Teradata Possible 2024 | Vedat Akgun, Ph. D & Pranay Dave, Teradata

Clips
News
More from Teradata Possible 2024

Vedat Akgun

VP, Data Science & AI

Teradata

Pranay Dave

Director, Product Marketing

Teradata

Teradata targets customer complaint resolution with its Vantage CX solution

With bottom lines at stake, companies must walk the walk, roll up their sleeves and get their hands dirty finding practical, in-business value areas for AI. Beyond viewing the technology as some sort of holy grail, companies must look through the weeds to pair individual AI capabilities with specific business operations. Teradata Corp.’s Vantage CX aims to enhance one such operational area: customer experience.Teradata’s Vedat Akgun and Pranav Dave talk with theCUBE about applying AI to CX use cases.“I think AI is impacting quite a lot in this customer experience kind of scenario,” said Pranay Dave (pictured, left), global director of product marketing at Teradata

play_circle_outline Importance of creating value with data and AI

play_circle_outline Customer complaint analyzer for resolving customer problems efficiently

play_circle_outline Focus on sustainability in AI journey and using smaller resources efficiently

play_circle_outline Integration of models with rest of data for operationalization and efficiency

play_circle_outline Unleashing the Untapped Potential: How GenAI Can Revolutionize Analytics by Reducing Failure Rates

Info
Transcript

Vedat Akgun, Ph. D & Pranay Dave, Teradata

Vedat Akgun

VP, Data Science & AI Teradata

Pranay Dave

Director, Product Marketing Teradata

At a Teradata Possible event in Los Angeles, Pranay and Vedat discuss the importance of AI in creating business value, especially in customer experience. They highlight their work on a customer complaint analyzer, emphasizing the need for efficient resolution to keep customers satisfied and reduce churn. The complexity of analyzing complaints in multiple formats like text and voice is addressed, as well as the importance of data preparation in AI projects. The integration of predictive AI with GenAI is seen as a way to improve resolution processes and create ... Read more

explore Keep Exploring

What is the most important thing for AI in terms of business value? add

What is the importance of analyzing customer complaints and how can AI help in resolving them efficiently? add

What are some strategies for customers to make their AI journey more sustainable? add

What is the new capability that was released today regarding bringing your own class language model inside your platform? add

What potential value have we not yet achieved in predictive and prescriptive AI, according to Martin Wilcox's presentation? add

bolt Powered by CUBE AI

Vedat Akgun, Ph. D & Pranay Dave, Teradata

search

Savannah Peterson

>> Good afternoon data community, and welcome back to downtown Los Angeles, California. We're here at Teradata Possible. My name is Savannah Peterson, and really enjoying my day next to Rob Strechay. Rob, have we had as many fun guests on before in a show as we've had today?

Rob Strechay

>> I don't know. It'd be tough to... It's a tough comparison. That's for sure. They're making it tough to compare to today. This has been a lot of fun.

Savannah Peterson

>> You're such a diplomat.

Rob Strechay

>> I know.

Savannah Peterson

>> I know. This has been a total lot of fun. And speaking of fun, we have two really fun guests sitting here on the desk with us. Pranay and Vedat, thank you so much for taking time today.

Rob Strechay

>> Thank you.

>> Thanks for having us.

Savannah Peterson

>> This has got to be a wild week for both of you getting to see your customers, your community, your teams. Vedat, what does it mean for you to be here this week?

>> It is a great experience to get to know multiple people, not only from our client side, but also our partners, as well as analysts and other prospects. So multiple industries, you get to know different people from different industries and background. That's been amazing.

Rob Strechay

>> And a big piece of what we've been talking about in getting to know that is that people are always inquisitive about how they get value out of their data and how are they creating value with their data. A lot are leaning on AI as, oh, this is the holy grail for doing that. AI has been around for more than a minute now. So how do you see it with the customers? And how have the customers really been leaning into creating value with their data?

Savannah Peterson

>> Yeah, when you said AI has been around for a long time, of course, I am an AI practitioner myself, so I did my PhD and worked in multiple industries and companies and trying to get AI and machine learning to value. So I know the challenges, what I went through. I have successes, I also have failures, and I can relate to it really well. And that's why business value is the most important thing for AI. Unless you can show that the cost savings or revenue increases more than the investment, then there's no point. That's why we purposefully defined our trusted AI as the way that people, data, and AI work together with transparency to create value. And when I explain this, I usually start at the end because the value is the ultimate goal that you need to achieve, that you need to have cost-effective, innovative, and performant AI in production. And the way to achieve that goes through transparency and people. So business value, the most important thing for AI.

Savannah Peterson

>> Absolutely. And the ability to... Like you said, we talk about making AI real. It's not real until it's out in the wild and my mom's interacting with it and getting some benefit and value add from it, not just in these POC use cases like we've been talking about. Speaking of value, CX is a big space here for AI, one of the earlier verticals of adoption within organizations or groups within organizations. We haven't had a chance to talk about it, but you have a custom complaint analyzer, correct?

>> Yeah.

Savannah Peterson

>> Can you tell me about that?

Rob Strechay

>> Yeah, sure. So first of all, thanks for having me on your channel.

Savannah Peterson

>> Our pleasure.

>> And I think AI is impacting quite a lot in this customer experience kind of scenario. And we have been developing a customer complaint analyzer, and we have also found a lot of insights into it. So for any customer, the top things, what he has on his mind is to resolve his problems, right? You as a customer, you may be interacting with your banks, with your retailers, but for you, the most important, okay, is my problem getting solved or not?

Savannah Peterson

>> Right.

>> And many of the customers would even walk away from a brand just after one bad experience, right? And to give you some statistic behind this, if you take the US financial industry, most of the banks, they receive about a million complaints just in a span of two or three months. And this data is coming from CFPB, which is the Consumer Financial Production Bureau. And you can imagine the volume, but also the types of complaints. You can have complaints which is coming from text format. You can have complaints which is coming the voice format. People complain about simple things like, "Okay, my creative card is not working." On the other hand, they can also complain about... Guys, you are doing some kind of insider trading, right?

Savannah Peterson

>> Right.

>> So you can have a vast variety of complaints. So it's very important to understand what are these complaints, what are these topics, and how to kind of quickly react back and give a resolution back to the customer because this is where you are adding value and you are reducing churn. So our customer complaint analyzer does that exactly. It tries to analyze, with the power of AI, intuitive AI, different type of complaints, and making sure that at the end, the customer is happy and we have perfect resolution for that particular customer.

Savannah Peterson

>> I can really see that helping out . A million complaints in a quarter is what you just said there, which is pretty nuts. How long does it take you to tune a model like this or to come up with that solution? I'll keep it with you for now just for a second.

Savannah Peterson

>> Yeah. So what we are currently doing is most of the time, it goes in the data preparation because we have to combine voice data, we have to combine text data. And I think from a timing point of view, it on different projects, but I can say it could take about 80% of the time goes in the data preparation.

Savannah Peterson

>> And even just with that voice and text, you've got people with different accents, lexicons, speaking different language. The Z axis of complexity starts to get pretty wild there.

Savannah Peterson

>> And also to give you... We're also doing a project in Saudi Arabia, and the text is reverse, right?

Savannah Peterson

>> Right.

>> And the language is not English, right? So you can imagine the complexity of all the data preparation, and also kind of integrating different data sets. You can have the same customer who has launched a complaint in text format. But the same time, if it doesn't get resolved, he will call, right? And I'm from France. I know that people like calling quite a lot, right? So you get a lot of complaints, same person calling again and again. And so it's important to process the data in a nice way. It's also important to relay it back, that who is the customer behind this complaint. Identify what is the topic of the complaint, and then we can start applying the generative AI and all that kind of stuff on how to best respond to this complaint, right? So data preparation takes quite a amount of time when it comes to AI, fortunately, we have our cloud partners who have excellent large language models who will go through all the different texts, all the different formats, and will tell you, okay, this is a topic. They also summarize the complaint for us. The complaints could be thousand words long, and they can summarize it into nice texts, like two, three line text so that bank people can understand what it is. And then, of course, the biggest thing, I think, is, as Vedat was telling, what is the value? So you have to somehow operationalize it, right? It's good to know what the people are complaining about, but then you have to come out with the resolution. Because you can apply all the models, but if you don't resolve the complaint, you don't add

Savannah Peterson

>> It doesn't matter.

>> Exactly.

Savannah Peterson

>> And then people are even more mad that-

>> Exactly....

Savannah Peterson

>> you wasted time messing around with AI-

>> Exactly. Exactly....

Savannah Peterson

>> instead of solving my stuff.

>> Exactly.

Rob Strechay

>> Right. And how does that really tie GenAI into this? Because I think for me, I can imagine how it would be, having filled out and had to over a thousand word complaint to a particular airline that I won't talk about, but they resolved it, they solved it, but it didn't fit in the space on their website, so I attached a word doc with the full text. To your point, I created more-

Savannah Peterson

>> You were determined to send that complaint.

Rob Strechay

>> . I was very determined. And I usually don't do that, but they pushed me, and it is what it is, but so... And I'm going to generally assume that it went into some GenAI to do summarization and pick out... How do you see, in this way or in other ways, GenAI playing a role in there?

>> Well, this is... As Pranay mentioned, those complaints come in different shapes and forms, right? If you go to predictive AI only, you need structured data, so you would spend time to convert or find a way to convert a script of an audio to structured data. But GenAI is a perfect tool and technology to bring all those different complaints together, and then analyze it and give you a perspective for resolution. So it is also a great example where you can actually combine predictive AI with GenAI. As Pranay said, after you analyze the complaints, you need to find resolution. At times, you may be able to use predictive model to say, "For this customer's complaint, this person or this department would be the best to resolve that so that you are able to reduce churn or increase retention, or increase conversion. So it's a fantastic example for GenAI to create business value through resolutions of those complaints coming from multiple data sources.

Savannah Peterson

>> Go for it. Yeah. Yeah.

Rob Strechay

>> Just because you hit on something that was really interesting to me, do you see GenAI as being able to help build the data that then can be used either in predictive or causal-

Savannah Peterson

>> Good question....

Rob Strechay

>> models afterwards? Is that-

>> Yeah. Actually, thanks a lot for reminding me. I was going to say that, and I forgot in the line of thought. Well, as you resolve those complaints, and as you analyze them and you classify them, that will become a learning for that GenAI model for the future complaints. And therefore, you are able to build the know-how to faster and better analyze them and faster and better resolve them because of that learning. So it creates a great opportunity from that perspective.

Savannah Peterson

>> Go for it.

>> I think this is very critical point, and it also shows the future of GenAI. So currently, projects, most of the people who are resolving our humans, right? But imagine a future where for certain type of complaints, you have AI resolving it, right? But that future would happen if it learns from the current resolutions. So this is where the data which is generated, and how actually do you resolve, right? In the future, it'll be augmenting what we call it agents, resolution agents,-

Rob Strechay

>> Right....

>> which would resolve certain type of complaints so that it is much faster and much efficient way to kind of get the resolution and the customers happy. You go the resolution faster, as well as a lot of saving also for the enterprise.

Rob Strechay

>> Yeah.

>> Yes.

Rob Strechay

>> We talked about them being small action models and groups and chaining of agents and things of that-

Savannah Peterson

>> Yeah, I wanted to connect it to sustainability because-

Savannah Peterson

>> Let's talk about it.

>> I am very passionate about it because we saw that everyone talks about the value, value of GenAI, but people sometimes don't talk about the cost or sustainability. If you build a 350,000 GPU data center, it's not going to run on air. It needs power and water. So if you are able to resolve problems such as customer complaints by using very well-trained small language models, that means you are not using as much energy. And those are really important for sustainability perspectives as well.

Savannah Peterson

>> What are some of the other ways that customers can make their AI journey sustainable? You mentioned using smaller computes so that you're using less energy, but there's got to be other ways that AI makes these companies more sustainable in the conversations that you're having.

Savannah Peterson

>> Yeah. I think some of our speakers today on the main stage talked about it. As you said, depending on the specific business problem, using sometimes small language model on CPU, not even on GPU,-

Savannah Peterson

>> Right....

>> will give you the value. And if you are very smart, in terms of finding the most impactful use cases that require optimal or optimized level of resources to create value, that's how you do that. And also, another thing our presenters mentioned, and I liked it a lot, sometimes people may be like, "Hey, let's do the best thing, right? Let's train and fine tune this large language model." Do you really need that? So you need to have trusted advisors, and Teradata is a trusted advisor, with your teams to get together, brainstorm the impactful use cases, and then score them and find the leading ones in terms of complexity and value, and you tackle them one by one. That's how you achieve sustainability across industries.

Rob Strechay

>> And I think one of the things about being sustainable is not moving things around all the time. And how do you really achieve... Because like you said, the data is multimodal, right? It's not just text. It's not just video, or voice, or what have you. How do you see that... How does Teradata really bring that together?

Savannah Peterson

>> Yeah. So I think what I can answer, it'll also answer your sustainability question together. So today, just today, we released this new capability about bring your own class language model. So instead of sending the data outside to make some prediction to two generative, why not just bring the model inside your platform? Right? So we provide this capability to do so. This also helps in the sustainability course because you're not moving the data around, but also it helps in different cases, for example, real time. If you want to achieve real time resolution of complaints, if you have the model within your platform, it is much, much more efficient. In addition, it's also about integrating your model with the rest of the data. So even though if I come... If I analyze a complaint in a very detailed way and I try to find what is the topic inside it, if I do not integrate it with the rest of the data, with the customer 360 degree data, it just remains a kind of insightful information, but it is not, which is operationalized, right?

Savannah Peterson

>> Right.

>> So this is the usefulness of bringing the model inside the platform so that... Model is like data. It gets integrated with the rest of the data leading to much, much more efficient operationalization, as well as sustainability.

>> Yeah. If I may add something to that,-

Savannah Peterson

>> Please....

>> my experience in financial industry, before Teradata, I was running a data science team working on multiple predictive models at the same time. And I realized that each of my team member is hitting the database, and through those complex queries, they're pulling the data, and most of those features are the same.

Savannah Peterson

>> Right.

>> But they're using it for different models. I'm like, there is something wrong here. So we have enterprise features still. Including customer complaint analyzer for any other model, if you are able to pull your features and build your analytics data set once you can then just append it with the new records or new data, rather than every time you run the model, you go to all those-

Savannah Peterson

>> Whole shabang....

>> data systems and pull them again. And imagine that we practitioners, it never works in the first time. Okay? You run a model, you realize that you forgot something. You run it again. And then you want to try a different algorithm. You also run that. And then you make a presentation to stakeholders, and they tell you, "Well, we need to change the time window. Here are new business rules. Here are some guardrails." So you have to go back and rerun everything. Imagine that how many times you are going to be hitting the data, right? So if you have a feature store that is already there, you don't need to consume a lot of resources or bend it. You will reuse it. So that's really important for our customers. And Teradata provides that.

Savannah Peterson

>> Absolutely, and making sure that you don't have to keep reinventing the wheel.

>> Yes.

Savannah Peterson

>> You've just got a great tire store. They're just waiting for you, whatever car you want to drive, and you're just ready to rock. Whether that be your adventure mobile or your Lamborghini, you're all set up, or your F1 car. We were talking about F1 earlier. I have one last question for both of you, and I'm very curious what you have to say. Pranay, I'm going to start with you first, since you're the lucky one immediately on my left.

>> Okay.

Savannah Peterson

>> You are obviously deep in this very exciting... I can sense the energy from both of you, and you're passionate about what you're working on. What do you hope to be able to say a year from now at the next Teradata Possible that you can't say yet today?

>> All right, so I would like to say that we help our customers to help their customers to resolve that complaints in an efficient way, leading value to both the end customer, as well as the enterprise. And I would like to, in the next Possible, give a tangible value to that.

Savannah Peterson

>> Yes. So actually being able to measure that too?

>> Exactly.

Savannah Peterson

>> Yeah. I love that.

>> And I think I would also like to state that... So this year, we are talking about different capabilities, but I would like to see the world evolve towards solving real problems.

Savannah Peterson

>> Yes.

>> This is what I wish for the next Possible.

Savannah Peterson

>> All right, we're with you on that one. What about you, Vedat?

Savannah Peterson

>> Well, as he mentioned, we announced bring your own large language model, right? So I know GenAI will become easier for our clients to be able to find the impactful use cases and bring them to value. So I would like to see those next year, large language model or GenAI use cases and presentations in the breakout sessions.

Savannah Peterson

>> Yes.

>> But in relation to that, as Martin Wilcox presented today, we have not still achieved the potential value of predictive and prescriptive ai yet.

Savannah Peterson

>> Right.

>> There are still a lot of work to do and the value to be gained. What I am hoping that the emphasis and benefits of GenAI as it materializes, it will also enable those failure percentages that you have heard today, 80%, 65%, or maybe even 85, to go down for predictive and prescriptive AI as well. And we have recognition of true bottom line value from clients in terms of all those use cases.

Savannah Peterson

>> I love that. Well, I cannot wait to tell those stories with you back on this stage next year. Vedat, Pranay, thank you so much for taking the time.

>> Thanks.

>> Thank you very much-

Savannah Peterson

>> Thank you....

Rob Strechay

>> for having us.

Savannah Peterson

>> This has been a blast, as always,-

Rob Strechay

>> Yes....

Savannah Peterson

>> Rob. And thank all of you for tuning in. We are midway through our full day of coverage here at Teradata Possible in Los Angeles, California. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.