← The AI in Business Podcast
The AI in Business Podcast
Breaking Free from AI Overwhelm in Banking and Financial Services - with Art Shectman of Elephant Ventures
00:33:23
Loading summary
Transcript33 lines
- [00:00]
A
Foreign.
- [00:13]
B
Welcome everyone to the Emerge AI in Business podcast. Today's guest is Art Shekhtman, CEO and founder at Elephant Ventures. Elephant Ventures delivers agentic AI solutions for complex enterprises, translating institutional knowledge into measurable competitive advantage. Shechtman examines why financial services and banking leaders remain stuck in evaluation mode despite growing board pressure to deliver AI results and what separates the teams that actually ship from those that don't. The conversation covers how to select a trustworthy first workflow, how to use time bounded sprints to reach production, and how early wins change the conversation in the boardroom. Today's episode is sponsored by Elephant Ventures. Getting in front of enterprise AI buyers
- [00:58]
C
is not about impressions.
- [01:00]
B
It's about trust. At Emerge, we help AI vendors engage decision making through research driven content and conversations that matter in the buying process. To learn the exact strategies we use to help leading AI brands and startups connect with their ideal enterprise AI buyers, visit go.emerge.com partner that's go.emerj.com P A R T E R Now the conversation with Art.
- [01:51]
C
Art, it's really great to have you in our Emerge AI in Business podcast studio today.
- [01:55]
A
Hey, thanks for having me on. I'm really excited to be here.
- [01:58]
C
I think it's going to be a great conversation and one that is very relevant to our executive audience at the moment. This entire series is for the leaders who aren't experimenting with AI. They're accountable to boards for outcomes. They cannot afford to experiment any longer. And inside financial services, those leaders are under a very different kind of pressure. They're being told to move faster, but it feels as if the ground is shifting underneath them continuously. And what we're seeing is that teams are basically paralyzed by the fear of building something that will be outdated in 100 days. AI is moving at a speed that is just beyond comprehension for me. From your seat. What is actually driving this overwhelm for board accountable AI leaders in the BFSI right now?
- [02:41]
A
It's a great question. The process of deploying AI solutions in highly regulated or complex industries, especially banking. You wind up with so many pressures that compete with each other. You have regulatory restrictions, you have infrastructure restrictions, you have your risk management team telling you not to do anything ever. And the executives that we're seeing, they have to go into some type of leadership meeting every Monday morning or the first month, whatever it is, and explain what progress they've made. And exactly as you described in your question, 100 days later it might be out of date. And these are very careful, calculating people culturally, like they're meant to make forecasts and then live up to those forecasts that sometimes span multiple years. And now they're making forecasts that are going to hold for a hundred days. And they're asked to work on the concept that the thing they're designing right now or the tools they're using right now might be outdated or might be wrong 100 days from now and might have security problems from the risk team when they try to get them to go on a path to production. And so really, it's just the overwhelm of how much is dynamic, how much is moving all at once, and this feeling of just really overwhelming change that they're being asked to discern and figure out, like, what is the path forward and then to take action on it. And it's just antithetical to the culture and environment that they're in day to day.
- [04:01]
C
And I'm sure you get these phone calls quite often where you get an executive phoning you, saying, art, we are overwhelmed. And that's a conversation that probably goes on for an hour or maybe even longer. But what is the first 10 minutes of that conversation sound like? What are they telling you? What is. Is that overwhelmed feeling? And how do they describe it?
- [04:20]
A
I think most people are tired of vendor pitches. They're reporting back. There's so much snake oil salesmen out there pitching them on all kinds of crazy things. AI they have pitch or platform exhaustion. They just. They don't want to hear another pitch. They want someone to just say, look, for right now, for this moment, we've seen and been successful building with this pattern, this what we would label as like a composable ecosystem of products. And for now, they work. We can drive through to success. But they'll say, I had 12 pitches this week. My inbox is full of talking head AI people. Can you do what you say you can do? Can you demonstrate the case studies that you have, and can you help me create an actionable path forward? Because I'm overwhelmed.
- [05:05]
C
Very much the truth and I. We kind of had this conversation just before our podcast recording where we said our listeners are also part of that group that's really over it of hearing pitches day in and day. They just want something that works. Another term that we've been hearing a lot is vendor fatigue. What does vendor fatigue look like in practice and how does it show up? How do we see it? Is it in the calendar, in the budget, in the team's energy? How do we identify vendor fatigue?
- [05:32]
A
I would say when you're. When you get the latest pitch in your inbox or you're in the middle of listening to someone who managed to get on your calendar and you find yourself just saying, oh man, I don't understand what this is. I don't really care. And I've heard three pitches that sound somewhat similar to this, and I'm unclear if any of them actually work. I just need something that works. I mean, that's kind of what it is. Or you just stop altogether. Like, I'm not taking another meeting. Like, I can't, I can't hear another, another pitch on another product that I partially understand. And it's a real problem because the challenge, like, we started this process for ourselves, like we were in their shoes from a innovation and software engineering delivery standpoint. We had a giant target painted on our back when LLM coding first became a thing three, three and a half years ago. And we saw it coming and we're like, man, we have to get ahead of this because our competitors are going to get ahead of it and we have to know what the answer is. And so we started looking at every product under the sun. We started evaluating, you know, taking all kinds of calls, sitting on all kinds of vendor pitch. And it is exhausting. But I do think it's the way we got over it for ourselves. You begin to have, There's a term that's coming out these days, it's gaining promise, this concept of harness engineering. Like, what are all the pieces that I need to make my AI project successful? What are they called? What are the generic boxes that five vendors have similar products and they fit into and having names for things so you know who to compare and which piece, like as they're pitching to you, where it fits in the landscape that's really important. And getting comfortable with what the descriptions or usefulness of each of those pieces in the harness, what does it do? For me that helps you kind of avoid vendor fatigue. And then honestly, giving yourself permission not to take the fifth pitch of the same thing or the slightly similar same thing, the things you'll learn from the first one or two pitches for different pieces of that harness, they're applicable to almost all of them. And then when you meet it in production, when you're like, okay, we're deploying that piece of our harness or that piece of our composable ecosystem of AI tooling, then you have a much better decision making framework and a way to choose between vendors. And so you kind of sidestep the fatigue because you very specifically know what you're shopping for in that Moment.
- [07:46]
C
I like it. It's kind of like window shopping is not an option. And we also have this fear of missing out and the pressure that is in all industries actually currently of being part of the AI movement and our competitors have AI integrated, why don't we or we should. So if, if we have the opportunity to hear from five to 10 vendors and we know that it's not, the solution is not hearing every pitch. How do we select which ones? Is there a certain term or a question that we should be asking to say, okay, is this the right pitch to be listening to? How do we make sure that we're not just exposing ourselves to more noise?
- [08:22]
A
That's a tough one. So sometimes when you're looking for differentiated functionality or something very specific to what you're trying to build, you want to talk to the small guy, you want to talk to the startup and listen what the founder has to say. She might know something that you don't know and that isn't popular wisdom. And that's why she's built a startup in AI and has a cool product that has a point solution need for you or you want just general education and so you kind of need to know what piece of the selection process you're in and why you're doing it. If you just want general vendor education, finding the three biggest, most popular the names you see all the time and just getting their regular standard pitch, it's a great way to educate yourself on what that piece of the harness does and to see what they have to say about it. But if you're, if you're seeking something more specific, well then maybe cast a little bit wider net to see what
- [09:13]
C
you can possibly and I feel like those smaller guys might also help you to customize more precisely. They still have the energy and the ambition to listen to you and to use your use cases specifically. Is that what we're seeing in reality?
- [09:30]
A
I would say thar be dragons. There's a sliding scale from like just implementing like one of the big hyperscalers, like fully drink the Kool Aid, use everything they have the way they do it, or using the tooling providers and building stuff for yourself for your purpose built needs. There's a kind of a difficult relationship with your vendor if you're saying, well I want your product, but I also want a bunch of stuff your product doesn't really do. And there's sort of some counter incentives there's for kind of trying to strong arm your vendor, especially a smaller vendor, because they want your business so badly into Customizing the product for you. Now that said, most of these AI companies these days, the different vendor space, they have some type of concept of for deployed engineering, they have some configurable or composable layer to their product so they can get fairly customized for your needs. And so you just have to balance. What are you asking? Are you configuring and customizing the implementation or are you actually trying to implement things or influence things to get implemented in their product roadmap? And you're probably not the best person to be putting pressure on their product roadmap. And maybe that will be a difficult situation for you going forward versus being on the configuration or composable side of it, which is easy for them to do. And then you need to know whether or not you want to invest and you have the capabilities on your own team or you have a vendor partner to help you implement and do truly custom stuff. But that's really, it's situational. But you can look yourself in the mirror and say, I think I know what we need. But if you find yourself in a territory where you're pushing on a vendor for their product to do stuff that it doesn't yet do, you should say, hey, am I trying to drag them in a strategically conflicting direction for their own needs? And then know that you're going to have some problems.
- [11:18]
C
So it's very much a two way street if you're going that route or thinking that direction. We've now looked at what is, what the overwhelm is and what is causing it and how we're basically adding to it. And I think if we trying to get someone to add to their product things that they have not developed and tasted fully, it might even increase that overwhelm. I want us to, to get into, let's move from overwhelm to action. When you work with banks, financial services organizations or insurers who actually get unstuck, those that get out of the pilot mode or out of the AI stuck mode, what is a practical confidence building path forward look like?
- [11:58]
A
Yeah, great question. So we have been there, we've been in their shoes. When we did it to ourselves, we got overwhelmed too. And the only way we found to get out of it was to start doing. And we would say like we have an overwhelming bias towards action. And what we realized when we were in their shoes, we kept trying to master everything. We tried to get through all of the different pieces of the system and know what they were doing. And then as soon as you're done, you have to repeat the research, because all new products came out and all new thinking came out and all kinds of things. And it's Sisyphusian, like you're pushing the stupid rock up the hill and then it falls back while you're sleeping and you just never get there. And so you have to take action. You have to put something into reality and confront what it still is not doing for you or what other piece of the system must come next to make it successful. And so we really try to help folks go as we had to internally to realize, like, let me, let me do something. Something built and something that's taken shape is way more important than taking the perfect right first action. And then I would say also like enlist partners. Your job, wherever you are, is likely not to spend full time studying all the products that are working and all the pieces of a composable ecosystem or a gentic harness to know exactly what's the right answer for this moment in time and then investing all the time to keep up to make sure that that view is also great for tomorrow and next week and next month. You have to repeat that activity. You have a day job. And so finding people who make that their job and partnering with them, learning from them, sharing knowledge together is important. I think we have a lot of thinking both from our own experiences, culturally, what you have to do to prepare people to go through the process of adopting and building AI solutions. And so really taking stock of what people are saying and having a clear point of view about how to get started and empowering people to get started. There's things we learned along the way that were super valuable. Especially maybe the top of the list is confronting people's fear that their job is going to be replaced and helping them get excited about the acceleratory power of AI solutions or agentic solutions and how it shifts their job to the more interesting parts and frees them up from administrative burdens. And then I also think, you know, everyone's tired of the conversation of 95% of pilots fail and like, great, thanks. That doesn't actually help me build something that works. And once you can pick a thing and be okay with feeling pressure that's coming from your risk group or feeling pressure from your internal IT people or some kind of governance body and saying, let me figure out what needs to be true to make that work at my company, instead of saying, I don't have all the answers and therefore nothing will ever work at my company, it's like, well, let me start. Let me put a stake in the sand. Let me Build towards a path to production and answer those questions each in turn as I'm pushing on them. And then I think the third magical thing for us was you have to pick the right initial workflow or project people naturally want to pilot these pie in the sky amazing. Like wouldn't it be miraculous if this pilot did all these amazing AI assisted super agent things for me? But they're never going to trust the agent to do that. So the pathway from experimentation to production for that pilot is not realistic. And so picking a simple workflow, picking a workflow where your organization already trusts the outcome, that's important. Because if you can't batch up and be deterministic, batch up all the context you need to execute something with AI and then trust the outcome when your people do it. Let's say you do a workflow and the people in your organization argue over what the right answer is or what the right approach and there's some internal religious war on, oh no, the TPS report has to have purple paper versus pink paper or my way's better or I've been here for 20 years, I just know how to do it. Those workflows are not going to be easy to implement. And if you don't trust and have declarative versions of success of what you're trying to accomplish in your organization and you couldn't trust a junior employee, let's say, to do the same thing that that 20 year senior experience person might do, you probably picked the wrong thing. And then there's so much capability building, so much context, repository, development of organizational wisdom and knowledge you need to distill and make available for your AI projects, all that stuff becomes a precursor to your path to production. And so kind of changing the lens that you look at and say what's trustworthy? Like what do we trust our people to do? Very simply, what has a reasonable amount of context associated with the, the inference parts, the judgment parts of how you're going to get that thing done. And if you pick the right size thing that you have a clear understanding of what context you need to do it, and your organization trusts the output, then you're usually likely to find a path to production that can be made to work and it forms a great forcing function, a kind of a strawman to say, well, what needs to be true at my company for this to get to production and then you can solve those challenges one at a time and get it into production and actually have some value coming out the other side. And then you get to go into your monthly or Monday morning board meeting and you're a hero because you launched something. You're not just talking about PowerPoints and strategy.
- [17:26]
C
I like that. And you touched on something that our financial services leaders really, I think it's a word that when they're kind of triggered by it and it's trust and trust for the normal human being, trust is an emotion but and financial services trust is very much keeping your job, not getting regulatory fines, not ending up in prison, not losing money, not allowing fraud to happen. So trust is a bigger thing for them. And if you're saying that it's essential to basically choose the right workflow to start with. Can you give us an example of a single workflow that you would pick first in a bank or at an insurance company? And why, why start there?
- [18:04]
A
I think we see two popular avenues of work. Everybody in their, you know, is, is trying to accomplish sales and marketing automation and expanding their sales and marketing outreach. There's a ton of workflows in the sales and marketing workflows at a bank and acquiring new customers and promoting new products. There's tons there that you can automate and gain success in pretty quickly. There's some regulation there about how what you say outside and what you say to the public. Then there's a bunch of internal workflow efficiency, just pure administrative tasks. And so you can pick kind of one or two of those. It's pretty easy to get some quick wins. We've seen some really interesting stuff around coaching or client development. So especially from a B2B financial services standpoint, using the coaching you might provide on operational efficiencies or growth or all kinds of things for an end client, you'll have internal like significant point of view as a, as a financial institution on what they should be considering or thinking about. You probably also have some soft product alignment to the things that you offer that help them do those things differently at their bank. And so figuring out which, figuring out which internal flows that kind of provide coaching or training to your end customers. Those are some pretty good places in banking, financial services or insurance where we're seeing some good traction. And then simple internal workflows that are inference based decision making and let me unpack a bunch of words. So at its core LLMs, they provide inference as a service. We all lived through the robotic process automation wave 10 years ago in banking and financial services and we automated all these deterministic workflows. Deterministic meaning like it had clear if then else kind of operations and steps. Well some of those workflows became hard, and you'd have to get a human involved because there were some inference, some judgment. You had to use the facts at hand, the context that you add, and infer the right next step, keeping certain regulatory frameworks in mind or sales touts in mind. And so this whole revolution, if you boil it all down, it's all about inference on demand or inference as a service. And so the level of workflows you can address versus the ones that you could do in the RPA days are now sort of one level more strategically valuable, where you had to pause and have humans do some inference work for you in the past. Now you can actually bridge those gaps programmatically. And so taking simple examples of workflows where you're spending a lot of time and effort and you have a lot of human involvement, and then unpacking the parts of that process that you can then actually automate or offload, that's where we've seen internal gains for some banks be pretty dramatic.
- [20:52]
C
And it makes sense because I think if we just take it back to basic human interaction with basic training new employees, you always give them the smaller tasks first and say, okay, if you can do this, then we can eat up to something more important. Or, this is still not high stake. You're not touching the client yet. You cannot get me in trouble just yet. So do this. And once you've perfected this, we'll move on to the next step. And so it makes sense to do it like that. I'm thinking that as an executive, the person that has to report to the board, and we know the boards want big transformative bets. They want. They want to boil the ocean. Most of the times. They don't want us to pitch them this, oh, let's do it in incremental workflows. How do we have that conversation with the board? What is the right way to approach it? What's the language that we should be using in. In telling them this is actually the correct way to start. This is the correct action to take. And this is why I think historically,
- [21:44]
A
it was about building a fortress. You build these amazing foundations, use these giant stones. You create the walls to defend against the barbarians at the gate. And you create this really trustworthy, gonna live forever castle that is just this fortress that's impenetrable, and then you fill it full of financial and insurance mathematicians and geniuses who can figure out the right moves to make, and you provide that surety and wisdom and guidance to people. And transformation in those environments has always been super hard right. We've spent 20 years helping Complex enterprises go through those transformational initiatives. And we've talked about this term transformational inertia, where it's very difficult to overcome a firm's transformational inertia because there are so many factors at play that make that not the right move in any sense of the imagination for anybody who works there or culture culturally or regulatory wise or anything else. And so what we're seeing now is you're asking these people to hop on a surfboard and surf and there is no solid ground. It is this moving wave. There are so many revolutions happening faster and faster every day. And you're asking your board to give you permission to live in that environment while they want to still be in the fortress. And so it starts off from a place of conflict. And I guess what we're seeing is that a lot of the board level folks are so tired of hearing strategy. They're so tired of seeing PowerPoints and slides and visions statements. What they really want to see is in dollars and cents. What did you do? What did we actually use AI for? And if you try to have the conversation without that in hand, it's very difficult because they want to see a result. But if you ask them to kind of get behind the concept that you're going to iterate and rework stuff and it's working for now and you're going to keep expanding and doing capability building, but you had a quick win. It's been much easier for the folks we've worked with to get that next tranche of strategic investment greenlit by their boards. We find that the boards are very excited for tangible progress and it loosens up the purse strings and tolerance for risk and kind of gets them to a place where they're willing to listen more. Once you've launched something, even if it's small, the fact that you've actually gotten all the way through, found the path to production and launched something is really the grease the wheel needs.
- [24:13]
C
That makes sense. I think the boards are also at this point where they've also been hearing pictures and they've also been promised the moon and the stars and everything. So to them, if you have a realistic end goal and you can can kind of show how you got to this, what the thought process was, why motivate, why it's enough for now, that's going to sound like a better investment than just throwing money at it and hoping for the best. I want us to land this conversation for our executive audience. They always like to know, okay, art You've given us some great insight. You've painted the picture. We know that you feel our pain, but what should we be doing right now? So what is the immediate path forward for AI leaders in financial services that need to show progress to the board without blowing up the risk if they had to go to that Monday morning meeting, what is step one of getting to that point where they're not stuck in pilot mode anymore?
- [25:09]
A
I think you have to organize yourself into some type of shortened, sprinted environment. You have to say, okay, we're going to have a 100 day initiative, we're going to have a 30 day initiative and go through a process of really actively listening to your people and what's achievable and making it time bound and say, by the end of this time, what could we have? What could we have that's working? And in some sense, in putting the pressure on yourself to get something alive, find the minimal path to production, you will cut some corners, you'll do some things that maybe aren't the greatest for the long haul, but really everything you're building right now might not be the greatest for the long haul. So historically your immune system would be like, no, don't do that. Build for the future always. But at this point, you don't know enough to know what that future is really going to be in any kind of a reasonable or solid way. So build for now, ish. And then get something into production. Solicit all those people from regulatory and governance and compliance and tech and data quality and everything else to get you to a place where you can have the factors in place that let something go live. But time bound and say, okay, we'll learn for 60 days, 90 days, whatever you want it to be, but have a time bound pressure ticker for yourself, your teams, everything to say, okay, by this date we're going to have something alive. And people are immensely creative when you make it time bound and then really do it again, because then they'll know very clearly, oh, well, we cut corners here, we used some duct tape over here. So let's go clean that up so that our next round will have better capabilities. But if you try to boil the ocean and have all the capabilities built in parallel at the same time, you're going through the cultural transformation of your people to use the tools, you just get nowhere. You wind up with a hundred things started and nothing finished. And so really limiting your work in progress so that you can finish something in a fixed amount of time, that's the best kind of recipe for success that we've seen. And then really, your people already have a laundry list of 15 or 20 or 30, 100, 200 workflows or agents or agentic things they want to do. If you can come up with a rubric internally of how well you're trusting them or how much you might trust them, how solid your understanding of the context that's required, you have to think like, I'm gonna, I want this thing to do the work of my best people. But if you expect it to do the work of your best people, but you don't have the knowledge in a place that's usable or consumable by AI from your best people, well, how do you expect it to do the same job they're doing without the context that they have? So you have some work to do. Building what we call like your context singularity. You have to build your context singularity. You have to have picked workflows where you have a reasonable chance of trusting it, which means you'll get a path to production and it'll survive the process of moving into production. And if you take a time minimalist view of that, you'll kind of get this prioritized list of workflows that you can trust, that you know what the outcomes are, that you know the context you need for it, so they'll likely survive. And then you apply the time filter and say, oh, well, given the five that bubbled to the top, these three are going to take us a year to build based on our current capabilities. These two we could finish in our time box. So let's start there.
- [28:26]
C
I like it. I like that you're not telling us to just do everything and pick what works. It's start with small blocks, have that specific time goals in mind. It's very true that people work best under pressure. Like you mentioned, the creativity is insane when the time restraint is there. I think our biggest takeaway from this conversation today is that our banking, financial services, and insurance, they don't need more AI noise. The noise is there. They're exposed to it. They've heard every pitch that there is. What they need is they need to focus on one or two workflows that they can actually ship, things that can actually work inside their organization. And from there they'll see what works and what does, and they'll know where to build. The overwhelm is not necessarily a leadership flaw. It is the reality of shifting models, louder vendors and pilots that that kind of age out within months. You're just getting started. You're just Getting used to it. And it's already outdated. And it sounds like the only teams that are making progress on this will be the ones that are not focused on their readiness, but focus on, let's just start building, let's start somewhere. That cliche of how to eat an elephant one bite at a time. And that is our key takeaway today is have that pre cleared workflow, know exactly what it is that you want from it, put that time constraint on it and just do it. Am I right? Is that our message that we want to send today?
- [29:49]
A
100%. We see so many people get caught up in trying to do everything and a lot of large corporations you might have very many competing initiatives and opinions. And once you give the folks that are having divergent opinions, once you give them the understanding like we're making the best decision for right now, we could re litigate this decision 100 days from now, but right now we have to get something in production and they all have that shared mission of okay, you can have all of your different opinions or different agendas, but here's how we're going to prioritize together. And we all have to get something into production in 100 days. That's where you start to be able to say, okay, well, I'll compromise here. We'll get on board with this one flavor of the decision of the five different variants that different people might want to do and then you can come back and iterate. But being able to pick something to get done, everyone's already ready, you're already ready. Nobody wants to know how ready or not ready they are. They want to go into their board meeting and say, I delivered a thing and here's the value it had. Here's the next 12 we're going to deliver. You're 100% on it. Just finding the thing that can work and then finding a pathway to coach your people into some amount of agreement with the permission to revisit it soon and short time cycles to come back and revisit in case the decision is wrong, I think helps move things towards getting everybody aligned to the goal of getting something into production and remove some of the barriers you see in larger complex corporate environments.
- [31:20]
C
Art, I think you've definitely given our audience something to think about and something to consider and they'll definitely enter that board meeting on Monday with a different type of confidence regarding this topic. Thank you so much for the time that you spent today with us talking about this.
- [31:33]
A
Yeah, thanks. It was a fun discussion. I appreciate it.
- [31:51]
C
Wrapping up today's episode, let's look at
- [31:52]
B
our three key takeaways from the conversation with Art.
- [31:55]
C
First, financial services leaders who move from
- [31:57]
B
evaluation to production do so by identifying a single workflow their organization already trusts, rather than designing for the most ambitious use case. Second, time bounding the first deployment. Committing to a live result within a fixed sprint is the forcing function that cuts through competing internal agendas and gets teams aligned around the shared goal. Finally, early production wins even small ones, resets the board conversations from strategy and slides to tangible value, unlocking the next round of investment and organizational permission to keep building. Position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap for the opportunity to showcase your solution to the executives currently funding and scaling global initiatives. Partner with Emerge to reach the decision makers holding the strategic mandate. Secure your partnership@go.emerge.com partner that's go emerj.com p a r tner for further executive level analysis and to join our network of leaders delivering workflow impact with AI, visit emerge.com on behalf of the team at Emerge.
- [33:06]
C
We'll see you on the next episode.