Good evening, everybody. My name is Eva Dong, and I'm here to give a talk for value realization for AI.

I just want to do a sense check. Have anybody been here this auditorium last week and saw me? One person?

Well, thank you for coming again. Three people, thank you for coming again. You are my VIP. I prepared special things for you.

But for everybody just meeting for the first time, my name is Eva Dong, and I want to give a short introduction about me. I work for Google Cloud.

So within Google Cloud, we are one of the major Cloud providers, obviously, and then we also provide professional services to our enterprise clients. and I'm part of that professional service, and our organization is called Delta, because we want to make the Delta impact for our clients.

Before joining Google Cloud, I was at McKinsey for eight years. It was a very humbling experience with terrible lifestyles, and I made it through it for eight years. I'm very proud of myself.

I really practice a lot of all kinds of skills, well-rounded. And before that, I was a machine learning data scientist at Visa.

So, let's dive into today's talk. We love to do like a quiz, or not a quiz, more like a guessing game with everybody. So, we interviewed a lot of enterprise leaders and asked them, what do you think is the number one criteria? when you evaluate and select a generic use case.

This is a histogram, which is a hint. So, 30 percent of the users think this is the number one criteria, and there's another 26 percent of leaders think this is the number one criteria. So, any guesses about what is the criteria?

Reducing staff. Reducing staff? Well, that's brutal.

Hallucinations. What is that? Hallucinations. Hallucinations. Yep.

Security. Security. Doing something faster? Yeah, faster. Faster, better.

Anybody else? Data? Like having enough data or generating enough data? generate enough data. Cool.

So, the third-party survey shows the number one criteria is easily quantifiable ROI, and then the second one is customizable. A lot of people think price will be very important, but actually, surprisingly, it's not even in the top one. not even in the top five.

If you feel a little bit surprised about why ROI becomes so important, I'd love to walk you through a timeline and break it down for you. I'd love to take you back on a journey since 2022. So 2022, November, ChatGPT come to people's life, triggered tremendous interest from the public and from the enterprise world.

Everybody's so excited about it. Everybody want to try it. So when it come to 2023, everybody's doing POC of some sort of Gen-AI, but everybody's innovating.

What is the latest model? I want to have the latest model now. So at that time, the entire industry is in the innovation race.

But what come with it is inflated expectation. So the technology was wonderful, but sometime is not completely mature. So that come to the next stage, was it like 2024, we come through like a period of disappointment.

That is a time when enterprise phase like integration issues, integration challenges, implementation challenges, so become so disappointed. Oh my God, this thing is so expensive.

It's not like $10 million on this. Why do I now have the return I'm expecting? Now, it's 2025.

we're on this slope of enlightenment, or at least I'm trying to take my clients to the slope of enlightenment. I want to have the right expectations for every application, every service they bought, and also, most importantly, I want to bring ROI and I want to bring the actual return for the spend they have spent on the ROI. experience we find the best winning strategy is not to keep innovating, not to keep adopting the latest model whatsoever, but it's actually pursue value.

But it's actually easier to say than done because AI brings so much uncertainty. What we try to do with this offer value realization for AI is navigate with our clients through the stages of uncertainty, reduce that, and increase ROI at different stages.

So at Google, we define the innovation into four different stages. So, the first one will be activate. So, this is time when enterprise want to adopt AI but don't know exactly what to do.

They might have a list of good ideas but don't know where to start. So, here we will try to help them like, what is your advantage? What's your disadvantage? What is your roadblocks in your current operation right now? How big is the opportunity? Then, what kind of use cases can you exhaust?

The second stage will be navigate. So, navigate will be, you already have a priority list of use cases, you know what you're going to do, you're planning, you have a good idea of what you're doing. But at this stage, I would say the RI is still very vague.

You have ideas but don't know exactly where they are, and sometimes uncertainty is still high because this is all ideas before you actually do anything. The next stage will be accelerate. So this is okay, you have three prioritized use case. I'm going to develop like AI co-pilot for my employee. I'm going to have a chatbot for my customers. I'm developing the AI, I'm integrating, I'm bringing my data to the model, I'm testing, I'm doing all kinds of A-B testing.

With this, probably your ROI is more and more clear and your uncertainty is pretty low, and the last stage will be transform. So, this will be a stage you scale the AI to entire business to encourage employees or customers, and this is the moment you should see the ROI become realized ROI instead of just like projected ROI.

So to summarize what I just said, so we at Google Cloud, we'd like to help our customers through every stage of the funnel. So regardless whether you're super early, like you just want to use AI and have a vague idea, or you're already prioritizing planning to a time you're already developing, to the time you're already like scaling, we want to help you through every way.

So, the second and third row, I think that's most interesting. So, an ROI would define value drivers or value levers for you. For everybody, it's a little bit different and people are thinking on this could be a little bit vague. So, for example, have a client who is developing an AI co-pilot for their employees, so their employees can save time on searching and generating documents.

and we projected, okay, therefore your employees can save 10 hours per week by using this pilot. My customer is very happy, great, awesome, done, done, done, out of the door.

So, I like to remind them at this moment that's not the end of the story yet. Just because they spend 10 hours less per week on working, it doesn't mean they're doing 10 hours yoga extra. Also doesn't mean you're paying them 10 hours less. Therefore, what are they doing for the 10 hours saved?

Are they picking up a new product? Are they working on a new work stream? How much incremental revenue does that new product or new work stream mean for you? So, eventually, how do we translate this from this AI co-pilot to save 10 hours per week to your bottom line revenue increase to your bottom line profitability increase?

So, that's the journey I like to push my customer on. But still, sometimes it's still easier to say it than do it.

So, I summarize some key learnings to share with the fellows today. So, there's four that I'd love to share. The first one will be, ideally you want to solve your customer's pain point as early as possible. The earlier you solve their pain point,

the bigger the impact will be. I'm now happy to bring two use cases after this. The next one will be value usually come from four Cs, that is coding, customer engagement, creative content, and concision, which means like virtual expert, usually a virtual expert for your call center, call center and for your employees. So, if you are still early in your journey adopting AI, those four different areas are usually the best area we see significant return.

This is the areas where the technology is very mature, there's a lot of application, there's a lot of use cases, and it's probably very easy for you to have impact here. The third one will be, Jenny and I look simple on the surface. So, a lot of people focus on IOM, but based on our research, that only generate about 15 percent of the impact.

The rest 85 percent of impact come from the underlying infrastructure. How good is your Cloud infrastructure? How stable is the infrastructure? How fluently or smoothly does your data pipeline flow?

How clean is your data? So, all of that is super important. So, I know there's a lot of service.

There's no code, just use our platform, blah, blah, blah. But at Google Cloud, we believe in differently. We believe you need to have a stable and strong infrastructure before you adopt any Gen AI use cases.

Lastly, is adopt a two-by-two approach. So, I think when it comes to genuine use cases, a lot of times people pick productivity as top of mind. How do you save call center staff time?

How do you save employee time? How do you save customer time? Productivity, obviously, great place to start, but a lot to encourage our customers to dig a little deeper into innovation.

So, we recommend If you have the energy and investment, we usually recommend two on productivity and two on growth and innovation.

I'd like to shine this in light into use cases because I have three VIPs here. I'm going to choose a different use case.

So, the first one I want to talk about Walmart. I'm sure more or less you have shopped at Walmart once or twice, although I'm talking about walmart.com today.

Let me actually just go into a demo. So, Walmart, you come here and you search. This is like the typical way.

But what we did is like semantic vertex search. So, here, so I was like trying before this. So, here like Easter is coming up.

So, I would say, help me plan an Easter party. I didn't say what, I didn't say I want decoration, I didn't say I want food, I didn't say who's coming, so I just say Easter party. So, this can help me brainstorm a lot of ideas like what I should buy and shouldn't buy for my party. So, let me make it a little bit more interesting.

What about Easter dinner party? It's slow. That's the Internet, that's not Vertex AI.

Okay, because what I'm hoping to see is some place like this because I said dinner party, but it seems like it's still concentrated on the decorations. Okay, I see some plates, some decorations, or you can say, oh, help me find a Mother's Day gift. Let me just type that. Help me. With this, it saves customer time to think about what they need to buy.

Maybe you're brainstorming in your mind originally. For my mother's day, should I buy flowers? Should I buy candles? Should I buy jewelry?

Should I buy her favorite beddings? With this, it saves you that brainstorming directly. This is more like mission-based search or semantic search instead of searching in a category or searching by keywords.

This is different type of search powered by Vertex AI. So with this, the impact we have seen is especially during holidays such as Valentine's Day or Mother's Day or Thanksgiving, 1we see mission-based search become top of search. It's like people search more mission-based than keyword category-based.

Also, because you want to implement this Vertex AI semantic search, 1the AI need to go through all your product catalog. For Walmart specifically, they have 850 million products currently selling and constantly updating.

And with AI, I save them 99 percent of the time. So, they're spending one percent of the time to go through the entire catalog than what they usually do. And then also because you're not searching for a key category, you're searching across category.

So, we find out it actually trigger a lot of high-margin cross category impulse purchase. So, you might come in and you just want to buy a necklace for your girlfriend. and you see this candle. Oh, this is awesome.

Let me add this. Oh, how about this flower? Add this too. So, that's the impact what AI is playing on you and watch out for that.

Eventually, also have an impact on loyalty. So, Walmart's membership is called Walmart Plus. Because of this good experience, they also triggered more than 10 percent growth and eventually boosted sales.

So, let's think about the learnings I just mentioned. First, I said solve customer's pain point as early as possible. So, this is at the beginning of their purchase journey. Before they search anything, help them bring from what to buy.

So, that's solving the customer pain point in the early on and secondly, I said like value come from four Cs. So, this one will be customer engagement.

So, like one of four Cs and also Walmart is a very loyal Google Cloud customers. So, we build all the fundamental infrastructure for them and this is one of the use case. So, do you think this is a productivity use case or innovation use case?

Innovation. Innovation. Yes, that's right. Cool.

I want to talk about another use case, Wendy's. So I'm sure we have all been through this kind of experience.

You're at a drive-through for a quick service and you've heard an order, and through this tiny speaker, you have to tilt your body outside a car and yell, can I have a cheeseburger without onion? And the staff on the other side of the speaker, okay, you want a cheeseburger with onion? and you yell, no, without onion. He said, okay, without onion. Eventually, when you get it, it's still with onion.

So, drive-through is actually most preferred way of ordering. So, for Wendy's specifically, 80 percent of their customers want to use drive-through at their preferred ordering system over person-to-person order or ordering on a tablet. However, this is also the channel that takes the most time and have the highest inaccurate rate for the order.

So, we implement this AI enhanced drive-through experience called Fresh AI with Wendy's. So, basically is a conversational agent taking order with you,

Then what we find out is one, the order accuracy improved to 99 percent, so only one percent of time it was wrong, and also it shortened the conversation a lot. We bring it to under 22 seconds of the market average, which is a lot because you think about it, it's usually like a one-minute conversation to order, and now it reduced by 22 seconds. And 86 percent of the order didn't really need a human intervention.

And eventually, because we saved one staff on the other side of the speaker, so we also increased the labor efficiency. Therefore, the increased margin by 0.8 percent. 0.8 percent for something as national and as large as Vindy's is a huge, huge win.

So this is critical. And eventually, we also boosted sales.

So with this use case, again, think about the learnings. So this is, This is solving customer's pain point early on because ordering the food is the beginning of customer's journey from ordering to eating to ordering again. So, you're solving their pain point early on.

This is coming from the four Cs, which customer engagement, Gen-A, like the AI infrastructure wouldn't just look very robust. Do you think this is a productivity play or innovation play? Yes, that's very good. It is. It saved some operation efficiency, which is productivity, but also innovation because people have a different experience at a drive-through.

Conclusion

Okay. With that, I'm at time, right? Yes. So with that, I want to conclude my talk.

This is my LinkedIn. Feel free to connect with me if you're interested in value realization for AI or interested in Google Cloud or my journey.

I think the last thing I'll just wrap it up as value is really important regardless whether what you're doing with AI, always keep that in mind because eventually think about like, why are we using AI? It's not really, you know, to make us cool, make us sound fancy, is really to bring value to our customer, to our employees, and to ourselves.

Thank you.