# Initiate Your AI-Powered Banking Service Journey Auto-transcribed by https://aliceapp.ai on Thursday, 19 Sep 2024. Synced media and text playback available on this page: https://aliceapp.ai/recordings/phkzVjfKc3EHDNW08Ye6Tyv_um5GnqOB * Words : 3,601 * Duration : 00:19:31 * Recorded on : Unknown date * Uploaded on : 2024-09-19 17:32:13 UTC * At : Unknown location * Using : Uploaded to aliceapp.ai ## Speakers: * Speaker A - 21.66% * Speaker B - 17.41% * Speaker C - 60.93% ---------------------------- Speaker A [00:00:00] Everybody, welcome. Welcome to Dreamforce. Those of you in the room, those of you watching us online, um, on Salesforce plus, thank you so much for joining us. We have a great session for you today, um, on how to initiate your AI powered banking journey. So before we get into the session, just a reminder that Salesforce is a publicly traded company. We do encourage customers to base their purchasing decisions off of products and services that are currently available. Thank you for joining us. We know that you have a lot of options here at Dreamforce and, um, we're so grateful that you decided to spend your time here with us. So who are we? I'm Kelly Fakis. I'm a product marketing manager here at Salesforce supporting financial services cloud. I'm joined today by Anshul Kanodia, principal solution engineer, who's been working very closely with Brian Tucker, um, director of CRM solution and design at CIBca. CIBC has been doing some really great work with generative AI and their complaints process and we're very excited to have Ryan here, um, to walk us through a demo that he actually created to obtain internal stakeholder alignment and move generative AI forward for CIBC. So let's address maybe some concerns that may be standing in your way of initiating your AI powered journey. We hear quite often from customers that there are still some security concerns, data security concerns, cybersecurity concerns, um, around implementing generative AI into their banking processes. Another one that we hear quite often is that regulators don't really know how to look at generative AI, which makes compliance teams feel very uneasy about implementing generative AI into their business processes and procedures. With that said, banks understand that they do need to leverage this technology. Now customers are requesting it, um, and to stay ahead of competitors. But the question still remains as to how. How do we initiate this journey? And we have some great news. We've broken it down into some simple steps that you can take and Ryan will validate these steps as these are very similar to what CIBC, um, took in their AI journey. The first one is educating yourself on the truths about aih. There's a lot of noise in market, there's a lot of conflicting messaging. It's really about zeroing in on what matters most to your bank and what matters most to your stakeholders and your leaders. Working with our team of AI experts to identify what really is the truth. And what does Salesforce have to offer to debunk some of the myths that may come up from your teams? The second one is showing identifying a use case that has small jobs where AI can come in and assist. CIBC chose to start with complaints. Complaints is a great process that is made up of a lot of small jobs where AI can come in and drive big value quickly. And then it's about showing versus telling. Working with somebody like Ancho, one of our solution engineers, who can help you develop a demo that you can show to your team versus just telling them so that they can see the tangible benefits that AI can drive today. And then, believe it or not, after that, you're in the transform pillar. You're looking for new use cases, you're looking to expand generative AI into other business processes. And the beauty of working with Salesforce is we're always innovating. So when you partner with us, we're innovating together. We always have a vision for the future of financial services, and we can do that together. So without further ado, I'm going to stop telling you, and I'm going to pass it over to Anshul and Ryan to show you what this actually looks like live, um, via our customer's voice. Speaker B [00:03:50] Thanks, Kelly. Hi, everyone. I am so excited to share the stage with Ryan. Ryan is one of my favorite people. He's a true trailblazer, and he's been spearheading all things AI at Cibc. All right, so let's talk complains. Ryan, walk us through what the complaints process looked like, and why did you use an AI solution to improve it. Speaker C [00:04:16] Uh, okay, so this is for our service, cloud case management complaints resolution process. Uh, and as part of that, uh, it was very manual, it was very tedious in order to get to the point where we were able to produce a, a written communication to go out to the client, which would summarize the activities that took place leading up to that resolution, uh, as well as all of the requirements that were required by our regulatory authority to provide. In that letter, there was a very manual and tedious process that involved multiple teams. It would start with the case agent who would be going through those notes. It would, uh, then produce a statement of facts that was based on information spread out across the platform in various different places, in notes, in fields, or perhaps even in connected systems, uh, further out in the enterprise. Uh, that would then have to go for approval, and then that would go to a communications team, because once we've got the facts, we would have to put that into something that's written that we could actually send out to our clients. M doing that work, uh, was with about 60,000 resolutions a year. Uh, it was hundreds of thousands of hours that we would spend just in writing a letter for a process that effectively was already complete. Speaker B [00:05:30] Wow, did you hear that? That's a lot of manual work. And I am so glad that you decided to use AI to help you with that. Well, we understand that you have choices for vendors when you choose AI solutions. So why don't you go with AI and Salesforce as your partner of choice? Speaker C [00:05:49] Uh, so selecting Salesforce was uh, uh, something that took quite a lot of time and process. I was involved in the assessment of various different solutions. Uh, one of the reasons that we landed and decided to go with the salesforce option is number, um, one, the focus on trust. And so we were able to demonstrate to our stakeholders that there's control and that there's governance through the trust layer. That's with the auditing of all requests and that's creating control and governance throughout that entire process from development through to consumption. So, uh, that's a big part of it. Um, but uh, I would say that the other reasons for selecting Salesforce would uh, have to do with the way in which it integrates with the platform. So we have existing workflows and when we looked at other solutions, they didn't quite plug into what we were doing already. They would sit to the side of that. Uh, so having Einstein sort of sit on the surface of exactly where our agents are doing their work on a day to day basis was pretty compelling in terms of its value. Uh, and then the modularity as well. Uh, so you build once and then you don't have to always build it over and over and over again. Uh, when you create a uh, prompt you can reuse it for different contexts. Uh, so being able to scale the solution to future use cases and uh, as well to adapt what we build into the future was uh, something that really made the salesforce, Einstein solution pull ahead of all of the others. Speaker B [00:07:17] That is so important. We don't want you to just build something and use it just for one use case. We want you to be able to use it over and over again. Well, I think we've talked a lot. It's time to show some demo. Can we show what Ryan's built and over to you? Speaker C [00:07:36] Yeah. So what we're looking at right now is we're inside service cloud, we're inside of a ah, case that has already been resolved. And so there is a history, there is notes, there's interactions that are all buried behind the surface. And our agent is about to go through the process of confirming that they are ready to engage with the AI on a regulated process, uh, and then, uh, ask the AI to produce a statement of fact. That statement of facts is being generated without very much effort. Uh, Einstein's been plugged into exactly what we're doing, and so it has the context, it knows what work it needs to perform. I don't have to tell Einstein to produce the statement of facts. It just gets generated for me with a single click. Speaker B [00:08:18] With a single click. Did you see how quick that was? Did you see how. And did I just see you actually give that thumbs up while you were in the process? Speaker C [00:08:31] Right. So the feedback mechanism is also really important to one of the reasons we selected Einstein. Uh, so, um, the human in the loop is absolutely required in order for us to show the diligence in how we're using AI across the platform. And so we're collecting that feedback with each and every single interaction that happens. So you click on that thumbs up, or you click on that thumbs down and you provide a rationale, folks who are then looking at the AI audit log and making sure that there's no toxicity in the responses and that our end users are satisfied with the output, which, uh, considering how many people are involved in this process, you need to have that kind of logging so that you can understand when it's failing or when it's not producing the level of quality that you're looking to maintain, uh, so that you can continue to refine and continue to ensure that your solutions are working the way that they're intended to work. Speaker B [00:09:24] And all of this is grounded in customer data. Speaker C [00:09:28] Right? So wait, wait, wait. Speaker B [00:09:30] I don't believe you. I really don't. I think this is all magic. I think the audience definitely thinks it's magic. So let's go behind the scenes and show them how it's done. Speaker C [00:09:38] Okay? It's not magic, I promise. So this is, uh, the prompt builder console, um, uh, which was released, uh, by Salesforce, I think, nearly a year ago. And what you get is you get basically a chat GPT, a big old text box, and you can type in your instructions. As a developer, this is your only user interface. You're looking at it. Uh, once you've provided the instructions to the AI, you need to ground it with data. So we've got these blue hyperlinks on the side that's going to bring the data into the prompt. And then when I test it, you can see here that I get to see how that resolves. The resolution gives me an example of what instructions are going to end up going to the AI while I'm going through that development process so I don't have to deploy it into production to get a sample of how it's going to actually behave. And then the response that comes from the LLM is also integrated into the development environment. Um, so yeah, it's a really simple solution. Uh, it's almost like working with Jack itself. Speaker B [00:10:37] Wow, that is really simple. And did I see you just enter a related record? So that means you can test out different use cases, tweak this on the go, and still not have to go and write code behind any of this. Speaker C [00:10:49] You can keep going back and forth. So this is my solution. I developed this and uh, I probably spent a good day just clicking on preview over and over and over and over again. That's what a good developer and that's what a good prompt engineer is going to do. They're going to test the output against multiple sources and they're going to continue to refine those prompts based on their prompt strategies. And bringing that data in, that is. Speaker B [00:11:13] So much better than actually implementing it, finding it out while testing it later and so on and so forth. But wait, did I hear you say that you brought data from outside of Salesforce into this as well? Speaker C [00:11:24] Right. Uh, so that's grounding to the source of truth. Um, instructions to the AI are meaningless without context and so I can't just put in a prompt and expect it to behave the way I want it to. I need to bring in information that's contextual and that's real time. So that would be data cloud. And so data cloud is able to bring in information that exists across, um, all the Salesforce orgs. Uh, so that we have an idea of the client profile that's not only within our, but across the enterprise as well as into some uh, other systems that aren't on the Salesforce platform. That grounding to the source of truth is what's going to enable the AI to provide more relevant information or more relevant summaries in what it's a producing. Speaker B [00:12:05] That's a great point because at the end of the day, without the right data, it's just a prompt. All right, but this isn't the end of the process. I remember you saying that there's a letter and an email that gets created. Are you using AI for that as well? Speaker C [00:12:19] That's right. So we've got the facts. Now that we have the facts, we actually have to produce a written communication. I'm sure everyone's aware that that's where AI excels and so this is a great use case and it's one of the reasons why we selected it. Uh, and so here I've got AI output feeding into AI. It's a multistage process. And again, Einstein has context. It knows where I am in this flow, it knows what I need. And I just had to click a button and it automatically took the statement of facts that I produced earlier and it pulled it in to some new instructions that were designed to write a letter. And within a couple clicks I've actually just shipped it out to the client. You didn't see the approval process, but a standard salesforce approval process would have been embedded in this flow so that we could get approval from our manager to send out those communications. Speaker B [00:13:08] Wow, that was still a really quick end to end process. How long did it take for you to build a solution? Speaker C [00:13:14] Um, so this is sort of an interesting story in the power of partnership. Uh, and so she invited me to the Toronto office really early in the Einstein days to take a look at prompt builder. And, uh, I got a demo environment that Salesforce set up for me. And I think it was a Thursday or a Friday and I was so excited. I knew about the SWR or I knew about the case resolution process already and I took it home over the weekend. I had a glass of wine in one hand and I had a mouse in the other and I banged it out by Monday. Uh, I'm not kidding. Within three days I was shopping the, uh, demo to all of my stakeholders internally to show what we could actually develop here. Speaker B [00:13:57] Did you guys hear that? He built this over the weekend. So if any of you are looking for something to do over the weekend. Speaker C [00:14:03] Create a prompt, you can do it, I promise you. Speaker B [00:14:07] All right, this looks great, but can you talk a little bit about the ROI and what CIBC is going to be achieving from that perspective? Speaker C [00:14:14] Sure. And of course, this is critical in both selecting the use case, uh, as well as in selecting Einstein. We need to be able to show benefits, we need to be able to show a return on investment because there's a lot of work that's involved in getting the alignments that we require. Uh, and so we're measuring, uh, our benefits and our return on investment, uh, in efficiencies to productivity, the gains that we're going to get in the hours that are reduced in time spent. Again, this is a process that's effectively been completed already. We've resolved the case. We are just required by our regulatory authority to send out that written communication. And so those productivity gains, uh, with hundreds of thousands of hours that are being spent annually in writing this letter are pretty important to us. Speaker B [00:15:00] That is amazing. But I'm sure you ran into some challenges deploying generative AI. We understand its new technology. So what did you do to overcome those challenges? Speaker C [00:15:10] Right. Uh, so there's always some challenges when you're working with a large enterprise. Um, and I would say that the demo was a key part of getting that alignment, uh, and being able to show my stakeholders exactly what we were going to get at the end of it. Um, but I think the biggest challenge that we had, uh, was in getting that alignment, uh, with the enterprise on architecture. And so, uh, in order to get that alignment, we need to be able to show a few different things. So the focus on trust, the trust layer is key there, uh, because that's going to demonstrate that we've got control, that we've got the ability to maintain our standards of governance within the organization using this software. Um, but I think that a couple other key components that were really important in getting alignment, uh, were the integration and the capability of using Einstein in a very flexible way. So the trust layer has something called the LLM gateway. And with the LLM gateway you can, uh, bring your own LLM. And if you don't know what LLM large language model, that's what's behind your chat. GPT user interface, it's your backend, it's your brain. And so for us, we've already got some architecture that we've built where our LLM is being governed by a central, uh, organization within the enterprise. And we were required to connect to that so that the enterprise could have control of AI workflows, not just with our application, but with all of the applications that we have at the bank. So those allowed us to connect into our existing, uh, infrastructure, uh, while still getting all of the benefits of speed and scale and plugging into our existing workflows and processes. Speaker B [00:16:53] That's amazing. And I think I, obviously he complains is a great use case, but you're just scratching the surface with that one. Right. So what's next? What is CIBC going to be doing next from a genai perspective? Speaker C [00:17:04] Uh, so I'm really excited about what's next. Um, so precisely what's next is we're going to be expanding from service cloud into sales cloud, uh, for meeting, uh, prep summary. And what meeting prep summary is going to do is it's going to do all of that hard work that our advisors would be doing in going through client interactions and notes and history and getting ready for those conversations that they're going to have. But most importantly, it's going to be based on context. It's going to know what the meeting is about and it's going to be able to pull the right information so that our advisors don't have to spend all of their time digging through records, digging through notes, trying to understand their clients a little bit better. And so for CIBC, um, uh, our statement is making our clients ambitions a reality. And so in order to do that, we need to be able to focus on the relationship. And by taking away all of that time and effort in reinterpreting that history, we're able to just give everything to our advisors, uh, via Genai, so that they know exactly how to take that conversation forward. Speaker B [00:18:08] That's amazing. And not just from a frontline or advisor perspective, but even from a customer perspective. I cannot even tell you how many times I've been on a call where I've been asked the same question over and over again that they should have known about me. So this is a great way to increase those NP's scores and deepen those client relationships. I know that's all the time that we have today. So thank you guys so much for joining us. And Ryan, thank you so much for being with us. Kelly, I'll take it, uh, give it back to you. Speaker A [00:18:35] Yes. Awesome. Thank you all for joining us both in the room and online. Just in summary, Ryan did a great job of showing us how CIBC has gone from AI initiation to true transformation. Taking those three simple steps, starting working with Anshul, educating himself on the truths about AI, and then really leaning into that show pillar, which is creating a demo and ensuring that you are able to show your stakeholders the true benefit of AI. Um, so thank you again. Speaker C [00:19:04] Thank you. Speaker A [00:19:11] For those of you in the room, please give us your feedback. We'd love to hear from you. Help us help you. We'll help you get some caffeine. Um, the first 4000 attendees get a five dollar Starbucks gift card. So please fill out the survey. Um, we'd love to hear from you. Speaker C [00:19:25] Get your Starbucks. Speaker B [00:19:26] Give us a thumbs up. Speaker C [00:19:30] Your agent force learning.