them to reduce the friction in answering that same routine information on servicing their policy so that advisor can engage in more authentic, holistic financial guidance that our policyholders expect and demand in 2024. Interesting. And tipping into data and cybersecurity. Dr.
Bloom off, how are you managing the sheer volume of cybersecurity and the type of pasteurization required by machine learning algorithms? It's a great question. Yeah. You know, we've been using machine learning or A.I. in general in our security products really since the dawn of deep learning. So you probably know, you know, deep learning dates to about a decade ago. That was kind of the invention of deep learning.
Arguably, before that, A.I. was not all that relevant to most enterprises. You just couldn't do all that much with these small neural nets. Things changed about a decade ago, and it turns out, you know, deep learning really is the ideal tool in cybersecurity because it's perfect for classifying large amounts of traffic in terms of what's normal versus what's abnormal, what's benign versus what's malicious, what's human versus what's a bot.
You know, more recently you get into generative AI and it has use cases, for example, well, really across all of our businesses, but use cases in simplifying the customer interaction, making employees more productive and things like that. But maybe I'm not exactly answering the question, but I did want to use this as a launch point just to make a will make a point about A.I. models in general, because I've been sort of envisioning things in a kind of an iceberg sense.
And what I mean by that is that above the surface, there's these there's models like, Hey, you've got Chad, GPT, you've got Gemini, you've got llama, you've got Anthropic. You know, these are the models that you've all heard of. They garner all the headlines and you know, anybody would be forgiven for thinking that's all there is to A.I., but there's a whole lot more below the surface. There's a myriad of much, much, much, much, much smaller, specialized, deep learning models that are doing everything from fraud detection to product recommendation to anomaly detection to time series modeling and things like that, and arguably are delivering probably more enterprise value today than those shiny models that are living above the surface. So when it comes to getting value out of
AI, that which your question is, you really have to pick the right tool and sometimes it's not even A.I. at all. AI is not the answer to every problem. And more specifically, Lem's generative AI is is far from being the right solution to every problem. And because of the hype, I think we're we live in a bit of danger.
Interesting points and you triggered some, you know, ideas in my head in terms of large language models. Keep your company is balancing using third party jenn-air end models versus internal. How are you going about that decision process and how are you leveraging some of the popular models out there? Yes, if you think about the serum space, which is the space, we're in it HubSpot. You're talking about customer facing use cases. And I, I think the point of you have to pick A.I. that's right for the job. But it is best that right now is really large scale guessing.
It's really good at large scale guessing, which comes down to from a use case perspective to Oh, maybe I can at scale recommend the right customer service article or answer. If you are a salesperson, you're trying to prospect out and get new prospective customers and you don't know that much about them. And I can infer a lot and make that outreach much more scalable and much more personalized. And in the marketing side of things, I can help you research, draft and get a lot of the messaging and content together for your different campaigns. It can also automate a lot of like the operational infrastructure for the campaign. So what we're trying to do is work with the the best LMS to basically provide those customer points of engagement.
And then obviously we are going to have to do fine tuning and kind of deeper learning on those models to deliver for those specific use cases in the product. And you service the small medium business community, specifically in the functions of sales and marketing and customer service. What are their top three use cases for A.I. today that you're hearing? Yeah, I mean, most small businesses right now, like they just can't get their head around, Hey, I like a guy is not in their kind of mode. So what we have found is that what, what
actually is valuable to them is a feature not needing to know that it's say, ISO, it's productivity features. So we have a feature our most one of our most popular features is content remix so that you can have a piece of content and it will remix it into multiple formats of content, do all of the workflow and automation around that, and that saves a small business a massive amount of time and they love that. The other thing that small businesses really like is copilot, because they're already used to chat and so they really like, you know, just asking their data, Oh, well why is my web traffic up this month instead of having to figure out what report to look at and what they need to know? They're not experts in this. They can just ask in natural language and get those insights. Those are a couple of things that people are really adopting early on.
Interesting. And, you know, as we sit here and we look across the tech landscape, we see a lot of comedy, a dichotomy in terms of investment scale. Can you speak early in terms of the scale investment required to bring these features to the customer? Yeah, absolutely. And I think it's probably different for every company, right. And where they are. I think we just heard some examples here on where you have it stored, how centralized you are already, how good the data is and how clean it is, and then what you're actually trying to accomplish with it.
We've had a very exciting, I think, progression at Hyatt, where we've been very fortunate to have our CEO. We operate in Agile pods and he joined and led the pod. So a little bit of a surprise. Welcome. Our couple made the into the room and I think from that perspective though, that helped unlock obviously urgency, attention and also garnered and helpful, you know, for the person at the top leading those investments, how that looks, we then I think took a pretty structured look at what is return versus investment on a lot of these things. And because we started with consumer facing first items while also looking at productivity but specifically in the consumer facing ideas, being able to wrap a business case around on those is a little bit easier, right? Because they're usually tied to revenue generating opportunities. And so for us to be able to put that out, have the entire executive committee really be engaged in those ideas as well and help us determine in the next fund those ideas has been a really productive thing for us internally. And I think also just obviously
garnering, you know, executive team attention to this item and having it be a really excited top down push has been nothing but an unlock for us. And when we think about investment in life insurance and the scale of policy and customer base that you have, Alex, what did you have to do your tech stack to enable some of the features? And you mentioned the mental health app that you have just pushed out over the past year. Yeah, I think well, first I want to comment on one theme that I think we've we've noticed across all our panelists, which is that generative AI is like this golden hammer that many stakeholders are coming in and saying, Hey, I would love to use this golden hammer, and you can look at that as a risk. And there's also an opportunity, right? The opportunity is, I'm so glad that you're interested in creating value with me leveraging this. Let me show you my toolbox where I have several other tools, many of which are cheaper to scale more efficiently and have less risk associated with, and that everything's a nail right on a golden. Exactly.
Precisely. And so the question of how we scaled our technology stack is really rooted in the way that I think most many enterprises explore this, which is in early 2023, we said, okay, there's a new. You'll hear everyone's excited. We have Jenny Media. Let's rapidly and responsibly explore what value opportunities might be unlocked by this technology. Right? And so our technology investment was not let's build it and hope they come. Let's start with business cases. That could be ideas, right? We're innovating here.
Does that be a fully baked business case? But start with ideas. Don't start with what does this tool do? And so the way that we've scaled up has really been opportunistically based on use case and success has been measured the same way that any business initiative would be measured by the success of the program rather than by this, let's say, the quality of the prediction or the the frequency of hallucinations. And along that line, Dr. Bloom, off, when we think about tech infrastructure, a lot of change in terms of transformation to the cloud. How has your company manage that and how does AI impact? Well, we've wanted to take an approach of sort of enabling air across the whole company.
Sort of to the point of that, you know, okay, not everything is a nail, but not everything is in Let me either to my earlier point, there's a lot of different AA models out there and a lot of different approaches. And given that breadth, there actually are good AA solutions to many, many problems, and it's applicable across large swaths of the company. So rather than trying to take a centralized approach and have a sort of an AA organization that builds the AA technology for the company or has to make a large investment there, the approach has been really one of sort of grassroots enablement, and that's been a big part of my job as CTO is to basically enable everybody at the company to understand what it is, how it works, what its limitations are, and basically enable anybody at the company to use AA, but do so in a responsible and insane fashion. And that means that we're adopting AI in many, many places across the company, but doing it in sort of a right sized fashion. Again, not everything is a giant limb and and that helps minimize the investment. Also from the point of view that it, you know, limbs are expensive to run. But if you can solve the problem with
something much like a smaller deep learning model at potentially 1/1000000000 and I mean a billionth of the cost, well, you don't need to buy those in video GPUs. You can get by with a much more efficient solution that actually is going to work better and have far less risk associated with it. So starting with an investment in enablement, I think is reaping rewards for us. And a question for us is, you know, there's so many internal stakeholders for A.I. when we think about different departments, marketing versus finance, they have a different goal, a different ambition for A.I.. Kip, How do you decipher what is a higher priority A.I.
project and which has a higher rely, which is several years out, which a lot of people don't have that type of visibility? Yeah, I think when it comes to A.I. right now, what we're trying to do is we want to prioritize projects that have like one very clear metric of success that we feel like is achievable in the short term, but will improve dramatic dramatically as the technology in the models improve. Right? Like we want to have the right infrastructure in the short term to take advantage of it, but also in the long term to get additional scale and upside. And so we essentially just have kind of
cross-functional pods aligned across a smaller subset of metrics and goals. And so, you know, sales, prospecting, for example, it's like it's already more, you know, a AI driven email prospecting is already more effective than email, human email prospecting. But as the models get better, you are then going to have the that same motion apply to specialized selling to other aspects of the sales process, and that we can kind of take the infrastructure that we're building now and apply that in future cases. And what is that metric that you looked at for sales prospecting specifically? Yes. If you think about the sales, prospecting is, you know, how much productivity per rep can we basically drive? So if you let's say most companies have kind of a business development rep or SDR function, you can say, hey, well I can, I can generate the equivalent of 30% of my team, 40% of my team, whatever that may be like. You set a benchmark, you kind of chase
after that. And that doesn't mean that you hire less people. It means you have better market coverage. For for us, we're in a big greenfield market and we want to go after that market as fast as we can.
Let's zoom out a little bit. You know, we're talking about A.I. early. How does A.I. rank in terms of your digital innovation priority list? I think it's in the mix. Right. I think that's what you kind of get used to working in this field is that there is nothing that's ever not net new or pushing. Right. And that could be a new language that you're developing in. It could be a new tool that's coming
out. And I think that the good thing about A.I. is it's been around, Right. I think, as Robert said, this is not a net new. Right. I think the attention isn't that new,
which is exciting. Right. And we're excited about that. And I think anytime you can garner excitement to corporate budgets, to investment, that's even more exciting, right, for us. And I think in that same vein, we see it coming up in a lot of our partners. I will call out as well.
Right. Because it's not just large businesses, it's partners that we partner with are also digging into this to find endpoint solutions that we want to engage with there. It's something that we now look out for in our net new contracts as we go up through re signings. It's also obviously deeply embedded and I think in the digital teams of product engineer design, under under my privileged watchful eye, I get to also just know that my team is honestly thinking about it all the time because they too are using it, right? They're using it in their personal lives, they're using it in their work lives and sort of be able to bring that forward. They are just used to being in the environment, right, in agile environment of wanting to progress in that way. And so I think when we take all of that into consideration, what we do in our normal day to day, it's absolutely just another vein in, in the same way that we felt mobile come in, right? We felt a mobile innovation wave come a few years ago and now we're feeling, you know, another wave of just new products and new services that we can integrate to make our experiences even that more robust.
And when we think about a framework for an organization or a corporate structure for an A.I. Council, Alex you know, high regulation and insurance and financial services, do you have that internally at MassMutual? Yeah, we have. I would say informal. We don't use the term counsel, but in our business, as with many trusts, is paramount. Right. So I mentioned earlier when we started in 2023 with this new tool, Jenn-air, we did a rapid and responsible exploration. So we did that by explicitly derisking how we explore these. Right?
And I think there's always a question, does does that counsel help or hinder the use of air in the business? And my answer is it depends on where you're standing, right? I think very often business partners feel like we're left out. Things are moving slower than we want it to move. Right. And some in some sense, that's that's the purpose.
Right? Was a famous Mario Andretti quote brakes aren't don't exist to make a car go slow. Brakes exist so that the car can go fast. Right. So in I council, we have a mass mutual's ensuring that we're making the right investments, making the right choices. And it's something that we've felt pretty passionately about at MassMutual for some time preceding Jeffrey Beatty and lln mania.
We independently stood up an air governance council. So kind of we're doing responsible AIB for responsible. I was cool because this is something that we think is important to continue delivering on the generations long trust and commitments that we've had with our policyholders. So again, you know, as we consider this topic of digital innovation and I what comes to my mind, Dr.
Bloom, off, I'm thinking what is your advantage relative to other peers? We hear good air stories, we hear the good investment cases. How can an investor differentiate? Akamai I pathway may have a higher ROIC versus company A, B or C, but don't think it's a high per se, but I think it has to live in sort of a broader context to evaluate one company against against others in any particular. Vertical or any particular area. When it comes to cybersecurity, I think we're at a point where it's hard to imagine a cybersecurity vendor not using A.I.. Now, that doesn't necessarily mean that they're really doing everything with low ends, but that means they're using some form of deep learning. Again, it's just inconceivable that you could be doing anything productive in cybersecurity without some form of deep learning. I mean, five limbs are a form of deep learning.
I think when it comes to cybersecurity, though, actually probably what's more interesting is what's happening on the other side of the fence, which is what's happening on the attack side. As good a tool as is in cyber defense, it's actually an even better tool in cyber attack. You know, I sometimes joke if if your job if you're a cyber criminal, if that's your job. November 30th, 2022, was the greatest date in your life, in your professional life, because that was the day that was released. And it is truly like manna from heaven. If you're a cyber criminal and I think we're still in the early days of of criminals adopting this thing, I think the world of cybersecurity is going to go through a very interesting. Put that word in quotes, five years to
come as we as the criminals start adopting these tools. And the landscape on the defense side is going to have to change quite a bit, I think. And by the way, there isn't a magic A.I. solution either. And anybody who's telling you that they
have an A.I. solution to all of your air attack problems. They're selling you snake oil. You know, I could, you know, wax for a long time on what it takes to defend against air power threats. But air power threats is a game changer in the world of cybersecurity. And I think what's going to
differentiate here to your question is how cybersecurity vendors respond to that threat. And on those points, how do you stay one step ahead of bad actors if they have access to similar type of technologies? Is it because of the scale of investment that you have? Scale is a very important part of it. The technology, the data you see, I think this is maybe your first comment. You know, it is absolutely true that the quality of AI depends on the data. I mean, everybody's got the same models,
right? Lens are all based on a thing called the transformer model. Everybody's got transformer models. It's there's no differentiation at all. Everybody's got access to GPUs. I know there's temporary shortages and things like that, but come on, everybody's got access to compute. Everybody's got access to the same models. It's only the data that makes a
difference. So you made that point at the beginning. I think a number of you also agreed with that point. And the volume does matter here. Now you do eventually reach a flattening point, but when it comes to a specific domain like cybersecurity, quality of data and volume of data, because when you're training A.I. models, the volume does matter. So those things can make can make a difference. And the points of maybe misinformation
from third party limbs, hallucinations. Is that an area that you have had to contend with early? Yeah, of course. I think a lot of it comes down to when we think about not just consumer facing, right. Because that's a very hot area for us, is the ability to do a customer service type based engagement. Right. So we watch that incredibly closely. But also on the internal colleagues side, right.
You have failures everywhere that could happen, right? If you're not taking care, doing your proper governance, watching things appropriately, using your own data to train it. Right. And I also think taking things one step further, which is toxicity filters and things of that nature. Right. So it's not just the model to be able to respond, it's the response itself being correct, but then also the tonality, the correctness of the information. You're not being able to ingest and answer a question back, right. That could cause some issues. So I think for us, it's an incredibly
hyper focused area of not just experimentation but also governance and how we feel comfortable with what we don't. And I think that goes back to a point that was made, which is what are those right solutions to make? Right. And I think that there's an easy button that everyone wants to hit and some people have hit it, and that has been out there. Right. And I think the ones who are
on the I would say, thoughtful and again, leading with what you want the outcome to be will have a slower pace because they will ensure that that experience isn't just the easy button, because it's easy for a moment. And that obviously can cause other issues. And I think that I'm very proud of the team, quite frankly, internally of how we've gone around this and again, doing it in phases. We didn't just launch an entire new feature, right? We did our question and then built from there.
And then once we felt confident in that one question, we added more. And so I think again, going in a phase where you can get that general antithesis of what you're trying to achieve out there, then you can go and go faster. I like that term easy, but can I still go for it? Just don't push it.
No. Alex, head of data science. You cut across all functional areas. h.R. Finance, marketing, sales, which area? When we look at ai impact internally is going to be most affected from your view, whether it's using ai assistance to do the scope of the work or even replace, you know, scope of work from a person to an ai system. If we're specifically talking about the opportunity that will be unlocked by generative AI, what's the scope it down to that? So it's precise. I think the largest impact will be on all of the functions that directly support policyholders and customers. And the impact will be by enabling the existing experts to outperform their previous selves, by increasing their efficiency, the quality of the service. And I don't necessarily just mean
customer care representatives, but to finish professionals more broadly, fiduciary is even lawyers to improve efficiency, quality and the the satisfaction that they have with their own jobs, as well as the satisfaction that they give to their customers. So things like reducing claims time, reducing the time to take a loan out, reducing the time to instantiate and support trusts and wills are all things that are actively getting work together. And you know, these cut across those functions of financial, professional fiduciary, lawyer, customer care representative to deliver that holistic experience that policyholders expect. Cap same question.
You know, a lot of automated tasks in sales are being addressed by new technology. How's that going to in fact impact the salesperson future and time prospects is employment. Yeah, look, if you look at the go to market sales, marketing and customer service there, there are a few I think fairly obvious things are going to happen. Sales and customer service representatives are going to get transformed with copilot and AI is going to make them incredibly more efficient. Just as Alex said, like that is 100% true. The other thing that may be slightly counterintuitive is go to market models will become a lot simpler because what happens now, let's say you are a big infrastructure technology company or big manufacturer. You might have a field sales rep, some
specialty product sales reps, then you've got a contract manager, you have ten people trying to get this deal done. You're going to be able to automate a lot of that technical knowledge and make that accessible directly to the end customer in customer and reduce a lot of the friction in the buying process. I think it's never, never been a better time to be a customer and to have a better, smoother transaction. And it's only going to get better over the next few years. I think that's what's going to really transform in the go to market. Interesting and thinking about
challenges. Dr. Bloom, off you have over two decades of experience at Akamai. What was an oh wow moment from an AI or a new technology over the last five years from your eyes? Well, deep learning was one of those. Again, before deep learning, I was by and large irrelevant. It was kind of an academic curiosity. You know, I studied in grad school at the, you know, grad level. In fact, I'm married to an air person,
so I've always sort of viewed as a very interesting academic curiosity. But with deep learning, all of a sudden you could see applications in actual using, you know, in enterprises by people, those that the term deep learning or people familiar with, you've heard the term. But are you aware that there were deep and deep learning has nothing to do with the depth of cognition. It only refers to the depth of the
neural network that's being used now anyway. Yeah, that was an AHA when I first saw it. Deep learning models that could actually do useful things. And in the context of Akamai, that really meant in the cybersecurity realm classifying traffic as good, bad human bot, things like that.
And then, you know, obviously the other one I think for a lot of people was when you first got first saw a demo of a large language model, you know. LLN This by the way, predate, I mentioned November 30th, 2020 to our predate that by quite a bit by four, three, four or five years, something like that. But the release of chat GPT was the first time anyone was really released in a chat bot form with a decent amount of alignment so that it could actually answer questions and do useful things. And I think for many people that first spectacular demo of carrying on a conversation with an ally was an aha moment. Certainly was for me. What I like what you just said, as can I think watching people then see that and the creativity that will come from just humans experiencing that themselves in the new businesses, in the new lines, will be a has for many, many moons to come if you think of it as watching and that I've ever played with an image generator image generator.
If you have access to and try this out so you can play with the really cool and you can generate these really cool images that'll just wow you. But if you have a very specific task in mind, you want an image that is very specific. It's actually pretty hard. They don't steer as well as you might think. And again, and I say that not to downplay the spectacular technology that these things are, but to just recognize the limitations and recognize that, you know, things maybe aren't quite what they seem at first to demo I can confirm this. I use this for my children who want very
specialized coloring pages. And when you're very specific, yeah, it's amazing. You try to use it like in a presentation. I need an image of a specific thing for a presentation. You may get what you want, but chances are you won't. And the more you try and steer, the worse it'll get. Let's wrap this up with one thought from
each of our speakers here. Revenue side or the cost side? What is going to be the biggest impact from adopting A.I. for your organization? Marley I think we'll be equal. I think we have a lot of promising activities, many of which you touched on, right? Because we have the customer doesn't care how complex we are to book a room. They don't care how complex it is to
price something one time a day versus four times a day based on availability. Right. But I think what will be equally interesting is we will be able to find our guests, will be able to find our members, will be able to grow our business because we will be out front with those use cases are and being experimental with it. So I think we'll see an even uneven return. Kip Sales our cost sales far and away. Look, you're going to get it's efficiency right now because that's what people are comfortable with.
That's what the models are kind of a little bit better at today. But long term, as we have like full societal change and impact from generative AI, it's going to be a growth driver more than an efficiency driver. Alex I was early on this. I think the cost savings are going to be huge upfront and that's going to feed top line growth in a variety of different ways. I think it's expectation management can do a lot less than in a year than most people think and a lot more in a decade than people think. Yeah, a tripling of the animals, I
think. I think are they had it had it right. It is reasonable balance probably on the revenue side, it's actually more deep learning on the cost savings, it's probably more low. But yeah, I think it's going to contribute meaningfully on both sides. Great. With that, will wrap this up.
Thank you, everyone. Thanks for and appreciate it. Thank you all.
2024-11-02