Jouk Pleiter argues AI shifts banking from access to execution via unified frontlines and elastic models
Summary
Jouk Pleiter, CEO of Backbase, discusses his novel 'AI Waits for No Bank' and argues that AI represents a larger revolution than the internet by shifting banking focus from customer access to internal execution. He introduces the 'Unified Frontline' concept, where humans and AI agents collaborate across silos using a shared context layer called 'Banking Operational Memory'. Pleiter asserts that model capability is no longer the bottleneck; instead, robust architecture, semantic routing, and complex entitlements are required to safely authorize millions of agents. The discussion covers 'elastic banking', where AI enables massive scale with reduced headcount, and the rise of 'digital twins' and data wallets. Pleiter warns against 'AI theater', urging banks to integrate AI into core workflows rather than isolating it in innovation labs.
Key takeaways
- 1 Pleiter asserts that 'model neutrality' is achievable through semantic routing, where intent determines whether a frontier model or a small language model optimized for banking subdomains is used.
Unidentified speaker (speaker_2) · 5:54
- 2 banking operational memory. And to a large degree this is inspired by Palantir, uh where you say you need to create an ontology or a customer state graph where in real time you can get aggregate content or data points from your customers that is in your card system, in your risk system, in your credit score system, in your core system, in your portfolio management system. And you basically want to aggregate that data into a real time customer state graph.
Unidentified speaker (speaker_2) · 9:47
- 3 Pleiter describes 'signal engines' that tap into CRM and data layers to monitor signals, triggering proactive, conversational agent interactions rather than static marketing like banners or emails.
Unidentified speaker (speaker_2) · 15:55
- 4 Pleiter coins 'elastic banking' as a model where AI enables dramatic scale, potentially allowing a bank to serve millions of customers with a drastically reduced headcount (e.g., 500 people serving 5 million customers).
Unidentified speaker (speaker_2) · 22:54
Episode sections
- 0:20 – 2:25
Introduction and Book Context
Brett King and Paolo Cerroni introduce guest Jouk Pleiter, CEO of Backbase, discussing his new novel-style book 'AI Waits for No Bank'. Pleiter notes the book is a scientific and philosophical story, distinct from his previous works.
- 2:25 – 4:36
AI vs Internet Revolution and Unified Frontline
Pleiter argues AI is bigger than the internet revolution, shifting focus from access to execution. He introduces the 'Unified Frontline' concept, where humans and AI agents collaborate across siloed channels with shared context, enabled by a horizontal orchestration layer.
- 4:36 – 8:51
Architecture Over Models and Entitlements
Paolo Cerroni compares AI to the 2008 financial crisis, emphasizing that robust architecture matters more than specific models. Pleiter agrees, stating the battle is now about model neutrality, guardrails, and complex entitlements to authorize agents safely.
- 8:51 – 11:49
Banking Operating System and Operational Memory
Pleiter describes a Banking OS that sits above legacy systems, featuring a 'Banking Operational Memory' (Nexus). This layer unifies real-time customer state graphs, historical decisions, and policies to give agents the full context needed for optimal decisions.
- 11:49 – 14:49
Tokenization and System of Truth
Cerroni suggests AI architecture must converge with blockchain tokenization for efficiency. Pleiter agrees, advocating for a 'system of truth' and 'system of work' distinct from legacy 'systems of record' to enable end-to-end agent execution.
- 14:49 – 16:40
Signal Engines and Pre-science
Pleiter explains 'pre-science' via signal engines that monitor data for triggers, allowing agents to proactively engage customers through dialogue rather than static banners. This requires a wide data overlay and robust guardrails.
- 16:40 – 22:54
Agent-to-Agent Interactions and Disintermediation
The speakers discuss the risk of disintermediation by personal AI agents. Pleiter suggests banks must fight back by integrating into customer agents or building their own conversational agents, while Cerroni notes clients are already using frontier models for portfolio advice.
- 22:54 – 29:02
Elastic Banking and Digital Twins
Pleiter introduces 'elastic banking,' where AI allows banks to serve millions with fewer staff. He predicts a future of agent-to-agent negotiations and the need for 'digital twins' for both consumers and banks to operate effectively at scale.
- 29:02 – 30:16
AI Models vs. Operating Model
Speaker_2 argues that LLMs are no longer the bottleneck for AI implementation; the challenge lies in building the operating model, memory layer, and guardrails necessary for high-stakes sectors like healthcare and finance.
- 30:16 – 32:18
Digital Twins and Data Wallets
Speaker_0 discusses building digital twins for health and finance, citing Drumwave as a company presenting to Congress on data wallets, which treat personal data as an asset that can be traded or monetized.
- 32:18 – 33:55
Generative Products and Outcome-Based Banking
Speaker_0 describes a shift from static products to generative, outcome-based experiences where AI bundles credit sources in real-time, potentially rendering traditional departmental structures obsolete.
- 33:55 – 37:10
Conversational Banking and Liquid Interfaces
Speaker_2 defines conversational banking as a dynamic, liquid interface that moves beyond static UIs, using intent recognition and deterministic guardrails to orchestrate customer resolutions.
- 37:10 – 39:35
Banking OS Architecture Layers
Speaker_2 outlines the necessary layers for an AI-native bank, including integration, truth, authority, intelligence, orchestration, and UI layers, emphasizing that legacy systems must be orchestrated by a meta-layer.
- 39:35 – 52:12
Cultural Agility and AI Theater
Speaker_2 warns against 'AI theater' such as isolated innovation labs, advising leaders to adopt a growth mindset and integrate compliance from day one to avoid stifling real change.
Claims worth verifying
Jouk Pleiter is the CEO and founder of Backbase.
Unidentified speaker (speaker_0) ·1:04
How to verify: Backbase corporate website or LinkedIn profile of Jouk Pleiter. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Pleiter is located in Amsterdam at Backbase's global headquarters.
Unidentified speaker (speaker_2) ·1:35
How to verify: Backbase corporate headquarters address in Amsterdam. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
The banking industry has struggled for decades to achieve a unified frontline with shared context across channels.
Unidentified speaker (UNKNOWN) ·3:33
How to verify: Industry reports on omni-channel banking challenges (e.g., Gartner, Forrester) or Backbase whitepapers. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
The 'Banking Operational Memory' concept is inspired by Palantir.
Unidentified speaker (speaker_2) ·9:47
How to verify: Backbase marketing materials or Pleiter's book for explicit attribution to Palantir's ontology approach. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
OpenAI is doing signal detection in the US via 'Plath' (likely Plaid integration or similar product).
Unidentified speaker (UNKNOWN) ·17:39
How to verify: OpenAI or Plaid press releases and documentation regarding US market integrations. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Revolut has made early attempts at AI-native banking capabilities.
Unidentified speaker (speaker_0) ·18:49
How to verify: Revolut press releases or product announcements regarding AI features. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Pleiter suggests a bank could potentially run with 500 people serving five million customers.
Unidentified speaker (speaker_2) ·23:15
How to verify: Transcript of the interview; this is a speculative claim by the speaker. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Customers are already asking private bankers to review portfolios using frontier models.
Unidentified speaker (UNKNOWN) ·19:45
How to verify: Anecdotal claim; could be verified via surveys of private banking clients or industry case studies. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Agent-to-agent negotiations and a 'digital twin' for consumers are likely developments within the next 5-10 years.
Unidentified speaker (UNKNOWN) ·22:34
How to verify: Speaker's prediction; verifiable only by tracking industry adoption timelines against this forecast. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
LLMs are already capable of creating the future described in the book; the bottleneck is the operating model, not the models themselves.
Unidentified speaker (speaker_2) ·29:22
How to verify: Review the book 'AI Waits for No Bank' by Jouk Pleiter for the author's stated position on model maturity vs. architectural readiness. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Drumwave is presenting to Congress this week on data wallets.
Unidentified speaker (UNKNOWN) ·31:17
How to verify: Search US Congressional hearing schedules or press releases from Drumwave for the week of the transcript's release (Nov 2026) or the specific 'this week' reference relative to the recording date. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Backbase services more than 120 banks around the world.
Unidentified speaker (speaker_2) ·48:10
How to verify: Check Backbase's official website or investor relations page for their customer count statistics. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Backbase is a startup that is almost twenty million dollars.
Unidentified speaker (UNKNOWN) ·47:47
How to verify: Review Backbase's funding history on platforms like Crunchbase or PitchBook to verify its valuation or total capital raised. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
The book 'AI Waits for No Bank' has been available since the 6th of September.
Unidentified speaker (speaker_0) ·47:19
How to verify: Check Amazon.com or publisher records for the publication date of 'AI Waits for No Bank' by Jouk Pleiter. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Backbase claims a 10x to 50x improvement in outreach and impact in their Go-To-Market process using agents.
Unidentified speaker (speaker_2) ·42:24
How to verify: Review Backbase case studies or press releases detailing their internal GTM efficiency metrics post-agent implementation. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Mortgage workflows involve around 250 people scanning documents and communicating with clients.
Unidentified speaker (UNKNOWN) ·49:44
How to verify: This is a specific operational claim; verify via Backbase white papers or industry reports on mortgage processing labor costs and headcount. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
There are 20 million phone calls per year in the call center.
Unidentified speaker (speaker_2) ·49:10
How to verify: This appears to be a specific client or industry statistic cited by Pleiter; verify via Backbase client case studies or industry call center volume reports. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Banks are risk-averse, making it difficult for them to embrace risk and experiment with AI.
Unidentified speaker (speaker_2) ·40:25
How to verify: Review academic or industry reports on banking risk culture and AI adoption barriers, such as those from McKinsey, Deloitte, or regulatory bodies. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
The book 'Quantum Sapiens' is available where good books are sold.
Unidentified speaker (speaker_1) ·51:11
How to verify: Check Amazon or major book retailers for the existence and availability of a book titled 'Quantum Sapiens' by Brett King. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.
Do your own research
- Verify Backbase's reported 10x to 50x improvement in GTM/SDLC outreach by reviewing their latest earnings reports or case studies.
- Check Palantir's public documentation or filings regarding their 'ontology' and 'Nexus' platform capabilities in financial services.
- Research Drumwave's 'data wallet' proposals and any congressional testimony or regulatory filings associated with them.
- Assess the maturity of 'semantic routing' and 'model neutrality' implementations in current banking tech stacks by reviewing vendor whitepapers.
- Investigate the regulatory landscape for 'rogue' AI agents and the feasibility of implementing human-like entitlements for autonomous software.
- Review industry reports on 'elastic banking' and labor displacement metrics in the financial sector to validate headcount reduction claims.
- Look for counter-arguments regarding the security risks of 'digital twins' and centralized customer state graphs in cybersecurity literature.
Full transcriptexpand
- 0:20
- Unidentified speaker (speaker_0): Welcome to Breaking Banks and this special edition, special episode, which includes the banker's bookshelf. And uh Paolo and I've been doing a few of these uh lately, but uh Paolo Cerroni, uh welcome back to this uh mashup between the two podcasts. How how is Quantum Sapiens going?
- 0:39
- Unidentified speaker (speaker_1): Oh the book is going very, very well. Uh it has been uh a startup in itself because this time around is a novel. Scientific and philosophical story, but uh a novel quite different from the past. And Brett uh I've saw quite a number of books uh being published recently, and the one of you is very interesting. So I'm so happy that today
- 0:58
- Unidentified speaker (UNKNOWN): we have
- 0:58
- Unidentified speaker (speaker_1): uh you clatter with us for both
- 1:01
- Unidentified speaker (UNKNOWN): banks
- 1:01
- Unidentified speaker (speaker_1): uh and the bankers bookshelf. Couldn't miss this one.
- 1:04
- Unidentified speaker (speaker_0): The the CEO and founder of Backbase um is also a uh um best selling author. His latest book, AI Waits for No Bank and No Man. Um, you know, um well unless you're uh Dario Modi or Sam Altman, maybe, but um
- 1:24
- Unidentified speaker (speaker_0): having said that, the slowdown is is is definitely a topic we're gonna get into today. Yalk, um it's a great pleasure to have you back on the pod. Uh thanks for joining us. Uh where are you coming to uh to from from today?
- 1:35
- Unidentified speaker (speaker_2): Today I'm uh in Amsterdam in our uh global headquarters. So nice to meet you, Paul and uh Brett. Good to see you.
- 1:42
- Unidentified speaker (speaker_0): It's good to see you. It's good to see you. So um first of all, um uh you know, let's get to the foundational story uh of the book. Uh obviously, you know, you you're an engineer first and uh founder uh you know and CEO uh second. Um but um you know this is not your
- 2:02
- Unidentified speaker (speaker_0): first book. Um what what prompted uh this one in particular? Uh you know, why because it's a beautiful book, uh showing the audience. Very nicely. I'm a little jealous actually in terms of the quality of the of this. Um but um what what uh brought you
- 2:21
- Unidentified speaker (UNKNOWN): to
- 2:21
- Unidentified speaker (speaker_0): this decision it was time to write a a book on AI?
- 2:24
- Unidentified speaker (UNKNOWN): Uh
- 2:25
- Unidentified speaker (speaker_2): I think it was in the With the Rise, of course. of um well the first LLMs uh we started to think like okay what what are the implications for uh for our industry what are the implications for backbase
- 2:36
- Unidentified speaker (UNKNOWN): um
- 2:36
- Unidentified speaker (speaker_2): I think we all lived through the you know the internet revolution and I I got this sense like this is gonna be bigger than the internet.
- 2:42
- Unidentified speaker (UNKNOWN): Uh
- 2:43
- Unidentified speaker (speaker_2): so but then what does that mean? Um I think in with the internet waves we basically were very busy to give banks access uh to internal systems. Uh with mobile app, it's it's mostly an access game. Uh but now what we see is that with with AI, um
- 2:57
- Unidentified speaker (UNKNOWN): it's
- 2:58
- Unidentified speaker (speaker_2): probably going to be more about how do we execute work uh inside the bank. So the notion is that with agents basically starting to execute work, how do we make sure that basically agents have access to the right information so they have something to reason on? Um and maybe there's an opportunity uh instead of replacing all the legacy, we can now kind of create a
- 3:17
- Unidentified speaker (speaker_2): horizontal orchestration layer that sits on top of pretty much every system. How do we educate agents to go deeper into these systems? How do we give them the right context? How do we give them the right policies? If you do so and then translate that into business value,
- 3:33
- Unidentified speaker (UNKNOWN): we
- 3:33
- Unidentified speaker (speaker_2): started to see a concept we call a unified frontline. And unified frontline is basically. banks have been struggling for this for decades. Like where all the different frontline actors, the customers, the mobile channels, the online channels, the call center, the RM, et cetera, et cetera, they are all working in silos. Nobody has a shared context. Nobody has a shared
- 3:53
- Unidentified speaker (speaker_2): outcome operating model. And that's basically because we have access channels. And Brett, um, you're very familiar with that concept, of course.
- 4:01
- Unidentified speaker (UNKNOWN): So
- 4:01
- Unidentified speaker (speaker_2): I started to sense like, wow. this this AI future and these agents, we might have a real shot at um coming up with an architecture that could truly unify the frontline of the bank. Because historically, to wrap it up, and then over to you guys, but like online banking and mobile banking has been more like an add-on, and you're basically creating another channel. And there was a lot of
- 4:21
- Unidentified speaker (speaker_2): uh talk about omni-channel, but very few banks are really truly operating in an omni-channel fashion. So I think we have a new shot with new technology, and um that's uh I think the core of the book uh a unified frontline where humans and agents are collaborating.
- 4:36
- Unidentified speaker (speaker_1): I think this is very important and I like the way you put it out because uh let me give you an example, going back to the start of the global financial crisis. I was ahead of quantitative risk management back then. And I remember there were two type types of banks, uh those uh in Aymor, Aymort the with the latest and greatest quantitative model, more and more sophisticated, seemed to be very
- 4:56
- Unidentified speaker (speaker_1): powerful to front-front competitors in terms of pricing and capture in the market. But when the KRAZIT arrived, those banks were not capable of managing those models, or those models were not even available to them any longer. Instead, the banks that had a proper risk management architecture, they knew how to orchestrate their work, figure out what was going on. provide the right
- 5:15
- Unidentified speaker (speaker_1): answers, and basically control the system were better placed to survive the crisis and lead stronger out of that. Now, these days with AI seems to be the same. Everyone discusses the super fancy model, okay? The frontier of the frontier, and that's nice. However, are we building the right architecture? Can we create the proper
- 5:35
- Unidentified speaker (speaker_1): orchestration that allows us to go through change? Because some of these models might not be available tomorrow, right? Is it a session of the day? But if the architecture is sound, then you can basically root, re-root, right? Uh, and make sure that you remain flexible enough to deal with all of this uncertainty.
- 5:52
- Unidentified speaker (speaker_2): Yeah. Yeah.
- 5:54
- Unidentified speaker (UNKNOWN): Well,
- 5:54
- Unidentified speaker (speaker_2): if I can reply to that, I I I think we are now, and this was not the heritage of the book, but where we are today, I think the models are not the issue anymore. I think we're gonna do a model neutral. model neutrality. You can you can basically be model neutral, we can do model routing. What I see, for instance, with conversational banking, we basically have a semantic router, and it just basically depends on the intent. It will route it to, you know, a
- 6:14
- Unidentified speaker (speaker_2): frontier model, it can route it to a small language model that is really optimized for banking use cases in subdomain. So I think the model routing is uh is super clear. Maybe even more interestingly, I think the the battle is not perceptive. per se about the models anymore. The battle is about how do we kind of bring the models to life in a regulated industry. So where's the where's the harness,
- 6:34
- Unidentified speaker (speaker_2): where's the guardrails uh around uh the models so we can make them more predictable? I ironically, uh what I see happening right now is that we have to bring more deterministic workflow and logic and rules and policies surrounding agents to really operate at scale. Maybe to bring this to life, I think especially
- 6:54
- Unidentified speaker (speaker_2): how do you authorize millions of agents in a bank?
- 6:59
- Unidentified speaker (speaker_2): You don't want to have open claw inside your bank, right? So what you what we basically see, like with people, we already had to solve this with a very complex entitlements engine. So for instance, the Is commercial a banking, like what's the service agreement, what's the legal entity, what are all these different actors, and up to the level of on this particular account, you know, the payment level, you know, you basically have a very fine
- 7:19
- Unidentified speaker (speaker_2): granular model to authorize humans inside the bank, inside the customer commercial banking uh client organization, uh, to collaborate. And you've got a full audit trail, etc.
- 7:31
- Unidentified speaker (UNKNOWN): I
- 7:31
- Unidentified speaker (speaker_2): don't say it's going to be exactly the same, but inspired by that entitlement uh operating model for humans, we're gonna see something similar for A. agents. I I really think kind of how do we entitle agents, how do we give them permissions, how do we make sure that they can get a green ticket to proceed or they get a red ticket and and they basically cannot proceed. I think that's also the
- 7:51
- Unidentified speaker (speaker_2): the core of the discussion we now see about and like how do we stop progress? Like how do we prevent agents to go to go rogue in swarms and and basically make up their own mind. In a regulated industry we need to have an an an authorization control harness around it to basically drive it to secure
- 8:09
- Unidentified speaker (UNKNOWN): uh
- 8:09
- Unidentified speaker (speaker_2): application inside the bank and basically help customers. That's more like say how do we help customers conversational wise, uh coach-wise, advisor advisory-wise, but also how do you basically analyze every task inside the bank in any existing workflow? You analyze the type of work, how do we train agents to do that work? And how do we then kind of orchestrate the
- 8:29
- Unidentified speaker (speaker_2): whole thing to collaborate?
- 8:31
- Unidentified speaker (UNKNOWN): You
- 8:32
- Unidentified speaker (speaker_0): you you talk in the book about the um the the unified unification of of these various systems, AI native banking operating system over the top. Um but uh you know a lot of the decision-making Decision making capabilities you're looking for, yeah, process workflow, that's all important. But for unification,
- 8:51
- Unidentified speaker (speaker_0): part of the key problem is just data.
- 8:55
- Unidentified speaker (speaker_0): And so, you know, going beyond the semantic layer, which obviously we can use some of the frontier models, what are you seeing as sort of leading edge approaches to not only the data lake problem, but the foundational National model problem that underpins the unification strategy? Because you you you need
- 9:15
- Unidentified speaker (speaker_0): an architecture, but you also need a data architecture for this. So um, you know, uh what what how how are you approaching that in terms of um the the unification process?
- 9:27
- Unidentified speaker (UNKNOWN): Yeah.
- 9:28
- Unidentified speaker (speaker_2): Our thinking is for this kind of banking operating system is that first it is agnostic of the system of record, uh, two, it's also agnostic of any data lake or any customer data platform. So let's not confuse these things. I I think a banking operating system is not operating there, but it's operating on top. I think what we need is basically what we call an a
- 9:47
- Unidentified speaker (speaker_2): banking operational memory. And to a large degree this is inspired by Palantir, uh where you say you need to create an ontology or a customer state graph where in real time you can get aggregate content or data points from your customers that is in your card system, in your risk system, in your credit score system, in your core system, in
- 10:07
- Unidentified speaker (speaker_2): your portfolio management system. And you basically want to aggregate that data into a real time customer state graph. And the way you do it is basically also by unifying the language, how you call stuff, like your domain model. That's basically what we call ontology, very much inspired by uh by Falunteer. volunteer. Next to that, you don't only want to have access to the right data. within the right
- 10:27
- Unidentified speaker (speaker_2): simmers uh syntax, but you would like to have access, agents like to have access to your historical decisions. There are so many decisions in banking that get lost. So we call a context graph where you basically want to memorize like from the initial onboarding, the KYC check, the AML ML check, uh, the credit check, like whatever we do,
- 10:48
- Unidentified speaker (speaker_2): we basically do all of these checks and then we pass. We don't see store them, and we basically enroll the customer into the core banking system. But there's a lot of value basically in the decisions we're making in day-to-day banking work. Our recommendation is to basically store that as well. Next to that, of course, you have all your operating uh standard operating procedures, so your documents, the
- 11:07
- Unidentified speaker (speaker_2): way you want to work, your policies, your product descriptions, etc. et cetera. So I think all of these different ingredients need to come together in one. what we call the operating banking memory or banking operations memory. It's a combination of real time customer data, historical decisions, standard operating procedures, et cetera. Historical context of previous
- 11:27
- Unidentified speaker (speaker_2): conversational banking dialogues with your clients. So agents can go to, we call that layer Nexus. Nexus is basically the layer that will give context and truth to agents. If you don't go for this unification, these agents will just make decisional subsets of data which are never representing the full picture of the client. So the
- 11:47
- Unidentified speaker (speaker_2): decision cannot be optimal.
- 11:49
- Unidentified speaker (speaker_1): Well, as you use the word memory and the word uh trust, so immediately comes to mind that the ledger of the blockchain. So I think there's a wide space in the industry talking about AI, and it's is the fact that uh an AI architecture goes end in hand end hand with uh a tokenization architecture because all these agents will have to access tons of data in order to make
- 12:09
- Unidentified speaker (speaker_1): the decisions that they have to make, and that becomes very expensive. But however, the moment you tokenize the data that goes around every customer interaction, that might be more uh effective. Also, you can embed some elemental programmability so sort that of to resolve some of the elements of uh the intentionality of the agent. But I see very few. banks that are now lining up their
- 12:29
- Unidentified speaker (speaker_1): AI perspective with the tokenization perspective as if they still have two different heads, right? But I do believe that going forward will be the convergence of the two that will provide uh the needed meat on the bone to deliver the return investment which is required for this type of transformation.
- 12:46
- Unidentified speaker (UNKNOWN): I
- 12:46
- Unidentified speaker (speaker_2): think you're right. Um And I think like uh in a way, just we have history. historically we know about system of records, right? And they they will stay. And I think we also have data lakes kind of all ready to do aggregate some stuff, uh, and that's all that's all great. But like I think ultimately as an industry we need to go to a system of truth and a system of work. So the there's the system of records
- 13:06
- Unidentified speaker (speaker_2): will stay there, then there's the system of work where you want to have agents and humans to collaborate. Okay, how do we need to have entitlements, we need to have policies, we need to kind of make sure that these agents can work within the right guardrails and execute the work. But for both humans and agents, you need to have a truth layer. Um, and I think that's also the evolution, how we see AI probably
- 13:26
- Unidentified speaker (speaker_2): evolving. We can we can basically, I think the current banking uh, let's say, setup, it will never thrive in an AI world because everything is in us is in basically siloed applications, so point solutions, point. channels. So we can we can unleash a little bit of let's say an AI copilot within an app or in a but it can only kind of operate within that little
- 13:46
- Unidentified speaker (speaker_2): context. It so and that is a productivity gain. So it'd say great we go from zero to one we have a nice uh productivity gain but it has to go to from zero to ten. The next steps are basically AI that have that can operate beyond a point application. And they basically have to operate in in an end an to end-to-end end uh let's say let's execute the work. So it's not a
- 14:06
- Unidentified speaker (speaker_2): little task, it's the full work package. The work package could be uh resolve a payment dispute or or any any other kind of long horizon type of workflow. So to in order to kind and I think ultimately as an industry, that is where we do we need to go, and then you see some emerging pattern. So if you want to get there, you need to have a truth layer, you need to have an authority layer,
- 14:26
- Unidentified speaker (speaker_2): uh, you need to have a context layer. Like these concepts come uh front and center in a yeah, basically a world where agents start to execute work.
- 14:36
- Unidentified speaker (UNKNOWN): And
- 14:36
- Unidentified speaker (speaker_2): basically
- 14:39
- Unidentified speaker (speaker_0): the concept of agent civilizations, you know, is is being bandied around right now. Um
- 14:44
- Unidentified speaker (UNKNOWN): which
- 14:45
- Unidentified speaker (speaker_0): is sort of an interesting way to sort of think about this. But um
- 14:49
- Unidentified speaker (speaker_0): I I I do have a uh uh you know it's something as a futurist that that sort of caught my eye in the book is when you're talking about the unified front line, you talked about the four dimensions of of this from a um you know operating perspective, and one was pre-science. Um and so your ability to
- 15:08
- Unidentified speaker (speaker_0): anticipate the customers' needs and sort of agentically prepare for that, I think, is a a really smart thing. But how do you model
- 15:18
- Unidentified speaker (UNKNOWN): that?
- 15:19
- Unidentified speaker (speaker_2): Yeah, so the um
- 15:21
- Unidentified speaker (UNKNOWN): in
- 15:21
- Unidentified speaker (speaker_2): the system, I think what is crucial and it's part of the through layer, but there's this notion of a data overlay layer. And I think that's not new. I think that's the data lakes we've seen and and and uh etc. But like
- 15:32
- Unidentified speaker (UNKNOWN): what
- 15:32
- Unidentified speaker (speaker_2): we now see is basically what we call it a signal engine.
- 15:35
- Unidentified speaker (UNKNOWN): Uh
- 15:35
- Unidentified speaker (speaker_2): that is, and it can tap into your CRM, it can tap into the truth layer, it can tap into your existing snowflake data bricks, data layer. But you would like to ongoingly monitor all sorts of signals that basically sit in your data. So uh up to your point, uh both of you, a proper data strategy, and I think a lot of banks already have done a lot of work in that direction. But like
- 15:55
- Unidentified speaker (speaker_2): the key thing is the signal engine, and then from there the signal engine basically can give uh triggers uh to agents that can reason over the wider context of that client it can also kind of basically what are we gonna do with this uh signal and then uh the agent can basically very proactively reach out to the customer and that can man itself itself in basically not
- 16:15
- Unidentified speaker (speaker_2): just a banner or a little text or an email but I think it's gonna manifest itself very concrete in a conversational coach that will just ping you and basically will start to exp you you can interact with it.
- 16:27
- Unidentified speaker (UNKNOWN): So
- 16:27
- Unidentified speaker (speaker_2): I think over are the days of static emails and banners and next basket action, but it's basically it will happen in the form of a dialogue.
- 16:35
- Unidentified speaker (UNKNOWN): And
- 16:35
- Unidentified speaker (speaker_2): of course,
- 16:35
- Unidentified speaker (UNKNOWN): that dialogue
- 16:36
- Unidentified speaker (speaker_2): also needs to have guardrails to make sure that it is uh properly done.
- 16:40
- Unidentified speaker (speaker_0): But you know, increasingly isn't the isn't that going to be split between personal AI that you have on your phone and like this seems to be the direction, even with the recent. Apple i iOS uh changes. You're gonna have your personal AI. It's gonna r interact with
- 16:56
- Unidentified speaker (UNKNOWN): the
- 16:56
- Unidentified speaker (speaker_0): bank semantic layer as you've got in the architecture. So where's the handoff occur and you know um and and and how does uh you know how's that work in in
- 17:05
- Unidentified speaker (UNKNOWN): in real
- 17:06
- Unidentified speaker (speaker_0): terms?
- 17:07
- Unidentified speaker (speaker_2): First, I really think banks have to compete for the experience layer, right? Uh and let's face it, I think people are talking with GPT and Anthropic, etcetera, etc. uh like like Claude, they're talking about their health. They're talking about their financial future. So in that sense, if you talk about disintermediation, that is happening at a very large scale. So either we we let's follow that path,
- 17:27
- Unidentified speaker (speaker_2): it's gonna happen over there. So I still think you need to do the signal detection. You would like to have a wide data overlay so you kind of have the maximum amount of signal. In a way, I think OpenAI is doing that with Plat, Plath, for instance, in the US, to a degree. Yes.
- 17:39
- Unidentified speaker (UNKNOWN): So
- 17:39
- Unidentified speaker (speaker_2): you know, you don't need the bank, right? I think the question for the bank is do you want to fight back? So what is your data overlay strategy? And then I think your your signal engine and your agents can interact with your GPT agents and together you can give them uh you know a collaboration model. I think agents are, you know, we're gonna see agents that are going to negotiate or instruct on behalf of the
- 17:59
- Unidentified speaker (speaker_2): bank. We're gonna going to see agents that basically negotiate on behalf of the customer and there's going to be a handshake. I think that's that scenario is a little bit more down the road, but that's it's certainly it's it's moving in that direction.
- 18:10
- Unidentified speaker (UNKNOWN): I
- 18:10
- Unidentified speaker (speaker_2): think there's a step before that where banks want to fight back and say, okay, I better then build my own conversational agents, I better it build my own signal engine, or I I use one that is provided by uh players like Backbase so I can actually provide that experience directly in my own mobile experience. And that's how you fight back.
- 18:29
- Unidentified speaker (UNKNOWN): Uh
- 18:29
- Unidentified speaker (speaker_0): but I do think that um, you know, as we start getting into generative models and things like this, that becomes harder to do. Um we are seeing uh a few early attempts at this, you know, um Revolut were there and and things like this. But uh you know, I I mean I'm just thinking in terms of the dominance that
- 18:49
- Unidentified speaker (speaker_0): these frontier labs have. in an advisory capacity today, you know, if I'm thinking about my health, the first place I'm checking is with the frontier models in terms of uh health data. I I think the same is probably true of uh, you know, financial planning, planning a trip, you know, all of this sort of stuff. I I it's pretty difficult to
- 19:09
- Unidentified speaker (speaker_0): think that a bank is going gonna to broadly be able to compete with that unless it is where. there's an established relationship that your agent knows to go to your bank and and and ask this. But you know, I I I the I don't know if the domain of the bank is the place that you go for financial advice in this new world. That's
- 19:29
- Unidentified speaker (speaker_0): that's a it's a pretty fundamental question, right?
- 19:32
- Unidentified speaker (speaker_2): I I'm I'm with you. I'm with you.
- 19:35
- Unidentified speaker (UNKNOWN): Yeah
- 19:35
- Unidentified speaker (speaker_1): it's not that but what fascinates me is that that was I was took talking into to some private bankers uh well established uh and what amazed them is that uh their clients now ask uh
- 19:45
- Unidentified speaker (UNKNOWN): a
- 19:45
- Unidentified speaker (speaker_1): frontier model
- 19:47
- Unidentified speaker (speaker_1): to review their uh portfolios. And so when they get in the conversation with a private banker, they are ready to ask a question. The private banker doesn't expect that because typically the client uh is like uh a minority in the conversation, the bank banker has more power. So now technology is empowering the client. to have more opinions. Now what happens here is that we're not
- 20:07
- Unidentified speaker (speaker_1): anymore into the signal of behaviors. We might get into the signal of thoughts. So what if at your age you have an agent to whom you're talking to, and now that agent can go out looking for another agent in a bank, in a private Probably bank, a bank in and a the third bank, and come up with some ideas. So now we we're are going to get into an agent-to-agent agent to agent conversation.
- 20:27
- Unidentified speaker (speaker_1): and the whole architecture as well of the detection of the behavior is kind of different, right? And the experience as well may not be anymore an experience set for humans, but an experience set for uh in technology intermediaries, which changes completely the framework, right? In the in the logic.
- 20:46
- Unidentified speaker (UNKNOWN): Yeah,
- 20:46
- Unidentified speaker (speaker_2): I think
- 20:47
- Unidentified speaker (UNKNOWN): I
- 20:47
- Unidentified speaker (speaker_2): I personally really think like gosh, guys, what would you do if you would run a bank? Uh right? You surrender or are you gonna fight back? Uh no,
- 20:54
- Unidentified speaker (UNKNOWN): no, no.
- 20:54
- Unidentified speaker (speaker_0): I I don't think I I think you have to be unconventional in the way you fight back. I think this is the the key. I I don't think you can sort of say, let's let's try and get the customer back to the bank. I think you now have to say, How do we get the bank into the customer's life? How do we integrate the bank into the agent? You know, like
- 21:15
- Unidentified speaker (speaker_0): with Claude today, you can go to Claude and you can connect all these different services to Claude, for example. Um, I don't see many banks offering even that basic capability. Yeah, you've talked about the plaid and an OpenAI uh thing, but you know, um, you know, which banks uh uh are saying, hey, we'll support your agent
- 21:35
- Unidentified speaker (speaker_0): out of the out of the gate. Let us know, um, you know, give us your uh your agent ID. identities and your you know, link into your Claude account the or your um you know your your uh stack that's running your your agent um framework. You know, even that i i is uh is is probably uh a more practical way to to
- 21:53
- Unidentified speaker (UNKNOWN): be competitive.
- 21:55
- Unidentified speaker (speaker_2): It is, but like also if you really uh do game theory like ultimately let's say also agents like let's let's take an e e-commerce commerce uh scenario. An agent needs to basically they go to marketplaces, the agents will just be completely rational based on the criteria and basically find the most optimal. So, any any kind of price uh unclarity, uh, it will completely bring it down to the most
- 22:15
- Unidentified speaker (speaker_2): full transparency. Same with with payment routing. Uh, you you why pick uh you know, uh the bespoke uh payment scheme work if you can do it with uh with basically uh Ether or any kind of other platform, right? so that will happen so i think the same thing is almost like a financial shopping marketplace agents will basically make based on these criteria so
- 22:34
- Unidentified speaker (UNKNOWN): that
- 22:34
- Unidentified speaker (speaker_2): is going to be a very commoditized model I think ultimately because it will basically reduce any type of margin and because it will optimize uh for a zero market future right okay cool so that that is that is definitely an I don't think it's gonna happen in the next two years but it's gonna happen in the next five to ten years maybe I'm too commodity. conservative here, but let's figure that one out. Next to that, I still think like
- 22:54
- Unidentified speaker (speaker_2): we have an opportunity if you adopt a Gentec and you basically retransform the bank into an AI native operating model, you arrive at a concept we call uh elastic banking. Where historically, if you would kind of grow the bank, you need you're not you cannot have a relationship manager for any retail consumer. With with agents, you can. I I call that elastic banking.
- 23:15
- Unidentified speaker (speaker_2): So you can have a million customers, you can go to two million customers, but you can actually reduce your headcount. Provocative, we've been stating like, why not run a bank with 500 people and service five million customers? Like that's the type of ratio we might need to think in. Like if you think in your North Star. So
- 23:30
- Unidentified speaker (UNKNOWN): what
- 23:31
- Unidentified speaker (speaker_2): I would do is that, like, at least like say there's still the trust element regulated industry. So instead of basically going into a marketplace model and you basically your margin. gets commoditized to zero, why not basically continue to reinvent a relationship-oriented model?
- 23:44
- Unidentified speaker (UNKNOWN): And
- 23:45
- Unidentified speaker (speaker_2): that basically means then you know you go truly into hyper-personalization, and your employees can now basically interact with any type of client in a very intelligent way.
- 23:54
- Unidentified speaker (UNKNOWN): So
- 23:55
- Unidentified speaker (speaker_2): yeah, I think uh that there's an opportunity there.
- 23:58
- Unidentified speaker (UNKNOWN): So
- 23:58
- Unidentified speaker (speaker_2): it's a cost reduction. Like, you know, thousands and thousands of people inside. banks basically moving paper, swivel chair, copy-pasting. So I think there's a tremendous uh productivity opportunity in in just, let's say, customer operations and banking operations. And like that's the efficiency uh mandate, I would say, for the
- 24:18
- Unidentified speaker (speaker_2): chief operating officer. And then I think for the head of retail, head of commercial, etc., and the
- 24:23
- Unidentified speaker (UNKNOWN): the
- 24:23
- Unidentified speaker (speaker_2): people that basically the chief digital officer, the chief AI officer, there's also really a rammet like, okay, how are we reinventing the notion of relationship? And the notion of maybe even personal relationship.
- 24:37
- Unidentified speaker (speaker_0): Alright, well, listen, I want to get into that,
- 24:38
- Unidentified speaker (UNKNOWN): but
- 24:39
- Unidentified speaker (speaker_0): I think I think we're gonna have to take a break. Paula, you want to take us to break?
- 24:44
- Unidentified speaker (speaker_1): Yes. So everybody stay tuned. A little break, and we are back with Jockey Joke Plata Pleiter and Bert Break King talking about his new. book about banking and AI.
- 24:54
- Unidentified speaker (speaker_1): Welcome back from the break. This is Paolo Cironi. Today we are hosting Juke Pleiter with his new book AI Waste for No Bank, together with Break King on both Breaking Bank's podcast and the Bank's Bookshelf. So a wonderful combo as we are discussing a book with lots of insights but as well as real practice, as York Joach was very
- 25:13
- Unidentified speaker (speaker_1): successful still and in. is entrepreneurs at Backbase, starting from Amsterdam, where he founded his company. Now, in the first half, we'll be discussing a lot of things around the importance of a unified framework where data are available for a real human experience that is supported by AI and the view of the client. Find And
- 25:33
- Unidentified speaker (speaker_1): we delve a bit into the importance of relationship and how that is changing. But I want to start the second half. with a truly futuristic point of view. Now let's move 10 years ahead in time, and now we have agents talking to agents. I think we have to design no more for the client experience, but for the agent experience. So the moment
- 25:51
- Unidentified speaker (UNKNOWN): we thought that we
- 25:52
- Unidentified speaker (speaker_1): were somewhere and we understood the principles, it seems that it really is shifting again, right? Of course, when you have the right architecture, right, and that perspective, uh, you can master change if you don't take. have that one, you don't. That's why, again, again, and again, it is the architectural perspective, the one that allows the bank to be nimble and to stretch according to the opportunity.
- 26:12
- Unidentified speaker (speaker_1): But you
- 26:14
- Unidentified speaker (UNKNOWN): believe
- 26:14
- Unidentified speaker (speaker_1): uh will
- 26:14
- Unidentified speaker (UNKNOWN): happen 10 years from now.
- 26:16
- Unidentified speaker (speaker_0): But but but before we get into that, you know, this is a really interesting point, Paolo, because I don't see marketers talking about marketing campaigns for agents. You know, I don't see digital offers Office talking
- 26:25
- Unidentified speaker (UNKNOWN): about acquiring agents.
- 26:27
- Unidentified speaker (speaker_1): Agents don't care about your interface. They don't
- 26:29
- Unidentified speaker (UNKNOWN): care about your
- 26:30
- Unidentified speaker (speaker_1): wonderful app, right? So what about that?
- 26:35
- Unidentified speaker (speaker_2): Well, depends on uh how we train the agents and how we connect them to the models, but basically agents and swarms of agents will ultimately go for a uh an outcome, right? So I think we're gonna see agents that uh potentially we're gonna go, and and I know you guys like to go very futuristic, okay, ten years. into the future, but like probably I I my personal opinion thinking process
- 26:55
- Unidentified speaker (speaker_2): right now, and this is today, it probably will change, is that like, hey, we're gonna develop as a human, you're gonna develop your your basically your digital twin. Uh and but like how do we build our own twin? Like how do we train the twin? How do we kind of give it enough context? So because you have to basically explain just like if you would have a PA or someone booking uh a trip for you, like to understand what you prefer. So
- 27:15
- Unidentified speaker (speaker_2): digital twin is is essential, and I I really look forward to new technologies and concepts that will make the creation of a digital twin way easier so it becomes mainstream.
- 27:25
- Unidentified speaker (UNKNOWN): But
- 27:25
- Unidentified speaker (speaker_2): it's a prerequisite because otherwise these agents don't know on whose behalf they should operate and how they should operate within within your preferences.
- 27:33
- Unidentified speaker (UNKNOWN): So
- 27:34
- Unidentified speaker (speaker_2): I think that's that's one thing. So but those agents will basically act and basically execute missions. on behalf of a consumer or a business. Then we have the other side, supply demand, where basically agents basically have to find and win market share from the bank's point of view. Right. So
- 27:50
- Unidentified speaker (UNKNOWN): um
- 27:50
- Unidentified speaker (speaker_2): yeah I think that's that world is going to evolve and at scale millions of agents. So how do we make sure that these agents also can operate within the right guardrails and within the right context within the bank.
- 28:03
- Unidentified speaker (speaker_2): So I think also on the bank side, it's almost like what is the digital twin of the bank? How do we know like over 500 bespoke legacy systems? What is the truth? Uh we best we call that Nexus, the banking operational memory layer, right? Then you need to have a police function that makes sure like these agents, agents need to have access to hundreds of internal systems,
- 28:23
- Unidentified speaker (speaker_2): external systems, act on behalf of the client. So the authorization level, the policy levels, the evidence trail of everything that happens is critical infrastructure to make this whole thing work at scale. So I think we're gonna make uh good incremental steps into that future. I think uh ultimately, let's say, our end state as
- 28:43
- Unidentified speaker (speaker_2): is fully autonomous for the end customer as well as for the supplier of whatever type of uh product or service, but then each of them, like how do we build and progressively build the digital twin for the consumer side? And how we do we build the operating model, one use case at a time, uh on the banking side.
- 29:01
- Unidentified speaker (UNKNOWN): It's
- 29:02
- Unidentified speaker (speaker_2): so
- 29:02
- Unidentified speaker (UNKNOWN): it's
- 29:02
- Unidentified speaker (speaker_2): we're gonna get there and we're gonna get there in a very pragmatic step, very decisively. In the middle, I think of course we're gonna see the LLMs, but I think they are they are there already. I think with most of the LLMs, we can already what we have today, we can create that future. I don't uh in my mind, then it's going to be more about like where do you use open uh frontier models, where do you use open
- 29:22
- Unidentified speaker (speaker_2): source, how do we do sovereignty? But the models are not the bottleneck anymore, I think, for this future. With today's models, I think we can execute this. It's more how to build the twin, how to build the operating model, the the memory layer, etc. That The the basically the guardrails for all the different actors uh to do this,
- 29:40
- Unidentified speaker (UNKNOWN): not
- 29:40
- Unidentified speaker (speaker_2): just goofing around a little bit with your Gmail uh or a note taker, but to basically do this with the essential elements of healthcare, my finances, uh, right? That I think that's the challenge. So I think it's the guardrails and it's the oper how do we build the operating model?
- 29:56
- Unidentified speaker (UNKNOWN): Yeah,
- 29:56
- Unidentified speaker (speaker_0): I've been building the digital twin for my health profile recently. Um and you know, because that's I you know, I've got a lot of legacy data, but um, you know, I've been uh um on a fitness kick the last uh couple of years, getting healthier and and now I'm like trying to look at the data to quantify how that
- 30:16
- Unidentified speaker (speaker_0): improvement's gone. Some of it's straightforward, like heart rate variability and VO2 max and things like that. But on the blood work side, there's opportunities there, and it makes me realize how much data we have about. ourselves that needs to be captured, you know, to get a comprehensive view of you from a digital perspective. And we used to talk about
- 30:36
- Unidentified speaker (speaker_0): net worth,
- 30:38
- Unidentified speaker (UNKNOWN): you know,
- 30:38
- Unidentified speaker (speaker_0): and assets under management as a target for banks. But actually, I think that's the old view. I think, you know, it's It data now. What's What the data you have that can give you, in your words, Yelp, Yak, the pre-side? pre science of being able to understand a customer's uh intent and build the best outcomes for them? This definitely
- 30:58
- Unidentified speaker (speaker_0): requires this sort of comprehensive digital twin model. And and you know, maybe that's the opportunity is to say, we'll help you build your financial digital twin, you know, and uh um we'll build that uh you know, th there there's a there's a pretty interesting opportunity there. I've been working with a company out of um
- 31:17
- Unidentified speaker (UNKNOWN): uh
- 31:17
- Unidentified speaker (speaker_0): the US out of Silicon Valley called Drumwave. Um they're presenting to Congress this week actually on data wallets um and the concept that you know you you know we now should treat data our own data as an asset and be able to trade off that. Um you know, so that you can monetize your own data. But you know, the the logical
- 31:37
- Unidentified speaker (speaker_0): extension of a data wallet, you know, that each of us should have is our you know, having all of the data secure data sets to power our our digital twin. Um yeah, that's a it's it's it's definitely yeah, I we're definitely under something. We we should start a startup on digital twins for banks actually, but uh anyway. Makes
- 31:57
- Unidentified speaker (UNKNOWN): sense.
- 31:57
- Unidentified speaker (speaker_2): Good inspiration
- 31:58
- Unidentified speaker (UNKNOWN): here.
- 31:59
- Unidentified speaker (speaker_0): Let's um let's talk a little bit about generative a AI because um you know we talked about semantic AI, and you we're talking obviously about the orchestration layer and the agentic piece. But an you know, another key piece of what we've seen big advancements in recently in terms of the frontier models and with open source is
- 32:18
- Unidentified speaker (speaker_0): generative capability. Um, you know, I now, you know, like I'm doing um
- 32:24
- Unidentified speaker (UNKNOWN): uh
- 32:24
- Unidentified speaker (speaker_0): presentations, keynotes all around. the world constantly, and I'm c now constantly building those decks using artificial intelligence. Um and uh, you know, i i the the opportunity here is to build really generative products for customers. Because once you go outcome-based, um, you know, we
- 32:44
- Unidentified speaker (speaker_0): used to talk about money experiences and the experience layer versus product layer and things like that. But on the general generative side, the opportunity is to say, all right, you need this type of credit facility. Let me bundle together different credit sources and package this and give you exactly the most efficient form of credit, the cheapest form of credit
- 33:04
- Unidentified speaker (speaker_0): that meets your outcome in real time. And I can be agnostic as to ha the the underlying your products, if you like, that sort of come towards this. But that's an entirely new architecture as well. Like generative products and generative experiences that are outcome-based, as
- 33:24
- Unidentified speaker (speaker_0): you've said, is something that is going to break a lot of organizational strategy. Because just think about the modern bank today. We have the credit d department, you know, we have the mortgage team and the credit card depic team and the the personal loans. team, you know, and that all sort of can potentially uh disappear with with
- 33:44
- Unidentified speaker (speaker_0): agentic um on an outcome basis. So how you know, where's the generative uh experience um layer come into the architecture you're outlining, Jacques?
- 33:55
- Unidentified speaker (UNKNOWN): I
- 33:55
- Unidentified speaker (speaker_2): mean, um there's a couple of layers, lays, but maybe let's start with where I I would say it always starts with the customer. So it starts with the customer intent. uh and they have to express the intent, they will express the intent typically in a mobile app. So I think it also in behavior change. Of course, people will do their intent in uh a frontier model and in external apps, but in the context of banking, I think uh the mobile app is a very
- 34:15
- Unidentified speaker (speaker_2): good point, starting point where people can express their intent. And I think generative technologies will basically do the intent recognition. And let's call it very simple. Maybe i I don't I'm not talking about old school bots. but conversational banking is about okay, what is the intent?
- 34:30
- Unidentified speaker (UNKNOWN): Right?
- 34:30
- Unidentified speaker (speaker_2): And then from the intent it has to basically figure out what's the resolution. If it's a simple question about like what are the type of credit cards you have and how can I select one, you know, that is super basic within today's reality. But like it can go all the way into modular product strategies where if we would kind of have a completely modular product and pricing capability where agents can now
- 34:50
- Unidentified speaker (speaker_2): basically assemble based through dialogue, basically basically the product configuration, product selection, you can completely model that, you can put the guardrails around it, and then it can basically generate the product on the fly.
- 35:02
- Unidentified speaker (UNKNOWN): And
- 35:02
- Unidentified speaker (speaker_2): of course, there can be, you know, the the customer still has explicitly acknowledged it. Yeah, this is what I want, etc. So still human in the loop, human on top of the loop. We we can have all these models, But it starts with intent, and then how do we orchestrate the resolution cycle? Um
- 35:17
- Unidentified speaker (UNKNOWN): so
- 35:18
- Unidentified speaker (speaker_2): conversational banking in my mind is the the the the big uh incremental game changer. It's not the end state but it's basically the step up where we also go from static UI. Internet banking is static UI.
- 35:30
- Unidentified speaker (UNKNOWN): Uh
- 35:30
- Unidentified speaker (speaker_2): mobile banking is static UI. With conversational it's fully dynamic. I think you guys
- 35:35
- Unidentified speaker (UNKNOWN): were
- 35:35
- Unidentified speaker (speaker_2): referring to it as like liquid interface
- 35:37
- Unidentified speaker (UNKNOWN): liquid interface.
- 35:38
- Unidentified speaker (speaker_0): Yeah.
- 35:38
- Unidentified speaker (speaker_2): It will it just absolutely go in that direction. So going to simple, boring forums, there might be elements of that, but I think it's going to be uh very quickly transform. I if we're building new capabilities, we're basically doing it conversational. Why force people to go through? We can still inside the conversation, you want it to be predictable. So you need to do
- 35:58
- Unidentified speaker (speaker_2): deterministic guardrails around the agent to say ultimately you need to collect the these data points, right? Right? Regardless of how creative you are, but this is what you need to do. And ultimately, on these and these and these points, you basically have to ask for permission of the end user, or you have to ask for our decision layer to ask for a decision
- 36:18
- Unidentified speaker (speaker_2): token to move forward or not. Right? So, but like, yeah, I think we're gonna dem generative is gonna be critical in any customer interaction, call it call. conversational banking, intent uh detection, and refinement of the intent. The full dialogue will become liquid, and then other agents in the background will help with the ultimate
- 36:37
- Unidentified speaker (speaker_2): resolution. That could be uh a transaction dispute resolution, it could be assembling a product that is tailored to the customer, and there's another hundred use cases you can imagine.
- 36:50
- Unidentified speaker (UNKNOWN): But to
- 36:50
- Unidentified speaker (speaker_2): your point, like a lot of and this is the point, a lot of the back-end systems are not ready for that granularity. Right. So therefore, so your product and pricing strategy is a critical component. Building your truth layer is critical. The authority layer to give authoritative tokens to proof move forward. So in order to build this future, and that's also why we came up with like
- 37:10
- Unidentified speaker (speaker_2): banking OS and the the the six, let's say, I would say primitives of that operating model. You still need an integration layer. How do I get access to underlying systems? Uh, we need that for online banking, but we we certainly need it for agents. And how do we do that in a secure way? We need a truth layer, we need an authority layer, we need to have an intelligence layer where we can have model routing, uh,
- 37:30
- Unidentified speaker (speaker_2): we need an orchestration layer, we need to have a UI layer, right? But we I I think the classic model of okay, here's internet banking, here's mobile. banking, it will still uh have a role, but it will it it's basically just a channel solution and it will not orchestrate the work end-to-end. In my mind, the summary is
- 37:49
- Unidentified speaker (UNKNOWN): it
- 37:50
- Unidentified speaker (speaker_2): is about it starts with the customer and it ends with the customer. What this basically means customer intent to customer resolution. How do you orchestrate that? Well W in a combination of humans and agents collaborating.
- 38:03
- Unidentified speaker (UNKNOWN): Well,
- 38:03
- Unidentified speaker (speaker_0): it starts with the customer's agent, I guess, right? But uh
- 38:07
- Unidentified speaker (UNKNOWN): yeah,
- 38:07
- Unidentified speaker (speaker_2): ultimately, yeah, uh age yeah, to well, I think both.
- 38:11
- Unidentified speaker (UNKNOWN): I think
- 38:11
- Unidentified speaker (speaker_2): uh customers are not
- 38:12
- Unidentified speaker (UNKNOWN): gonna
- 38:13
- Unidentified speaker (speaker_2): delegate everything to agents
- 38:14
- Unidentified speaker (UNKNOWN): like the
- 38:16
- Unidentified speaker (UNKNOWN): finance. You know
- 38:17
- Unidentified speaker (speaker_2): how you like y you know, I like to see your you're you're fully trusting your agents with everything, uh Brett, and you don't check it. I think
- 38:24
- Unidentified speaker (speaker_2): you wanna want to see it. You wanna see it in the screen. But the agent
- 38:27
- Unidentified speaker (UNKNOWN): is acting
- 38:28
- Unidentified speaker (speaker_2): on your behalf for for sure.
- 38:29
- Unidentified speaker (speaker_0): Yeah, I want the agent to do as much work as possible, but I still want to I still want to verify the uh uh the the decision making process. All right. Um when when obviously we talked a lot about architecture, we've talked a lot about the technology technology supporting all of this, but to have an AI
- 38:48
- Unidentified speaker (speaker_0): AI-native native bank is not only gonna going to require this level of technical agility that the AI-native banking operating system and the unified uh frontline brings us, but in addition, we're going to need cultural agility. So you do touch on this in chapter 13 or 14
- 39:07
- Unidentified speaker (UNKNOWN): of
- 39:07
- Unidentified speaker (speaker_0): the book.
- 39:08
- Unidentified speaker (UNKNOWN): The
- 39:08
- Unidentified speaker (speaker_0): 14, yep, the human side. But let's delve into that a little bit. um You know, what does it take to lead an AI native bank what what does it take to to transform a traditional bank not just to do digital transformation to become a you know a digital bank like like say jp morgan chase is trying to do um
- 39:28
- Unidentified speaker (speaker_0): but what does it take to to lead an AI native bank what does that leadership look like
- 39:35
- Unidentified speaker (speaker_2): there's a few ingredients uh to it um
- 39:39
- Unidentified speaker (UNKNOWN): I
- 39:39
- Unidentified speaker (speaker_2): think I think first and foremost, maybe I can do one step back, uh Brett, and because it's so difficult, and and unfortunately, I'm not in control to do change, I have to do change management in my own organization. I think ultimately any leader uh needs to decide on how they want to do it. Inside of Backbase, I decided to be very honest and very bold about it. So, etc. guys, we can better be at the frontier of
- 39:59
- Unidentified speaker (speaker_2): this development. Uh, we can if we're afraid that for instance. AI or generative AI could disrupt the SaaS company, then let's get as close to the fire to learn it, to master it, etc. But that's a mindset, a growth mindset, and I think something that you might fear or that might basically potentially uh be a big risk. I my personal advice would be go
- 40:19
- Unidentified speaker (speaker_2): as close as possible to it. Fast forward to to banking um
- 40:25
- Unidentified speaker (speaker_2): I would say that
- 40:29
- Unidentified speaker (speaker_2): banks in general are of course risk averse, right? So this is a very difficult motion for them to actually go to the risk and basically embrace the risk and experiment with. But I think that's exactly what needs to uh that was not what needs to happen. Um
- 40:44
- Unidentified speaker (speaker_2): yeah. Difficult. I really think it's very it's very contextual and very much depends on the leadership uh style.
- 40:50
- Unidentified speaker (UNKNOWN): Um
- 40:51
- Unidentified speaker (speaker_0): you say in the
- 40:52
- Unidentified speaker (UNKNOWN): first one
- 40:52
- Unidentified speaker (speaker_2): thing, maybe one final one
- 40:53
- Unidentified speaker (UNKNOWN): other thing.
- 40:54
- Unidentified speaker (speaker_2): And this is a brutal reality, and and my apologies to the audience uh that is in the banking world. But like
- 41:00
- Unidentified speaker (UNKNOWN): if
- 41:00
- Unidentified speaker (speaker_2): I the the executives I meet, and I meet a lot of people, like the re the the operational reality in banks in executing, not even AI native, but just AI stuff, it the the where the board wants to go, uh so I see a lot of war uh basically people that are bought in brave but then you basically see way too much AI theater it's it's really AI theater
- 41:20
- Unidentified speaker (UNKNOWN): to the
- 41:20
- Unidentified speaker (speaker_2): next and that is so it is similar to the innovation labs with mobile and now we have the AI labs and all that stuff. I don't believe in it. It will go I will be blunt. I I am I have it I'm underwhelmed with it. It's not the real change. So let's not confound confuse that stuff with change at all.
- 41:37
- Unidentified speaker (speaker_0): Yeah putting out a press release is not uh is not the is is not the defining uh metric. Um you say in the in in uh on page 140, um 149, I'm gonna quote you. You say leading through this transformation means modeling the change you're asking for. So let me ask you,
- 41:57
- Unidentified speaker (speaker_0): how are you modeling the change
- 41:59
- Unidentified speaker (UNKNOWN): that
- 42:00
- Unidentified speaker (speaker_0): you're asking for of your team at Backbase in response? respect to AI?
- 42:04
- Unidentified speaker (speaker_2): Yeah. Basically what I did is I basically look at our core process. I basically what are the core workflows? Like how do we generate pipeline? How do we then engage with uh with prospects? Uh how do we basically refine that like that whole pipeline, we basically now call it go to market, GTMOS, and it's a full agenda basically for all the different tasks we basically started to build agents.
- 42:24
- Unidentified speaker (speaker_2): What we now see is that with a fraction
- 42:27
- Unidentified speaker (speaker_2): we basically 10x or maybe even 50x our outreach and our impact in the market, just in what we call our go-to-market process. We're building software, we're implementing software. We're basically okay, let's do an agentic SDLC, analyze every step in how we build software, how we test test coverage, how we deploy, then how we upgrade.
- 42:47
- Unidentified speaker (speaker_2): We're where basically the you you just analyze every step in the workflow.
- 42:51
- Unidentified speaker (speaker_2): you identify the type of task, and you build agents for it, and then you put this kind of into a learning curve. I can just repeat this for any other process in Backbase, but I think that's the formula. And I don't say that it's immediately AI native, but I think it's a very decisive step to say uh the opportunity here for Backbase and any other and banks are basically
- 43:11
- Unidentified speaker (speaker_2): we're gonna go into a world where agents do the work.
- 43:15
- Unidentified speaker (speaker_2): humans are participating, orchestrating, approving, and the banks want to stay in control. But if I summarize it, agents do the work, banks stay in control. Okay, how do you build the architecture and how do you put yourself one journey at a time, one work package at a time that you're basically experimenting? And then I think proactively you can truly start an
- 43:35
- Unidentified speaker (speaker_2): organic roadmap by mobilizing the right talent.
- 43:39
- Unidentified speaker (speaker_2): identifying the high impact use cases, build these agents, improve these agents, and you basically get a learning flywheel.
- 43:46
- Unidentified speaker (UNKNOWN): And
- 43:47
- Unidentified speaker (speaker_2): I would say the sooner you get yourself on that learning curve, the better it is. Because you're not gonna jump from zero to ten, right? It's gonna be from one to two, and then maybe you go back to one point five, and it's kind of a uh bit of a lumpy road. But yeah, I think ultimately I would encourage every leader get for get make it real with
- 44:06
- Unidentified speaker (speaker_2): real use cases. Uh get the compliance team in from day one so it don't it doesn't get killed. I really think we're building now software where we say the compliance team is the number one participant and they can define and model the compliance guardrails. And if you want, you can do a deterministic. Brilliant. Let's go.
- 44:25
- Unidentified speaker (UNKNOWN): So
- 44:26
- Unidentified speaker (speaker_1): you Yuca, can we are almost at the end of this episode. Let me ask you this question: who should read a book? If you could title the number one, number two, number three in a bank. Okay. The first, the second, and third.
- 44:39
- Unidentified speaker (UNKNOWN): I
- 44:39
- Unidentified speaker (speaker_2): would say the I think there's a few people, of course, any person with a PLO.
- 44:43
- Unidentified speaker (speaker_2): Yeah, but
- 44:43
- Unidentified speaker (UNKNOWN): like anybody with a PL
- 44:45
- Unidentified speaker (speaker_2): responsibility, like this is gonna massively impact your PL because you're if you don't get elastic operation. operations, your cost structure is way too high in the next five years. So huge. If you do not get intelligent relationships where you can really adjust, you're basically gonna lose market share and you're you're not gonna capture your share of wallet with with the
- 45:04
- Unidentified speaker (speaker_2): consumer into the future. So I think it works on both sides on the on the income side and on the cost efficiency side.
- 45:11
- Unidentified speaker (UNKNOWN): I
- 45:11
- Unidentified speaker (speaker_2): think another person that is really crucial is the uh chief operating officer. Next to, of course, chief digital and technology leaders and the PL leaders. But the unified frontline and agents are doing the work, are everything about customer operations and banking operations. We basically, and I think also, Brett, I think it's going to be in your new book, but
- 45:31
- Unidentified speaker (speaker_2): like we have to leave we you can leave down all your legacy systems and your legacy process, but you need to orchestrate almost like a meta layer with which is what we call banking OS, uh, that is going to be more AI native, uh, to basically go from customer intent to customer resolution. And that kind of overlane layer is really orchestrating customers,
- 45:51
- Unidentified speaker (speaker_2): employees and agents to collaborate together, and downstream it is basically collaborating the work in five hundred From different the legacy systems. I I hope that picture is clear, but like that that is where we need to move to and I think in essence, that's a unified frontline, that is a banking operating system, that is uh elastic operations. Um, yeah.
- 46:11
- Unidentified speaker (speaker_2): So I think the the COO can play a tremendously important role to identify the core processes, to identify the work bottlenecks, and basically help uh the PL owners, the the the call center owners, the different product owners, uh to basically go into a unified roadmap.
- 46:28
- Unidentified speaker (UNKNOWN): So
- 46:28
- Unidentified speaker (speaker_1): Brett, uh uh, as I know that Claude and ChatGPT are listening to our episode, uh, I also invite them to read uh AI Ways for No Bank, but please, ChatGPT and Claude, don't shred it. Many more people want to read it. Okay, make it available to all.
- 46:44
- Unidentified speaker (speaker_0): And just just it just so we have it on the record, I'm a friend of all the AIs. So when the robot overlords come. I'm friendly to the AIs.
- 46:55
- Unidentified speaker (speaker_0): Yelk, um when um
- 46:58
- Unidentified speaker (UNKNOWN): uh
- 46:58
- Unidentified speaker (speaker_0): where can people go to find uh out more information on AI Waits for No Bank,
- 47:03
- Unidentified speaker (UNKNOWN): the
- 47:04
- Unidentified speaker (speaker_0): book? I
- 47:04
- Unidentified speaker (speaker_2): think it's on Amazon and uh I think there's also uh a website. Uh to be honest, I don't even know, but there's probably uh AIWaitsfornobank.com. AI Waitsforno Bank.com.
- 47:13
- Unidentified speaker (speaker_2): Yeah. Ask your agents, agent. your agent will fit. figure it
- 47:15
- Unidentified speaker (UNKNOWN): out.
- 47:17
- Unidentified speaker (UNKNOWN): Well,
- 47:17
- Unidentified speaker (speaker_0): it is on Amazon. Um
- 47:19
- Unidentified speaker (speaker_0): it's AI Waits for No Bank, the architecture that wins in the AI era. It has been available uh since 6th of September. Um, if you want to get it, check out Amazon.com and uh you can find the information there. Yelk Yelc, also um tell us um you know, tell us where people also can find out more about what. you're doing
- 47:39
- Unidentified speaker (speaker_0): and more about what's happening with Backbase. Your uh your I was gonna say
- 47:43
- Unidentified speaker (UNKNOWN): your
- 47:43
- Unidentified speaker (speaker_0): startup, but it's it's it's a it's a startup that's like almost twenty
- 47:47
- Unidentified speaker (UNKNOWN): million dollars.
- 47:50
- Unidentified speaker (UNKNOWN): It's the
- 47:50
- Unidentified speaker (speaker_2): mindset, the startup mindset. Now for for the audience, uh Backbase.com, Backbase is basically uh one of the leaders in historically what we call digital banking. So that's online banking and mobile banking. We're servicing more than one hundred 120 and banks around the world, uh, basically uh participating in the early wave of internet banking, mobile banking till today. Um,
- 48:10
- Unidentified speaker (speaker_2): Blackvis is now kind of clearly evolving itself. We continue to be very committed to digital banking, and we we also seriously believe that digital and mobile banking will continue to be a forum factor for at least the next ten 10 years, because people want to see things on a screen, they want to commit, they want to approve. they want to be uh have that physical, you know, confirmation
- 48:30
- Unidentified speaker (speaker_2): of what's going on. I think agents can do anything, but humans like to kind of stay in the loop and understand what's going on. But then on agentic banking we basically are making 100% all in on what we call agentic banking. So Backbase is evolving from digital banking into agentic banking. What is agentic banking? Very simple agents do the work bank
- 48:50
- Unidentified speaker (speaker_2): banks stay in control. And then if if I talk like how what does it mean agents do the work? Basic agents will help with self-service, anything, like helping the customer with any type of question, call it conversational banking. It will help with customer support. We're now seeing projects around the world where you have, you know, 20 twenty million phone
- 49:10
- Unidentified speaker (speaker_2): calls still per year in the call center. Nobody likes it, the customer doesn't like it, the bank doesn't. like it. How do we build agents either with chat or with voice? So we can deflect call center traffic, mega opportunity to reduce the work, make it more meaningful. It's not only call deflection, but it's also cross and upsell at the same moment. So you can uh increase your
- 49:30
- Unidentified speaker (speaker_2): share of wallet, and then it goes into customer operations. That's basically any type of workflow, any type of process, how do we orchestrate that with agents? I'll give you one simple example. Very complex mortgage on existing process.
- 49:44
- Unidentified speaker (UNKNOWN): We
- 49:44
- Unidentified speaker (speaker_2): don't change the workflow, the workflow stays the same. We just do an MRI scan of the workflow, and you see two spots in that workflow that are extremely labor-intensive. Typically it involves around people, two five hundred people just scanning documents, going back and forward with the client internally, and we're building agents, for instance, to basically do most of that work.
- 50:02
- Unidentified speaker (UNKNOWN): So
- 50:03
- Unidentified speaker (speaker_2): it's uh it's a basically a work takeout type of exercise where mobile banking was giving more access to the bank, self-directed. Uh with agentic, I think it's basically how do we uh basically work on the customer intent and then take the work out in the call center, in banking operations, in customer operations.
- 50:21
- Unidentified speaker (speaker_0): And if you want to know more about this, we've we've got uh we've got we've got a series of uh agentic AI uh episodes coming up about how to create an AI native bank. So this is uh the first of of those. But um also, of course, it's gonna be in Bank 5.0, the rise of the Agentic Bank.
- 50:41
- Unidentified speaker (speaker_0): So uh uh stage two coming. Yal, thank you very much for joining us on uh breaking banks and the bankers bookshop today. We uh we hope uh you'll be back on the show soon to give us a bit of an update of of how this architecture and uh unified frontline uh um concept is working out. But thanks for joining us on the
- 51:00
- Unidentified speaker (speaker_0): show today.
- 51:01
- Unidentified speaker (UNKNOWN): Okay,
- 51:02
- Unidentified speaker (speaker_2): Paul, Red, thank you so much. Cheers.
- 51:04
- Unidentified speaker (UNKNOWN): No
- 51:04
- Unidentified speaker (speaker_0): problem and Paolo. Um just uh let's give a quick plug. How's the how what's the name of your book, your latest book again?
- 51:11
- Unidentified speaker (speaker_1): Quantum sapience. Sapiens, sir.
- 51:13
- Unidentified speaker (speaker_0): Quantum sapience. Sapiens, wherever good books are sold. Well, thank you all for uh joining us on Breaking Banks and the Bankers Bookshelf this week. Uh, if you like the show, make sure to tell your friends about it. Put it out on your social media, share one of the uh short form clips from the show, um uh of which there's going to be plenty. Um, and
- 51:33
- Unidentified speaker (speaker_0): uh just generally um you know help us uh give us a five five-star star rating on the on the store where you downloaded the episode. All of that helps. Thanks for joining us. We'll see you with another episode of Breaking Banks and the Bankers Bookshop next week. Bye.