WEBVTT 00:00:00.000 --> 00:00:03.439 Welcome to this deep dive. We are really glad 00:00:03.439 --> 00:00:05.059 you're here with us today. Yeah, definitely great 00:00:05.059 --> 00:00:08.369 to have you listening If you are looking to just 00:00:08.369 --> 00:00:11.710 bypass all that dense corporate jargon and get 00:00:11.710 --> 00:00:14.330 straight to the actual mechanics of where technology 00:00:14.330 --> 00:00:17.329 is taking high stakes finance, you are in the 00:00:17.329 --> 00:00:19.710 exact right place. It is a totally fascinating 00:00:19.710 --> 00:00:21.910 landscape to explore right now. I mean, we are 00:00:21.910 --> 00:00:24.250 moving way past the theoretical stuff. Right. 00:00:24.250 --> 00:00:26.429 And we're looking at the actual plumbing of how 00:00:26.429 --> 00:00:28.910 these major institutions operate on a daily basis. 00:00:29.089 --> 00:00:31.750 Exactly. So our mission today is to explore how 00:00:31.750 --> 00:00:35.049 artificial intelligence is just completely rewiring 00:00:35.240 --> 00:00:37.659 set management. Down to the studs. Down to the 00:00:37.659 --> 00:00:40.719 studs. We're pulling insights from a really comprehensive 00:00:40.719 --> 00:00:43.539 panel discussion today. It features digital finance 00:00:43.539 --> 00:00:46.240 and security leaders from MetLife, Capgemini 00:00:46.240 --> 00:00:50.179 Invent, PNC Bank, and PGM Global Services. Heavy 00:00:50.179 --> 00:00:53.140 hitters. Yeah, very. And the sheer scale of what 00:00:53.140 --> 00:00:55.380 they are discussing is honestly, it's staggering. 00:00:55.679 --> 00:00:58.219 We are not talking about using a chat bot to 00:00:58.219 --> 00:01:00.829 summarize an article for you. No. We are looking 00:01:00.829 --> 00:01:03.490 at a fundamental shift in how trillions of dollars 00:01:03.490 --> 00:01:06.489 in global wealth are managed, secured, and optimized. 00:01:06.849 --> 00:01:09.969 And that scale, honestly, that is precisely why 00:01:09.969 --> 00:01:12.049 this matters to you, even if you don't manage 00:01:12.049 --> 00:01:14.689 a massive institutional portfolio. Right. Even 00:01:14.689 --> 00:01:17.609 if you're just interested in how enterprise tech 00:01:17.609 --> 00:01:20.370 is evolving. Exactly. What we are really examining 00:01:20.370 --> 00:01:24.030 today is the ultimate modern tightrope walk for 00:01:24.030 --> 00:01:27.010 these massive firms. They're trying to achieve 00:01:27.010 --> 00:01:31.310 this this hyper -efficiency through AI. Which 00:01:31.310 --> 00:01:33.250 everyone wants. Which everyone wants, but while 00:01:33.250 --> 00:01:36.310 maintaining the absolute non -negotiable necessity 00:01:36.310 --> 00:01:38.689 of security. Yeah. One of the panelists from 00:01:38.689 --> 00:01:40.609 PNC Bank framed this perfectly, I think. They 00:01:40.609 --> 00:01:43.170 said the guiding principle for their AI adoption 00:01:43.170 --> 00:01:46.310 is to have foresight, not fear. Foresight, not 00:01:46.310 --> 00:01:48.150 fear. I love that. I think that's a great anchor 00:01:48.150 --> 00:01:50.549 for our discussion, especially when it is just 00:01:50.549 --> 00:01:53.450 so easy to either panic about security or get 00:01:53.450 --> 00:01:55.469 completely swept away by the vendor hype. Oh, 00:01:55.469 --> 00:01:57.980 the hype is everywhere. It is. OK. Let's unpack 00:01:57.980 --> 00:02:01.219 this, starting with the actual productivity leap 00:02:01.219 --> 00:02:04.159 we are seeing on the ground. The panel sources 00:02:04.159 --> 00:02:07.260 cited that generative AI is currently being used 00:02:07.260 --> 00:02:11.099 to reduce manual processes by as much as 80%. 00:02:11.099 --> 00:02:14.099 80%. Yeah, and now, I have to admit, whenever 00:02:14.099 --> 00:02:17.319 I hear a perfectly round number like 80%, I get 00:02:17.319 --> 00:02:19.879 a little skeptical. Sure. It sounds like marketing 00:02:19.879 --> 00:02:22.360 math. What does that actually look like in the 00:02:22.360 --> 00:02:24.340 trenches? That's a healthy skepticism to have. 00:02:24.740 --> 00:02:26.840 But when you look at the specific use cases they 00:02:26.840 --> 00:02:29.520 highlighted, that number starts to make a lot 00:02:29.520 --> 00:02:32.520 of sense. OK. Take the traditional KYC process, 00:02:32.560 --> 00:02:35.560 for example. Know your customer. Right. We all 00:02:35.560 --> 00:02:38.419 know how notoriously bottlenecked that can get. 00:02:39.080 --> 00:02:41.020 Historically, you have analysts just bogged down 00:02:41.020 --> 00:02:43.659 in the mechanics of verification. Pulling disparate 00:02:43.659 --> 00:02:46.159 records. Right. Cross -referencing complex corporate 00:02:46.159 --> 00:02:49.819 hierarchies. and digging through massive unstructured 00:02:49.819 --> 00:02:53.000 PDF documents? just to validate an entity. Oh, 00:02:53.120 --> 00:02:56.340 the dreaded unstructured PDF. Anyone who has 00:02:56.340 --> 00:02:59.379 ever had to manually pull specific data points 00:02:59.379 --> 00:03:02.699 out of a messy 200 -page legal document and put 00:03:02.699 --> 00:03:05.580 it into a spreadsheet knows the sheer physical 00:03:05.580 --> 00:03:08.060 eye strain involved. You're just living with 00:03:08.060 --> 00:03:11.199 dual monitors, using keyboard shortcuts, and 00:03:11.199 --> 00:03:13.979 praying you didn't misread a decimal point. Exactly. 00:03:14.139 --> 00:03:17.379 And what the AI is doing is sweeping that mechanical 00:03:17.379 --> 00:03:20.750 friction away. It effortlessly parses those unstructured 00:03:20.750 --> 00:03:23.969 documents, pulls the relevant entities out, and 00:03:23.969 --> 00:03:26.650 formats the data for the human analysts to review. 00:03:26.689 --> 00:03:29.830 Wow. But here is the critical pivot in the technology. 00:03:30.669 --> 00:03:34.259 We are evolving from simple AI into what the 00:03:34.259 --> 00:03:37.159 industry is calling agentic AI. Agentic AI. Okay, 00:03:37.180 --> 00:03:39.219 that feels like a massive turning point. Break 00:03:39.219 --> 00:03:41.360 down agentic AI for us because that is a term 00:03:41.360 --> 00:03:43.319 a lot of people are hearing but might not fully 00:03:43.319 --> 00:03:45.659 grasp in a corporate context. Sure. Up until 00:03:45.659 --> 00:03:48.060 recently, enterprise AI has mostly been a prompt 00:03:48.060 --> 00:03:50.759 and response tool. You ask a system a question, 00:03:51.020 --> 00:03:53.740 it generates text or code in return. Like a really 00:03:53.740 --> 00:03:58.300 smart search engine. Exactly. Agentic AI is fundamentally 00:03:58.300 --> 00:04:01.930 different. It acts as a proactive assistant that 00:04:01.930 --> 00:04:05.389 has actually granted permission to manage end 00:04:05.389 --> 00:04:07.689 -to -end processes across different software 00:04:07.689 --> 00:04:10.250 platforms. Oh, wow. Yeah. It isn't just waiting 00:04:10.250 --> 00:04:13.669 for a prompt. It is anticipating the next logical 00:04:13.669 --> 00:04:17.060 step. making API calls, and queuing up actions 00:04:17.060 --> 00:04:19.720 for a human to approve. So it shifts the user's 00:04:19.720 --> 00:04:22.740 role from being a fast data gatherer into an 00:04:22.740 --> 00:04:25.279 actual strategic decision maker. Spot on. That 00:04:25.279 --> 00:04:27.699 makes a lot of sense. To make this tangible for 00:04:27.699 --> 00:04:29.620 you, the listener, let's look at a case study 00:04:29.620 --> 00:04:31.920 the Capgemini panelists shared. They applied 00:04:31.920 --> 00:04:34.519 this to the client contact center. A great example. 00:04:34.639 --> 00:04:36.759 Yeah. And instead of a generic retail banking 00:04:36.759 --> 00:04:39.199 example, let's put this in the context of high 00:04:39.199 --> 00:04:42.500 stakes asset management. Imagine you are an institutional 00:04:42.500 --> 00:04:44.980 investor, the market has just taken a massive 00:04:44.980 --> 00:04:47.879 unexpected dip, and you are calling your asset 00:04:47.879 --> 00:04:50.759 manager to figure out your exposure. Right. And 00:04:50.759 --> 00:04:52.740 in a high stress scenario like that, the last 00:04:52.740 --> 00:04:55.000 thing the client wants is to navigate an automated 00:04:55.000 --> 00:04:57.480 phone tree. Press one for panic. Exactly. Or 00:04:57.480 --> 00:04:59.439 sit on hold while an agent frantically tries 00:04:59.439 --> 00:05:02.240 to pull up their specific portfolio. Right. So 00:05:02.240 --> 00:05:05.360 with agentic AI, the system actually listens 00:05:05.360 --> 00:05:08.120 to the voice of the caller. It recognizes the 00:05:08.120 --> 00:05:11.170 client. analyzes the urgency in their voice or 00:05:11.170 --> 00:05:14.230 the keywords they are using, and instantly routes 00:05:14.230 --> 00:05:16.810 them to the appropriate senior desk. Seamlessly. 00:05:16.930 --> 00:05:19.689 Seamlessly. But it's what happens on the agent's 00:05:19.689 --> 00:05:22.629 side that is truly impressive. Before the human 00:05:22.629 --> 00:05:26.029 agent even answers the line, The agentic AI has 00:05:26.029 --> 00:05:29.250 generated a comprehensive pre -call dashboard. 00:05:29.470 --> 00:05:31.230 It connects all the dots before the conversation 00:05:31.230 --> 00:05:34.449 even begins. Exactly. The agent instantly sees 00:05:34.449 --> 00:05:36.850 the client's total exposure to the current market 00:05:36.850 --> 00:05:39.930 volatility, their recent trade history, and even 00:05:39.930 --> 00:05:42.389 automated suggestions for mitigation strategies 00:05:42.389 --> 00:05:44.970 based on the firm's current macroeconomic stance. 00:05:45.089 --> 00:05:48.029 That's huge. The AI has already done the 20 minutes 00:05:48.029 --> 00:05:50.189 of frantic background research and the two seconds 00:05:50.189 --> 00:05:52.430 the phone was ringing. And we shouldn't overlook 00:05:52.430 --> 00:05:54.149 what happens after the client hangs up either. 00:05:54.240 --> 00:05:57.139 Right. The agentic AI performs automated proactive 00:05:57.139 --> 00:05:59.920 post call follow ups. It drafts the summary of 00:05:59.920 --> 00:06:02.399 the agreed upon actions. It queues up the necessary 00:06:02.399 --> 00:06:04.939 trade tickets for the agent to approve and address 00:06:04.939 --> 00:06:07.100 an email to the client outlining the next steps. 00:06:07.420 --> 00:06:10.160 It's just a seamless end to end orchestration. 00:06:10.240 --> 00:06:13.000 Completely. That automated follow up is incredible 00:06:13.000 --> 00:06:16.279 for the client experience. But I'm actually more 00:06:16.279 --> 00:06:18.819 interested in what this means for the person 00:06:18.819 --> 00:06:21.459 sitting on the other side of the desk. How is 00:06:21.459 --> 00:06:24.300 this changing the internal worker's day -to -day 00:06:24.300 --> 00:06:26.699 reality? That's the real question. Because the 00:06:26.699 --> 00:06:29.060 goal here shouldn't just be replacing people. 00:06:29.379 --> 00:06:31.860 It should be giving them better tools. And one 00:06:31.860 --> 00:06:34.399 of those tools the panel discussed is the ability 00:06:34.399 --> 00:06:36.860 to essentially chat with your own proprietary 00:06:36.860 --> 00:06:40.839 data. That capability alone. represents a massive 00:06:40.839 --> 00:06:43.959 paradigm shift in corporate environments. Think 00:06:43.959 --> 00:06:46.399 about the traditional latency involved in getting 00:06:46.399 --> 00:06:48.620 an answer to a complex business question. It 00:06:48.620 --> 00:06:51.660 takes forever. Right. If a portfolio manager 00:06:51.660 --> 00:06:54.519 needed to understand a subtle trend buried in 00:06:54.519 --> 00:06:57.279 three years of transaction data, they couldn't 00:06:57.279 --> 00:06:59.720 just find that themselves. They'd have to submit 00:06:59.720 --> 00:07:02.540 a JIRA ticket to a data engineering team. Oh, 00:07:02.540 --> 00:07:05.399 yeah. Wait a week? and hope the SQL query the 00:07:05.399 --> 00:07:07.759 engineer wrote actually captured the nuance of 00:07:07.759 --> 00:07:10.100 the original request. And half the time it doesn't, 00:07:10.120 --> 00:07:11.860 and you have to start the process all over again. 00:07:12.000 --> 00:07:14.759 But now, that same portfolio manager can use 00:07:14.759 --> 00:07:17.899 natural language to directly ask their data a 00:07:17.899 --> 00:07:21.399 question. Under the hood, the AI translates that 00:07:21.399 --> 00:07:24.120 plain English into a complex query, runs it against 00:07:24.120 --> 00:07:26.920 the database, and instantly generates a custom 00:07:26.920 --> 00:07:29.879 bashboard on the fly. You are completely removing 00:07:29.879 --> 00:07:32.920 the latency between having a question and getting 00:07:32.920 --> 00:07:35.470 an actionable insight. Yeah. And we are seeing 00:07:35.470 --> 00:07:37.810 this acceleration outside of the data science 00:07:37.810 --> 00:07:40.110 realm too. The panel brought up a great example 00:07:40.110 --> 00:07:42.370 regarding sales teams and wholesalers out in 00:07:42.370 --> 00:07:44.509 the field. Yes, the wholesaler example was brilliant. 00:07:44.610 --> 00:07:46.550 If you've ever worked in that side of the industry, 00:07:47.050 --> 00:07:49.310 you know the pain of competitor research. It's 00:07:49.310 --> 00:07:52.110 brutal. You're traveling, meeting with institutional 00:07:52.110 --> 00:07:54.470 clients, and you have to be prepared to explain 00:07:54.470 --> 00:07:57.290 exactly why your firm's product is better than 00:07:57.290 --> 00:08:00.389 a rival's. Historically, that meant digging through 00:08:00.389 --> 00:08:03.269 hundreds, sometimes thousands of pages of incredibly 00:08:03.269 --> 00:08:06.730 dense compa - I always joke that this specific 00:08:06.730 --> 00:08:09.529 kind of tedious manual research used to be a 00:08:09.529 --> 00:08:12.550 rite of passage for junior analysts, fueled entirely 00:08:12.550 --> 00:08:15.290 by bad office coffee and sheer willpower. It 00:08:15.290 --> 00:08:18.829 is incredibly grueling work. You are looking 00:08:18.829 --> 00:08:21.949 for very specific details, subtle differences 00:08:21.949 --> 00:08:25.540 in risk profiles, fee structures, or investment 00:08:25.540 --> 00:08:28.439 restrictions, all buried deep within legal text. 00:08:28.560 --> 00:08:31.379 But armed with this new AI, those teams in the 00:08:31.379 --> 00:08:33.679 field can just query the system directly from 00:08:33.679 --> 00:08:36.279 their phone or tablet. They can ask a highly 00:08:36.279 --> 00:08:39.100 specific complex question about a competitor's 00:08:39.100 --> 00:08:42.639 newly updated policy, and the AI synthesizes 00:08:42.639 --> 00:08:45.159 the answer for them right there, in real time, 00:08:45.279 --> 00:08:48.200 citing the exact page of the prospectus it pulled 00:08:48.200 --> 00:08:50.620 from. It gives them the answer when they actually 00:08:50.620 --> 00:08:52.620 need it, right before they walk into the client 00:08:52.620 --> 00:08:54.980 meeting. And it isn't just data retrieval, it 00:08:54.980 --> 00:08:58.009 is also content creation. The panelists from 00:08:58.009 --> 00:09:00.549 PGM Global Services shared how their marketing 00:09:00.549 --> 00:09:03.110 and comms teams are utilizing this. They're taking 00:09:03.110 --> 00:09:05.429 their past thought leadership materials, white 00:09:05.429 --> 00:09:08.330 papers, event copy, and feeding it into an AI 00:09:08.330 --> 00:09:11.450 so the model learns their highly specific proprietary 00:09:11.450 --> 00:09:13.789 brand voice. They mentioned it took their standard 00:09:13.789 --> 00:09:16.490 campaign creation time from 14 days down to six 00:09:16.490 --> 00:09:18.570 days. Which is a massive jump in efficiency. 00:09:18.830 --> 00:09:21.429 Massive. But my immediate question when I read 00:09:21.429 --> 00:09:23.529 that was, does it actually capture the nuance 00:09:23.529 --> 00:09:26.269 of a premier financial firm, or does it sound 00:09:26.269 --> 00:09:29.710 like a generic robot wrote it? That is the crucial 00:09:29.710 --> 00:09:33.289 distinction. Because they are training it specifically 00:09:33.289 --> 00:09:37.190 on their own curated historical data, the output 00:09:37.190 --> 00:09:40.190 isn't generic. It incorporates their specific 00:09:40.190 --> 00:09:43.370 terminology, their required compliance disclaimers, 00:09:43.370 --> 00:09:46.620 and their house style. OK. The human editors 00:09:46.620 --> 00:09:49.460 are then spending their time refining a highly 00:09:49.460 --> 00:09:52.200 competent first draft rather than staring at 00:09:52.200 --> 00:09:54.960 a blank page. What's fascinating here is how 00:09:54.960 --> 00:09:56.960 this completely reframes the mainstream narrative 00:09:56.960 --> 00:10:00.100 we often hear about AI coming to eliminate everyone's 00:10:00.100 --> 00:10:02.799 jobs. When you look at these enterprise use cases, 00:10:02.919 --> 00:10:04.700 it isn't about human replacement. It is about 00:10:04.700 --> 00:10:08.120 human augmentation. It removed the absolute drudgery 00:10:08.120 --> 00:10:10.679 of the day -to -day workflow. It allows people 00:10:10.679 --> 00:10:12.960 to actually perform the strategic, creative, 00:10:13.240 --> 00:10:14.740 and interpersonal relationship building they 00:10:14.740 --> 00:10:17.159 were originally hired to do. But of course, to 00:10:17.159 --> 00:10:19.120 do any of this effectively, whether we're talking 00:10:19.120 --> 00:10:21.340 about call center dashboards, querying databases, 00:10:21.539 --> 00:10:23.679 or generating thought leadership, we have to 00:10:23.679 --> 00:10:26.360 talk about the underlying engine. And that is 00:10:26.360 --> 00:10:29.549 the data itself. There is a very old, very true 00:10:29.549 --> 00:10:31.450 saying in computer science, garbage in, garbage 00:10:31.450 --> 00:10:34.090 out. Always true. And when you are dealing with 00:10:34.090 --> 00:10:36.789 high stakes finance, you absolutely cannot afford 00:10:36.789 --> 00:10:40.769 for your AI to be making decisions based on unverified, 00:10:41.169 --> 00:10:44.149 noisy public information. It is the single biggest 00:10:44.149 --> 00:10:47.450 risk factor in enterprise AI adoption. An AI 00:10:47.450 --> 00:10:50.129 model is only as accurate and unbiased as the 00:10:50.129 --> 00:10:52.710 data it is trained on. This is why the panelists 00:10:52.710 --> 00:10:56.289 from PGM heavily emphasize a strategy they call 00:10:56.289 --> 00:10:58.750 the walled garden approach. I love that term 00:10:58.750 --> 00:11:01.009 walled garden. Explain the mechanics of that 00:11:01.009 --> 00:11:02.730 for us because it sounds like they are basically 00:11:02.730 --> 00:11:04.669 fencing the AI off from the rest of the world. 00:11:04.830 --> 00:11:07.269 That is essentially what it is. Instead of allowing 00:11:07.269 --> 00:11:10.070 their AI to roam free across the open Internet, 00:11:10.110 --> 00:11:13.049 which is terrifying, which is filled with unverified 00:11:13.049 --> 00:11:15.769 claims, outdated regulations and inherent biases. 00:11:16.210 --> 00:11:19.129 Instead of that, these firms deploy their LLMs 00:11:19.129 --> 00:11:21.820 within a strict internal perimeter. The AI is 00:11:21.820 --> 00:11:24.940 grounded only in curated proprietary data sets. 00:11:25.259 --> 00:11:27.419 It only reads the firm's vetted internal knowledge 00:11:27.419 --> 00:11:29.940 bases, their secure data lakes, and their approved 00:11:29.940 --> 00:11:32.559 policy documents. So it doesn't know what people 00:11:32.559 --> 00:11:35.440 are saying on Reddit. It only knows what the 00:11:35.440 --> 00:11:38.000 firm's chief investment officer has officially 00:11:38.000 --> 00:11:40.940 published. Precisely. And establishing that Waldgarten 00:11:40.940 --> 00:11:44.820 matters immensely. First and foremost, it drastically 00:11:44.820 --> 00:11:48.220 reduces the risk of AI hallucinations. If the 00:11:48.220 --> 00:11:50.629 model doesn't know the answer, It is constrained 00:11:50.629 --> 00:11:53.730 to say, I don't know, rather than confidently 00:11:53.730 --> 00:11:56.090 inventing a false statistic. Which we've all 00:11:56.090 --> 00:11:59.389 seen happen. Exactly. Furthermore, this walled 00:11:59.389 --> 00:12:01.970 garden creates a secure sandbox for the employees. 00:12:02.490 --> 00:12:04.690 It gives them a protected environment to experiment, 00:12:05.370 --> 00:12:07.750 to learn how to properly prompt the system and 00:12:07.750 --> 00:12:10.169 to build their digital literacy without any risk 00:12:10.169 --> 00:12:12.750 of accidentally leaking sensitive internal data 00:12:12.750 --> 00:12:15.600 to a public model. But setting up that walled 00:12:15.600 --> 00:12:17.480 garden isn't just a matter of buying a software 00:12:17.480 --> 00:12:20.200 license. No. The panel really stressed that this 00:12:20.200 --> 00:12:23.000 requires a massive culture shift. There has to 00:12:23.000 --> 00:12:25.639 be an incredibly tight ongoing partnership between 00:12:25.639 --> 00:12:27.620 the business teams and the technology teams. 00:12:28.059 --> 00:12:31.000 If tech operates in a silo, you run into that 00:12:31.000 --> 00:12:33.919 classic pitfall where engineers build a highly 00:12:33.919 --> 00:12:36.639 sophisticated, shiny new tool, they hand it over 00:12:36.639 --> 00:12:39.740 to the portfolio managers, and no one uses it 00:12:39.740 --> 00:12:41.820 because it doesn't actually solve a real business 00:12:41.820 --> 00:12:44.919 problem. We see that happen constantly. The technology 00:12:44.919 --> 00:12:47.759 must be driven by actual business use cases, 00:12:48.240 --> 00:12:51.100 not just the desire to deploy AI for the sake 00:12:51.100 --> 00:12:54.620 of deploying AI. And that requires comprehensive 00:12:54.620 --> 00:12:57.879 enterprise -wide training from the executive 00:12:57.879 --> 00:13:00.440 board all the way down to the junior analysts. 00:13:00.940 --> 00:13:03.159 Everyone needs a baseline understanding of how 00:13:03.159 --> 00:13:05.519 these generative models function, how to evaluate 00:13:05.519 --> 00:13:08.299 their outputs, and most importantly, what their 00:13:08.299 --> 00:13:10.799 limitations are. Which brings us to perhaps the 00:13:10.799 --> 00:13:12.820 most critical part of this entire deep dive, 00:13:13.039 --> 00:13:15.460 the security aspect. Here we go. Because everyone 00:13:15.460 --> 00:13:17.759 in this industry is moving fast. The security 00:13:17.759 --> 00:13:19.940 leader from TNC Bank used a really memorable 00:13:19.940 --> 00:13:22.120 analogy for this. They said that in the current 00:13:22.120 --> 00:13:24.639 market, everyone wants to drive a fast car. Everyone 00:13:24.639 --> 00:13:26.960 wants to adopt AI quickly to stay competitive. 00:13:27.440 --> 00:13:30.100 But to drive a fast car safely, you need really, 00:13:30.100 --> 00:13:32.460 really good brakes. The guardrails have to be 00:13:32.460 --> 00:13:34.440 established before you step on the gas. Here's 00:13:34.440 --> 00:13:37.039 where it gets really interesting. Because the 00:13:37.039 --> 00:13:39.259 consequences of driving this fast car without 00:13:39.259 --> 00:13:41.919 breaks are not theoretical, they are already 00:13:41.919 --> 00:13:45.000 playing out. The sources detailed some severe 00:13:45.000 --> 00:13:48.179 incidents from 2024, where companies, suffering 00:13:48.179 --> 00:13:50.519 from a fear of missing out, allowed employees 00:13:50.519 --> 00:13:53.360 to use publicly accessible Gen. AI chatbots for 00:13:53.360 --> 00:13:56.000 financial analysis without implementing proper 00:13:56.000 --> 00:13:58.740 data loss prevention controls. It is a nightmare 00:13:58.740 --> 00:14:01.500 scenario for any chief information security officer. 00:14:01.659 --> 00:14:04.179 It truly is. You had well -meaning employees 00:14:04.179 --> 00:14:06.600 taking sensitive client portfolios, including 00:14:06.600 --> 00:14:09.279 massive amounts of personally identifiable information, 00:14:09.740 --> 00:14:12.379 and pasting that raw data directly into public 00:14:12.269 --> 00:14:15.169 chatbots just to format a report. Just trying 00:14:15.169 --> 00:14:17.429 to save time. Yeah. They didn't realize that 00:14:17.429 --> 00:14:19.350 by doing so, they were feeding their clients 00:14:19.350 --> 00:14:21.690 private data into the public models training 00:14:21.690 --> 00:14:25.190 set. The result was massive data exposures and 00:14:25.190 --> 00:14:27.389 millions of dollars in regulatory fines, all 00:14:27.389 --> 00:14:29.830 because they bypassed basic internal guardrails. 00:14:30.230 --> 00:14:32.470 And accidental data leakage is only one side 00:14:32.470 --> 00:14:35.070 of the coin. The threat landscape has fundamentally 00:14:35.070 --> 00:14:37.710 changed because malicious actors are also weaponizing 00:14:37.710 --> 00:14:39.899 this technology. Oh, the panel did not hold back 00:14:39.899 --> 00:14:42.740 on this. They highlighted the terrifying reality 00:14:42.740 --> 00:14:45.399 of AI generated phishing and social engineering. 00:14:45.919 --> 00:14:47.980 We are no longer talking about poorly worded 00:14:47.980 --> 00:14:51.279 scam emails full of typos. Right. Attackers are 00:14:51.279 --> 00:14:54.220 using AI to generate hyper realistic deep fakes 00:14:54.220 --> 00:14:57.220 of executives. They will clone a CFO's voice 00:14:57.220 --> 00:14:59.639 and call a junior employee to authorize an urgent 00:14:59.639 --> 00:15:02.639 wire transfer. Or they will execute incredibly 00:15:02.639 --> 00:15:05.460 sophisticated business email compromises that 00:15:05.460 --> 00:15:07.620 perfectly mimic the writing style of internal 00:15:07.620 --> 00:15:10.500 leadership. This is what we refer to as the cybersecurity 00:15:10.500 --> 00:15:13.740 arms race. The adversaries are using AI to scale 00:15:13.740 --> 00:15:16.419 their attacks, automate their vulnerability scanning, 00:15:16.919 --> 00:15:19.379 and generate highly convincing synthetic media. 00:15:19.799 --> 00:15:21.779 So you have to fight fire with fire. Exactly. 00:15:22.179 --> 00:15:24.019 The defending teams have absolutely no choice 00:15:24.019 --> 00:15:27.200 but to utilize AI as well. A security operations 00:15:27.200 --> 00:15:29.840 center at a major asset manager simply cannot 00:15:29.840 --> 00:15:32.740 rely on human analysts to manually sift through 00:15:32.740 --> 00:15:35.190 millions of daily network logs anymore. So how 00:15:35.190 --> 00:15:37.730 are the defenders utilizing AI to actually fight 00:15:37.730 --> 00:15:40.750 back and gain the upper hand? They're using specialized 00:15:40.750 --> 00:15:45.129 internal AI models to process those unimaginable 00:15:45.129 --> 00:15:48.490 volumes of event data at machine speed. The AI 00:15:48.490 --> 00:15:50.850 establishes a baseline of normal network behavior 00:15:50.850 --> 00:15:53.429 and then constantly hunts for anomalies. Like 00:15:53.429 --> 00:15:55.909 finding a needle in a haystack? Faster than any 00:15:55.909 --> 00:15:58.450 human could. Yeah. It helps defenders spot the 00:15:58.450 --> 00:16:01.289 unknown unknowns, the subtle complex attack patterns 00:16:01.289 --> 00:16:03.409 that a human analyst would never catch in time. 00:16:03.629 --> 00:16:06.190 But the panel also offered a stark reminder. 00:16:06.590 --> 00:16:09.470 AI models are, at their core, just software. 00:16:09.570 --> 00:16:11.789 Right. And like any software, they have vulnerabilities. 00:16:12.669 --> 00:16:14.610 The security leaders emphasized that firms must 00:16:14.610 --> 00:16:17.250 continuously stress test these models using established 00:16:17.250 --> 00:16:19.649 cybersecurity frameworks. They specifically mentioned 00:16:19.649 --> 00:16:22.090 leveraging resources like OWAS, which is essentially 00:16:22.090 --> 00:16:24.169 the global standard for application security, 00:16:24.429 --> 00:16:27.970 and OpenAI primers to rigorously hunt for flaws 00:16:27.970 --> 00:16:30.149 in their own walled gardens. It sounds like a 00:16:30.149 --> 00:16:32.370 relentless high stakes battle. But what I found 00:16:32.370 --> 00:16:34.230 really encouraging in the sources is that AI 00:16:34.230 --> 00:16:36.490 isn't just being used to fight off global hackers. 00:16:37.029 --> 00:16:38.769 It is also being aimed at solving one of the 00:16:38.769 --> 00:16:41.049 most massive, tedious pain points in the corporate 00:16:41.049 --> 00:16:44.649 world. I am talking about compliance. Ugh, yes. 00:16:45.289 --> 00:16:47.309 The labyrinth of modern financial regulation. 00:16:47.659 --> 00:16:49.940 The panel called this the dream of killing the 00:16:49.940 --> 00:16:52.059 questionnaire. If you operate anywhere in asset 00:16:52.059 --> 00:16:54.879 management, you know the absolute agony of conflicting 00:16:54.879 --> 00:16:58.000 regulations. You have state privacy laws colliding 00:16:58.000 --> 00:17:00.720 with federal financial regulations, which then 00:17:00.720 --> 00:17:03.580 collide with various industry certifying authorities. 00:17:03.659 --> 00:17:06.000 And it never ends. It doesn't. And typically, 00:17:06.140 --> 00:17:09.059 a firm handles this by deploying armies of compliance 00:17:09.059 --> 00:17:11.859 analysts to do what they call stare and compare 00:17:11.859 --> 00:17:15.559 work. They sit there. manually cross -referencing 00:17:15.559 --> 00:17:18.160 hundreds of spreadsheet cells, trying to figure 00:17:18.160 --> 00:17:21.619 out how to satisfy a 50 -page vendor risk assessment. 00:17:21.680 --> 00:17:24.680 It is a massive drain on resources and it is 00:17:24.680 --> 00:17:26.920 highly prone to human error. But the experts 00:17:26.920 --> 00:17:29.940 on our panel detailed how agentic AI is actively 00:17:29.940 --> 00:17:32.740 being used to harmonize these conflicting frameworks. 00:17:33.240 --> 00:17:36.420 The AI can ingest a massive new federal regulation, 00:17:36.920 --> 00:17:38.900 cross -reference it against the firm's existing 00:17:38.900 --> 00:17:41.579 internal controls, and instantly highlight the 00:17:41.579 --> 00:17:44.319 gaps. It maps the requirements across different 00:17:44.319 --> 00:17:46.940 jurisdictions to create a single rationalized 00:17:46.940 --> 00:17:50.000 set of controls. The ultimate goal these firms 00:17:50.000 --> 00:17:52.700 are driving toward is automating compliance detection 00:17:52.700 --> 00:17:56.140 entirely. They want the AI to continuously monitor 00:17:56.140 --> 00:17:58.980 the firm's systems and output the required reporting 00:17:58.980 --> 00:18:01.660 automatically, effectively cutting out the middleman 00:18:01.660 --> 00:18:04.140 and killing the endless compliance questionnaires 00:18:04.140 --> 00:18:06.480 once and for all. That would change the industry 00:18:06.480 --> 00:18:09.180 overnight. So what does this all mean? We have 00:18:09.180 --> 00:18:12.220 covered incredible ground today. Looking at everything 00:18:12.220 --> 00:18:15.059 these leaders from MetLife, Capgemini, PNC, and 00:18:15.059 --> 00:18:18.759 PGM have laid out, it is abundantly clear that 00:18:18.759 --> 00:18:22.000 AI in asset management is no longer an experimental 00:18:22.000 --> 00:18:25.400 toy. Not at all. It is a top to bottom restructuring 00:18:25.400 --> 00:18:27.779 of the financial plumbing. We are looking at 00:18:27.779 --> 00:18:30.299 a future powered by agentic assistance capable 00:18:30.299 --> 00:18:33.339 of eradicating the bulk of manual friction. But 00:18:33.339 --> 00:18:36.019 it is a future that completely hinges on pristine 00:18:36.019 --> 00:18:39.119 data governance through walled gardens and ironclad, 00:18:39.299 --> 00:18:41.799 thoroughly tested security guardrails. If we 00:18:41.799 --> 00:18:44.019 step back and view this in a broader historical 00:18:44.019 --> 00:18:46.940 context, we are squarely in the middle of the 00:18:46.940 --> 00:18:49.380 Fourth Industrial Revolution. And what this deep 00:18:49.380 --> 00:18:52.279 dive truly underscores is the pivot in the value 00:18:52.279 --> 00:18:55.220 of human capital. For decades, the financial 00:18:55.220 --> 00:18:57.539 industry rewarded the manual gathering, sorting, 00:18:57.700 --> 00:19:00.619 and verifying of data. Right. That era is ending. 00:19:01.299 --> 00:19:03.640 The true value of the human worker is moving 00:19:03.640 --> 00:19:06.759 entirely toward interpreting the profound, synthesized 00:19:06.759 --> 00:19:10.500 insights that these machines deliver. The machine 00:19:10.500 --> 00:19:13.220 does the heavy lifting of data aggregation, while 00:19:13.220 --> 00:19:15.480 the human provides the strategic judgment and 00:19:15.480 --> 00:19:17.880 the ethical oversight. It's a total shift in 00:19:17.880 --> 00:19:20.680 skill sets. It is. But as we look ahead, this 00:19:20.680 --> 00:19:23.119 raises a profound question. Something I want 00:19:23.119 --> 00:19:26.240 you, the listener, to consider long after this 00:19:26.240 --> 00:19:28.339 deep dive ends. Let's fast forward five or ten 00:19:28.339 --> 00:19:31.559 years. If agentic AI successfully automates the 00:19:31.559 --> 00:19:34.420 vast majority of manual analysis, if it flawlessly 00:19:34.420 --> 00:19:37.099 harmonizes global compliance, and if it seamlessly 00:19:37.099 --> 00:19:39.759 manages our client interactions, how will firms 00:19:39.759 --> 00:19:41.819 actually differentiate themselves in the marketplace? 00:19:41.960 --> 00:19:44.319 That's a great point. When every single competitor 00:19:44.319 --> 00:19:46.480 has purchased the exact same hyper -efficient 00:19:46.480 --> 00:19:49.359 AI capabilities, technological speed is no longer 00:19:49.359 --> 00:19:53.049 an advantage. It is just the baseline. In a fully 00:19:53.049 --> 00:19:55.750 automated, flawlessly efficient financial world, 00:19:56.190 --> 00:19:59.130 will deeply held human intuition, empathy, and 00:19:59.130 --> 00:20:01.349 the art of creative relationship building become 00:20:01.349 --> 00:20:04.490 the ultimate scarce premium asset? That is the 00:20:04.490 --> 00:20:06.369 ultimate question we are all going to have to 00:20:06.369 --> 00:20:08.630 answer in the next few years. Thank you so much 00:20:08.630 --> 00:20:10.950 for joining us on this deep dive. We hope this 00:20:10.950 --> 00:20:12.910 exploration helped you cut through the hype and 00:20:12.910 --> 00:20:14.970 grab hold of the real mechanics shaping the future 00:20:14.970 --> 00:20:17.769 of finance. Keep learning, keep questioning the 00:20:17.769 --> 00:20:19.910 information landscape around you, and we will 00:20:19.910 --> 00:20:21.109 catch you on the next one.