Solving Data Challenges for AI In 2024

In this session, banking strategic consultant Mike Greer discusses the data challenges retail and commercial banks face in their AI strategies and execution. With siloed systems, customer data in different formats across customer profiles and business functions impedes the path forward and delays the efficiencies executives seek to achieve. At DIGITAL BANKING 2024, you will hear about solutions available today and how other organizations are addressing them to deliver on the promise of AI.


Transcription:

Mike Greer (00:11):
All right. Good Afternoon everybody. Help yourself to a seat.

(00:20):
So today we're going to talk about solving data challenges for AI. My name is Mike Greer and I'm with a company called Tealium and the pudgy fellow on the screen right now is what we call the beast. That's our mask out because we all know that data is a beast. Any of us who work at data, any of us who have to wrangle data, any of us who have to struggle with ETL, and that's why you're all in the room today. Well, in addition to getting lunch. So welcome and thank you. So this would help Mike. There we go. So we're going to talk about three themes this afternoon. This presentation is only going to be about 20 to 30 minutes, and then we're going to open up questions for Q&A. So shaping AI's learnings is all based on data, and there's a lot of talk right now in the industry about the significance between good data and bad data.

(01:18):
Its impact on AI in shaping its learnings. And then we'll go into a topic around data for AI is a double-edged sword. We'll talk about the pros and the cons or the challenges associated with that. And then we'll talk about the need for real time data supply chain in order to fuel that AI learning and to be able to respond to prompts and be able to take a look at the volume of transactions, make decisions, and have it go forward. And you need that with real time data. Anything that's lagged, anything that's going to data warehouse that needs to be pre-processed and then executed against is going to create a lag and it's going to create missed opportunities for you folks, actually all of us. So shaping the learnings is all about garbage in, garbage out. We all know this, right? Bad data leads to poor training and poor results.

(02:12):
In fact, you folks will receive a copy of this presentation and a link here is a conversation between Dell, Nvidia and ServiceNow on AI factories. And Michael Dell's quote is, bad data equals bad AI. And the majority of the conversation in this video is probably about 30, 45 minutes. It's all about the impacts that they talk about around data and its influence in AI. And really there's six categories of data that cause it to be bad, right? So you've got inaccurate data, you've got incomplete data, conflicting data, duplicate data, invalid data, and unsynchronized data. And it's that UNSYNCHRONIZED data that's the most challenging component that we're all dealing with right now as a part of the banking industry. Oh, sorry. Thanks.

(03:12):
I keep forgetting. I'm looking at my screen. So these six areas, especially the unsynchronized data, is the challenge that most systems have because they're not made to work with each other. You have front office capabilities, you have back office capabilities, you have marketing tools, and all these tools are purpose built for a uniform purpose. However, they're not made to talk to each other, and that's where the challenge comes in, right? They're in different formats, there were different contexts, they're different fields. And being able to synchronize that data across all those tools is what makes AI work. And without that synchronization, you have challenges. So according to McKinsey, bad data impacts about $15 million a year for each enterprise. And on top of that has causes a technical debt tax of 25%. That means and represents a $3.1 trillion annual cost relative to bad data. In the US alone, with 50% of all of our time being used in order to deal with that data,

(04:34):
The economic impact of gen AI will likely benefit all sectors of banking with about 56 billion for the commercial side and about 54 billion for the retail side. According to McKinsey last year, they see, and the CEOs are asking for impact on making these investments from a strategic level to see results. And if I don't see those results because you have bad data or because you're not getting the incremental growth and impact of financial performance because you have those gaps within the data, within the models, within the learnings, those gaps are why so many 85% of AI projects fail right now. It's that important to get it right. And this isn't talking about, we're not talking about creating a new master data model or anything like that, is talking about using the tools and systems you have today and being able to integrate those in a way that they can talk to each other and execute amongst each other to optimize the opportunity. That's the impact we're talking about.

(05:47):
And here's why. The importance of data quality is all about $1 of prevention protects against $10 of remediation versus a hundred dollars for a failure. The cost to wait and delay and not not working to prevent it upfront costs you a hundred times more per dollar than it does to deal with it upfront. Now, many of us have day jobs, many of us have a full workload today. We don't have the time, at least we don't think we have the time to go back and fix the things we know. So we build workarounds, we build things to say, Hey, look, we're going to just roll with it. We're going to just take it this way and make it happen, and we will deal with it another day. Well, that day is here because if you don't do this stuff now and take the opportunity to be strategic about reevaluating your workflows, reevaluating your data architecture, reevaluating how the tools talk to each other, if you don't take those steps, the impact is going to be the a hundred dollars impact and missed opportunities. Is that resonating? Cool. So let's talk about a couple examples. H&M is a retailer, clothing retailer, global, about 26 billion a year in revenue. They had irrelevant recommendations because of the AI accuracies, which led to a 5% decrease in customer click through for the marketing campaigns, and a 3% drop in online sales. That impact for one year was $16 million to their top line.

(07:42):
And increasing the marketing cost by 8% represented another $3 million. Now, let's talk about something a little bit closer to home. HSBC. Incorrect customer segmentation led to ineffective marketing efforts resulting in lost sales opportunities and increase the marketing costs to correct the segmentation errors and retarget customers appropriately. But the damage that was done from customer trust and loyalty impacted future revenue. Now that there's no disclosure on what the amount is, but you folks can make an assumption of what that would be. Inaccurate segmentation, not meeting your customers where they are, not being able to ensure that you're communicating 'em in such a way that's relevant to them, and you turn AI loose on that. It's exponential impact. That's why you have to tame the data beast now and not wait for it to grow bit more and get bigger. So the question is, what would happen to your organization if you had those type of stories? What's the impact to yourself, your role, your team? What's the trust for your executives that say, our people are doing the right thing and they're making the right impact? This is a very, very serious subject, and you folks are very serious people and that's why you're in the room right now. So data is a double-edged sword with ai. Are we clear up to this point? Am I following? Okay, cool.

(09:36):
So the sharp side, how does data fuel AI success? Enhanced fraud detection is one area. The AI algorithms can process so much transaction data and microseconds to identify parameters, patterns, and anomalies. One of our customers currently does this, not with ai, but just with their current capabilities and they're going to go down this path. Next, personalized banking using customer behavior and financial goals allow AI to tailor products and services. Let me give you a personal example for my life. I bank with a large organization. It's a global bank. I go to the ATM on a regular basis. I travel a lot. When I go to their ATM, I get the same message every single time, and it's not relative at all to what I do with them, how I bank with them, the services that I use and the stuff I might be interested in. A customer of ours in Spain has the ability to adjust that messaging based on your ATM card when you go in, takes a look at your transactions, takes a look at your balances, takes a look at products that may fit your profile and promotes that on the ATM screen.

(11:04):
That's a difference a personalization can make and it can grow 30 x with AI. Improved efficiency. This is the gem, right? This is the gold standard for what AI should be, right? It's what we're going to talk about is being able to automate those repetitive tasks so humans can do more important work. But as we talked about, if you release the beast with inefficient models, bad data, the cost can go negative. And here's the other side, pitfalls, bias and discrimination algorithms are only as good as what the data is that they are trained on. And you folks know this. And if the data contains historical biases, hallucinations, the models might persist. Issues in loan approvals, credit scoring and investment recommendations, security concerns. You all hold vast amounts of customer data, sensitive customer data. AI systems that rely on the data become potential targets to cyber attacks. And you folks have organizations that deal with this every day within your teams and data breaches. Every data breach gets a press headline. Now these days, AI can give you both the signals to prevent that, but also can be the signal to create that opacity and explainability. AI models can be black boxes, they think differently than we do. And being able to optimize on that and being able to understand that is really important. And it all comes down to being able to wield that sword responsibly.

(13:14):
Data quality and fairness need to ensure that their data sets are accurate, unbiased, and representative of your customer base in their current state. Robust security measures. You folks already know that we don't have to go into that particular topic. Explainable AI, this is where the magic happens, right? Being able to open up that black box and help people understand humans understand what's actually happening under the hood and being able to optimize that, understand it, and being able to manage it. If you don't understand it, you don't know what to change. If you don't change it, you don't know what the impact's going to be. So it all comes down to the importance of data quality. And that starts with data collection. Data collection is important for artificial intelligence because it's defined as a process gathering data from various sources. And its important importance lies in the ability to help banks to better understand their customers. Real-time data means subsecond speed. As forester reports, real-time data needs to be fast enough to keep up with the customer. And as their behavior changes in the moment, treating them like they did, treating them with the information that they did yesterday doesn't help you activate and integrate with them or satisfy their needs today.

(14:53):
And it has even more importance when it comes to fueling ai. If you don't fuel your AI initiatives with clean data, you risk diminished reliability and usability of its outputs. And real time accuracy and data collection boosts consistency, proactive engagement, and personalization at scale. And as you know, being able to treat and meet the customer in the moment, in the moment that they're in is the most important aspect of competitive advantage. Whether it's a loan application, whether it's deposit account, whether it's customer acquisition, retention or cross sell or engagement in the call center, being able to be able to activate that and real time with accurate data is the most important attribute you folks have to create a powerful relationship, a lifelong relationship with your customers. So key considerations, real-time data for AI improves the time to model and value applying data preparation, transformation, enrichment, encryption to incoming data for immediate data availability in the moment and directly send consented, organized and filtered data in real time to major AI platforms and other tools to activate across all the channels of engagement. Provide a real-time activation engine for AI insights and scores, integrating those models and platforms with the rest of your marketing tools, your customer touch points, and being able to do that in the moment that they're in. And reduce the risk of blocking any non-consent or noncompliant data from AI models.

(16:52):
That's where we come in. So for better customer experiences, the common goals to improve the CX, the challenge lies in integrating data with diverse sources to forge personalized and predictive customer experiences. Understanding not just where they're at, but where they're going to go next or where you want to guide them in that journey to the next step. And at the root centralized data architecture of real-time data processing can aid in making data compliant, secure and primed for successful AI outcomes. This is where the magic happens. It's at the data, at the collection source, at the enrichment, at the consented, and the filtered components in order to be able to do that. So key considerations. Oops, hold on. The best in class banks are 30% more likely to leverage technology such as AI and automation, especially across CRM, billing, loyalty and payments, and establish a single view of the customer's data, 95% versus 73%. Those firms are also more likely to use automation to update and integrate customer feedback across all touch points. 89% of the firms versus 58. This is a real time data supply chain,

(18:26):
And this is what we do. So from left to right, collect the data, be able to standardize that data in real time at the ingestion point before it feeds into the rest of your stack. And then being able to connect that data, not just through your data warehouse, not just through your AI models, but also across all your data enrichment in order to make sure that the data is synchronized and talks in the language of the tools that are going to receive that downstream. And then being able to orchestrate that data to make sure it goes where it needs to go and where it doesn't go, where it shouldn't go. To protect privacy and compliance. And then activating that across all the channels, call center, billing, marketing, CRM, and having a unified view of the customer in their state, in their capabilities, in their desires in the moment. So here's our commercial. I had to throw it a commercial. You knew that, right? So we have a suite of solutions to close the innovation gap, sorry, by fueling models with privacy compliant customer data for enhanced experiences. You'll receive a copy of this deck and you'll be able to click on these links to see the reference materials as we go to our website for more information. And my colleagues and I are here to answer questions. Can't see the clock where we at on time, gang?

(20:19):
Sorry. 24. Okay, so here's our learnings for today. Learning is best based on data quality. Data is the foundation for AI, and AI requires a real-time data supply chain. I'm Mike Greer. I'm here for questions and we're going to take some questions now. Do we have a microphone to pass around? Does anybody have questions? Thank you. Is the food that good? Good. Well, I hope you enjoyed the presentation. My colleagues in the back are Adam Abowitz, he's a regional vice president for the southeast, Nina Nino. I'm so sorry, man, is our SVP of Global Solutions and Strategic Consulting, and then I'm the banking consulting strategist. Happy to answer any questions.

(21:33):
No. Yes.

Audience Member 1 (21:36):
We just ask, do you integrate with the bank core systems because all the data selected in core, right?

Mike Greer (21:42):
Correct. Yeah. So the question is do we integrate with core systems? And the answer is, yeah, we can do that. Some of 'em are native and some of them we have to actually build connectors for other questions.

Audience Member 2 (21:59):
The quality of data starts data collection. How do you think about how to equip the people who are actually collecting the data or at the front end of that process to get it right in the first time?

Mike Greer (22:14):
That's a great question. So the question is, how do you work with your forward looking staff to actually input the data correctly the first time? Did I rephrase that right? Yeah. Okay. So there's always the opportunity for human error, but being able to take a look at the data that's inputted and qualify it against some standards before it feeds into the model is a way to correct against that. But no system is a hundred percent, as we all know. That's not the perfect answer, but I'm an imperfect individual, so

Audience Member 2 (22:50):
That's the right answer though.

Mike Greer (22:54):
Other questions? Well, thank you very much for your time. I hope you enjoyed the luncheon. I hope you found the information valuable and we're around and if not, you know how to get ahold of us. Thank you so much.