Delto proposes an agentic Skill-based architecture for banking: a single AI orchestrates 300+ pre-built banking Skills (the Banking Large Action Model) instead of multiple isolated bots. The result: greater scalability, security and traceability. McKinsey estimates generative and action AI could add USD 200 to 340 billion a year to global banking.
Financial institutions are at the forefront of innovation, and conversational artificial intelligence has become a foundational pillar. We have seen many banks explore banking chatbots or virtual assistants, trying to solve everything from customer service to collections management. Often, this means training multiple agents in isolation or integrating external tools for each new feature. Yet this traditional path carries significant limitations in scalability, maintainability and control. At Delto we built an approach that transforms this landscape: a platform based on Agentic Skills (intelligent skills), designed specifically for the needs of digital banking . In this article, we will explore why this innovative approach is more scalable, secure, maintainable and analytical than conventional solutions. We will also dig into what a "Skill" means in Delto's architecture, how our model translates into real solutions (from collections agents to digital onboarding), and the technical advantages over the "tool-based" approach. Find out why Delto's Banking Large Action Model , with more than 300 pre-built banking Skills, is the key to the next generation of conversational automation in banking . From the multi-agent approach to Skill-based: the necessary evolution in AI for banks Deploying several isolated agents or bots for different functions brings significant challenges: each agent requires specific training, dedicated infrastructure and individual maintenance. This can create information silos and inconsistent customer experiences. A Skill-based approach, by contrast, consolidates capabilities into a unified platform. Why is this evolution so advantageous for the banking industry in Latin America? Simpler scalability: Skills act as reusable modules. Adding a new use case no longer means building another bot from scratch, but rather incorporating or activating the corresponding skill within the same system. Once the platform is deployed, scaling to new processes or products is far more agile and at a lower marginal cost. A centralized platform can also handle volume peaks by efficiently distributing resources across Skills, whereas a multi-bot scheme might require over-provisioning each one for its own peaks. This capability is crucial for the future of banking. Maintainability and efficiency: With a single AI managing multiple Skills, duplication drops. Updates (for example, improving language understanding) happen in one place and benefit every capability. This simplifies operations and lowers maintenance costs. Skills also encourage reuse: a user authentication or balance-check skill can be invoked in different contexts without re-implementing it each time. It is a solution for banks looking to optimize resources. Unified security and control: A bank must ensure every automated interaction complies with strict security and regulatory policies. In a scattered traditional approach, guaranteeing homogeneous standards across multiple agents is complex. With centralized Skills, it is easier to apply cross-cutting security controls (authentication, authorizations, action auditing) within a single framework. It also minimizes the exposure of sensitive data to multiple external systems; the intelligence operates inside an environment controlled by the bank. For example, Delto runs on enterprise-grade cloud infrastructure (Azure), with strict data governance, ensuring privacy and regulatory compliance across all operations. Superior analytics: By concentrating all interactions on a single platform, the bank gains 360 visibility into what is happening. It becomes possible to analyze conversations from different domains together, identify trends, common friction points and improvement opportunities. Delto, for instance, includes native AI-powered analytics to monitor every interaction in real time, identifying what customers ask, how each Skill responds and measuring effectiveness, which enables data-driven continuous improvement. With multiple isolated bots, that aggregation and global learning would be far harder. This is key to the banking customer experience. In short, the Skill-based approach offers a scalable, consistent and governable foundation for conversational automation in banking , versus the fragmentation of traditional approaches. What is a Skill and how does it work in Delto's architecture? In Delto's platform, a Skill is an encapsulated conversational capability that solves a specific user need. We can think of each Skill as a "mini-agent" specialized in one task, but unlike traditional bots, it does not operate in isolation: it is coordinated within a central AI brain. A Skill typically includes: Detection of relevant intent and context (for example, identifying that the user wants to check the balance of their checking account). The business logic or action defined to serve that intent (e.g., running the balance query against the bank's core systems). Generating the response to the customer with the information or result obtained, in natural language. When a user asks a question or makes a request through any channel (web chat, WhatsApp, voice or another), Delto's architecture interprets the input and decides which Skill (or combination of Skills) to apply. This intelligent orchestration is handled by the Delto Intelligence module, the cognitive core of the system. That module uses state-of-the-art language models (Large Language Models) adapted to the banking domain to understand the customer's request and plan a response. If the response requires a concrete action (for example, "Transfer $500 to my savings account" involves checking the balance, debiting the account and generating confirmation), Delto Intelligence will activate the relevant Skills (balance check, transfer execution, etc.), each with its own validations and predefined steps. In essence, the AI model acts as an orchestrator, selecting and chaining skills to fulfill the user's request autonomously. Within Delto's component architecture, Skills sit at the functional heart: Channels: Our platform is omnichannel. Messages can come from different channels (the bank's mobile app, website, WhatsApp, Messenger, phone IVR, etc.) and all converge into the same system. Delto integrates easily with the bank's existing channels, leveraging APIs or native connectors, so the customer gets a smooth, uniform experience regardless of the contact medium. Pre-Processor: Before reaching the intelligence core, each interaction passes through a pre-processor. This component performs initial tasks such as normalizing the input (for example, transcribing voice to text if the interaction is spoken), detecting the language, and extracting relevant information (identifying the customer from the phone number or session ID, recognizing entities mentioned such as account numbers, dates, amounts, etc.). It can also filter content or apply simple business rules. The goal is to prepare the right context so the AI makes better decisions. Delto Intelligence: This is where the cognitive and decision-making engine lives. Built on generative AI , Delto Intelligence understands the user's intent from the text (or the voice transcription) and plans the response. Unlike a simple FAQ chatbot, this module can reason and break complex requests into actionable steps. If it needs data or operations, it will invoke the corresponding Skills in Delto Core. This component ensures the conversation follows a coherent flow, handling context (for example, remembering what was said earlier in the session) and maintaining the right tone and language. Delto Intelligence can be thought of as the "brain" that understands and decides, using Skills as its "hands" to execute actions. Delto Core: This is the execution core and Skill library. Here live the 300-plus pre-built banking skills along with any custom-built ones. Each Skill in Delto Core has the logic to interact with the bank's systems or with authorized external services (for example, a bill-payment Skill knows how to format the order and communicate with the bank's payments API). When Delto Intelligence invokes a skill, Delto Core executes the corresponding code securely and at scale, obtaining results (queries, transactions, reports, etc.). It also handles integration with legacy systems: for instance, connectivity with the core banking system, CRM, ERP or other tools, through standardized APIs or connectors already included in the platform. After the Skill runs, Delto Core sends the return information to Delto Intelligence so it can craft the final response to the user. Post-processing and learning: Finally, the generated response travels back through the channel to the customer. The platform can include post-processing steps such as masking sensitive data in the response or translating into the user's preferred language if needed. Every interaction is logged for analysis, feeding analytics modules that learn from successes and errors. Delto includes an AI Analytics module that monitors these conversations to identify optimization opportunities (for example, detecting that a Skill needs additional training because users often phrase the intent in an unexpected way). This modular design ensures each component can scale and innovate independently (for example, adding a new channel without changing the core logic, or improving a Skill without retraining the entire model). In particular, the extensive catalog of pre-built Skills, Delto's Banking Large Action Model , is a key differentiator. Let's look at what it consists of and how it applies to concrete banking automation use cases. Banking Large Action Model: 300+ banking Skills ready to use Unlike generic conversational AI solutions, which usually start from an empty base or require training an agent for each task, Delto provides more than 300 pre-configured banking Skills. These skills cover a broad range of operations typical of financial institutions, from informational queries to complex transactions, and are designed specifically for banking. In practice, this means a project can start with much of the work already done: the core capabilities (checking balances, transferring funds, blocking a card, quoting a loan, collecting a debt, etc.) already exist and have been tested. The bank's team only needs to adjust integration details (for example, mapping the "transaction history" Skill to its specific core banking API) and customize certain behaviors or vocabulary to fit its operations. The Banking Large Action Model (BLAM = Banking Large Action Model) is not simply a static list of functions, but a set of skills backed by sector knowledge and best practices. Each Skill comes with built-in compliance considerations (for example, amount limits, mandatory identity verification before certain operations, messages pre-approved by the legal team for specific sensitive responses). This accelerates implementation while giving the bank's Risk and Compliance teams peace of mind, since automated interactions stay within the bounds set by financial regulation. To learn more about our specific banking solutions, visit our solutions for banks page. From a technical standpoint, having a broad catalog of pre-built Skills is similar to having a library of high-level functions: instead of coding the logic of each new feature from scratch, you reuse and assemble what already exists. This also reduces errors, because you start from components already tested in real scenarios. Of course, the platform is extensible: if the bank needs a completely new skill (for example, an integration with an uncommon external service), Delto offers a development Framework where programmers can create custom Skills in Python, with supporting tools such as code copilots and integrated test environments. Once developed, the new Skill is added to the Core and becomes available to be orchestrated by the intelligence alongside the rest. Banking solution use cases with Skills: automation for banks in action This approach takes shape in concrete solutions inside a bank, going beyond the traditional view: Personal banking and a conversational agent for individuals: A Skill-based assistant can cover the retail customer's financial life end to end without fragmenting the experience. A single conversation moves, for example, from "What is my card balance?" to "Block it because I lost it" or "Pay my electricity bill" without channel jumps. Behind the scenes, Delto Intelligence chains specialized Skills: identity verification, balance checks, card blocking, payment execution, limit adjustments or certificate generation. If the user asks "Should I take a 90-day term deposit or invest in funds?", the AI activates recommendation skills based on profile and risk, explains the pros and cons and offers to open the product right away. The customer never leaves the conversation. The bank gains fine-grained metrics on which contact reasons dominate, resolution times and cross-sell opportunities, while reducing operating costs by automating routine tasks that previously required human agents. Corporate banking and agents for businesses: Corporate clients tend to have more complex queries than an individual user, and to carry out larger operations (for example, treasury management, mass payments, foreign trade queries). A multi-bot approach might mean one assistant for payments, another for trade finance queries, another for technical support, etc., fragmenting the interaction. With the Skill-based platform, a single corporate agent can handle multiple topics: one conversation can go from "What is my consolidated balance today?" to "Quote me the exchange rate to transfer USD 50,000 abroad" or "I need help loading the payroll." Behind the scenes, different specialized Skills serve each request (querying corporate account balances, quoting a preferential exchange rate, executing mass payments integrated with the ERP, etc.), but to the user it is the same intelligent assistant that understands their context (it knows which company it is, which products are contracted, what their last operations were) and never hands them off from one department to another. This greatly improves the corporate customer experience while reducing operating costs, by automating tasks that previously required dedicated officers. The platform ensures that, despite the complexity, every step is validated (for example, multiple authorizations for high amounts, compliance with anti-money-laundering policies) since those rules are embedded in the relevant skills. Digital customer onboarding (product enrollment): Enrolling a new customer or selling them an additional product typically involves collecting data, running assessments (credit scoring, KYC/AML), signing contracts, etc. Traditionally, this is handled with separate web forms or with executives guiding the user step by step. A Skill-based conversational agent can orchestrate the entire onboarding flow naturally through chat. For example, a new customer starts by asking "I want to open a checking account." The assistant requests data through registration Skills (capturing personal data, photographing a document, validating against regulatory lists). If mid-process the user asks "What are the benefits of this account?", the same agent can activate an informational Skill that answers that question in clear language, leveraging the integrated knowledge base, and then resume the opening process. Once the data is gathered and verified, another Skill generates the digital contract and presents it for acceptance. All without leaving the conversational channel. The interaction is proactive and guided, with the AI anticipating common doubts and offering contextual help at each step (for example, if the user is slow to respond during the ID upload stage, the bot can offer assistance or alternatives). This dramatically reduces friction when onboarding new customers. Operationally, the bank gains higher conversion and can track at which steps users drop off to optimize the funnel, thanks to the fine-grained traceability the platform provides. Technical advantages over the traditional "tool-based" approach: why is Delto the best AI solution for banks? Beyond experience and scalability, the Skill-based architecture delivers significant technical and operational benefits compared with approaches based on external tools or multiple agents. Some key differentiators: Optimized use of resources and costs: A single AI orchestrating Skills consumes fewer resources than maintaining several independent AI engines. The Delto platform leverages generative models cost-efficiently, combining its own instances and public services as appropriate. Instead of making multiple calls to external APIs or spinning up several processes per query, many decisions are resolved internally with local Skills. This reduces latency and third-party API costs. The infrastructure also scales on demand (especially being cloud-based), so the bank pays only for what it uses, absorbing interaction peaks without high fixed investments. Faster response times: With Skills integrated into the same platform close to the cognitive engine, a lot of orchestration "dead time" is eliminated. Latency drops because the AI does not have to wait for an external service to process something synchronously at each step; many actions happen almost in real time within Delto Core itself. Even when there are integrations with external systems (e.g., an internal database), these are optimized through persistent connections and direct calls from the Skills, avoiding the overhead of going through intermediate layers. The result is smoother conversations, where the user perceives immediate responses, even in complex operations. This contrasts with some external-tool approaches where each LLM query plus a separate API query can introduce noticeable delays. Built-in security and compliance: In financial environments, confidentiality and accuracy are paramount. An approach based on generic tools could involve sending sensitive data to external services or losing control over how certain operations are handled. With internal Skills, all data stays within the bank's secure environment (or its private cloud) during processing. Delto, for example, hosts the solution on Azure under strict data policies and banking privacy agreements. Likewise, each Skill can keep detailed audit records (what was queried, who approved a given operation, etc.), easing traceability and regulatory compliance. The platform can anonymize or mask information when appropriate, ensuring the AI never reveals unauthorized information. Centralized consistency makes it easier to apply urgent security updates in a single place, rather than patching multiple disparate systems. Ease of integrating new technologies: The AI world moves fast; new models, services and libraries appear constantly. In a traditional approach, adopting a new technology (for example, changing the bot's NLU engine, or adding a sentiment-analysis tool) would require updating agent by agent or developing ad-hoc integrations with each system. In Delto's architecture, components are decoupled and standardized, which lets new pieces fit in with less effort. For example: if tomorrow a more powerful or Spanish-specific language model appears, Delto Intelligence could integrate or replace it without the Skills underneath changing; or if a useful new open-banking API emerges, a Skill can be developed to consume it and immediately all conversations could leverage that capability. This evolutionary flexibility protects the bank's investment, since the platform can incorporate innovations without rebuilding the solution from scratch. Delto includes a continuous-improvement methodology (Skill Growth Strategy) to detect the next skill or adjustment to implement as needs change. Context curation and accuracy: A common complaint about implementations based solely on generic LLMs with tool access is the tendency to produce hallucinated or off-context responses when the model misinterprets the question. The Skill approach mitigates this by defining specialized contexts for each skill. That is, when the user asks something that activates the investment-recommendation Skill, say, the AI works with the specific data and rules of that skill (for example, it will only suggest permitted products based on the customer's profile). The conversation is bounded to the relevant context, reducing room for the language model to wander. In addition, Skills act as "deterministic functions" in many cases: if an installment-calculation Skill receives an amount and a term, it will always return the correct result according to the formula, regardless of how the question was phrased. This combination of generative AI plus specific logic delivers the best of both worlds: naturalness and broad understanding, but with precision and control in executing responses. Full traceability of interactions: In a multi-agent scheme or with scattered tools, tracking what was said or done can be complex, with data fragmented across different logs. Delto, by contrast, centralizes the record: every conversational turn is tied to a customer, a session and an applied Skill, with timestamps and results. This eases not only technical debugging (knowing what happened if something goes wrong) but also obtaining operational insights. For example, you can trace the full path of a transaction started via chat, see where the user may have gotten stuck, how many times a human intervened, etc. The transparency this traceability provides increases trust in the AI: the bank's executives can audit interactions and verify that the agent follows the defined policies. It also helps train the models better, with supervised feedback based on real histories, and demonstrate compliance to regulators by presenting clear logs of every automated interaction. The future of conversational automation in banking: are you ready for the digital banking transformation? Adopting Skill-based conversational agents represents a paradigm shift in how banks approach automation and customer experience. We are moving from static bots that answer questions to intelligent agents that can reason and execute actions on the user's behalf, integrating deeply with banking processes. This advance aligns with the trend of Large Action Models (LAMs), where AI not only understands language but takes concrete actions. A McKinsey report estimates that generative AI and action AI could add between $200 and $340 billion a year in value to the global banking sector , reflecting the enormous efficiency and personalization potential still to be captured. In the near future, it is plausible to imagine banking operations that are almost fully autonomous for routine tasks: from approving a simple loan in minutes to resolving a complaint without human intervention, with the virtual agent coordinating the necessary checks and decisions. But beyond efficiency, the focus will be on personalization and proactivity. Thanks to Skill-centric architectures and integrated analytics, banks will be able to anticipate customer needs, for example, proactively recommending financial optimizations or flagging deviations, contextually and on time. Channel and experience executives will find in these platforms a lever to raise customer satisfaction, by delivering instant, consistent responses available 24/7, without losing the human touch. In fact, an AI well trained in skills can achieve interactions so natural that the customer feels the bank "knows them" and advises them as a personal manager would, but with the speed of automation. Of course, this future comes with challenges: it will be vital to keep a balance between AI autonomy and human oversight. This is where approaches like Delto's show their strength, by enabling centralized governance of skills and transparency in the decisions made. Banks that adopt these Skill methodologies early will be better positioned to adapt to new AI regulations, integrate emerging technologies (such as advanced biometric identification, sentiment analysis, etc.) and scale automation coherently across all their operations. Institutions that seize this opportunity will not only reduce operating costs, but redefine how they relate to their customers in the digital era, delivering intelligent service at scale, without losing the warmth and trust that characterize banking done well. The path is already laid out: platforms like Delto show that it is possible to achieve today what a few years ago looked like science fiction, setting the foundations for a new standard in tomorrow's banking experience
What is a Skill in Delto's architecture? A Skill is an encapsulated conversational capability that solves a specific customer need: it detects intent, runs the business logic (for example, checking a balance or executing a transfer) and generates the response in natural language. It works as a specialized mini-agent, coordinated by Delto Intelligence within a single AI brain, rather than in isolation like a traditional bot.
What is Delto's Banking Large Action Model (BLAM)? The Banking Large Action Model (BLAM) is Delto's catalog of 300+ pre-configured banking Skills, from checking balances to collecting a debt. Each Skill ships with built-in compliance considerations (amount limits, identity verification, legally pre-approved messages), so the bank starts with much of the work already done and tested in real-world scenarios.
Why does the Skill-based approach beat the traditional tool-based one? A single AI orchestrating Skills consumes fewer resources than several isolated engines, cuts latency and third-party API costs, and keeps data inside the bank's secure environment. It also centralizes traceability: every turn is tied to a customer, session and Skill, which simplifies auditing, regulatory compliance and continuous improvement.