Actionable Decision Intelligence in Banking: 3 Need-to-Know Accelerating Trends in AI

In early 2024, Curinos proprietary research had revealed either no adoption of enterprise-wide AI applications or an extremely cautionary posture among banks. It now shows that more than 85% of institutions have undergone some level of AI adoption. Why such rapid deployment? We believe it can be explained by three layers of transformative impact, all of which are accelerating: the increasing user accessibility of enterprise-wide software; agents enabling a much more advanced user experience; and decision intelligence systems that orchestrate and execute against long-term business goals.  

Impact #1: Access to insights and analytics in the user experience of enterprise software is being rapidly opened up.  

We see a near-immediate future in which an increasingly broad set of users will interact with enterprise software. Data will increasingly be converted, accurately and securely, into action for non-analyst managers and executives by, nearly instantly, offering complex analytics, data visualization and workflow orchestration, and outcome measurement. This is in marked contrast to enterprise software that has historically required specific controls and configuration parameters, together with data-driven metrics with nuanced definitions. 

Credit the introduction of large language models (LLM), which have revolutionized the interface between human language and human-driven technical concepts like programming languages, web-services API interactions and data formats. By the end of 2026, we expect that enterprise software products will exhibit conversational or “chat” interfaces by default. They’ll allow the conversational creation of data visualizations and highly interactive and flexible multimodal incorporation of insights directly into business output. Graphics, text, tabular data and animations will automatically populate financial models, progress reports and presentations. 

Initially, we expect these capabilities to sit alongside conventional reporting and dashboard interfaces. But ultimately they’ll fully replace them with application real estate that’s consumed with more visual expressions of data and with system workflows, dataflows and customer behavioral signals. Both the volume and speed of insights available to drive decision-making at all levels of the bank will increase, with an associated improvement in bank performance. 

Impact #2: Conversational systems will sit on top of a series of agents and drive action through workflows, software configuration and orchestration that enable a more advanced user experience.  

As insights and analytics open up to non-analytics individuals, we anticipate a wave of increased sophistication in what can be accomplished by highly trained subject-matter-expert analysts, economists and data scientists. This is a transformation from what has been the state-of-the-art decision intelligence systems of the past three to five years. In those systems, intricate orchestration workflows and experience sequences have been burdened by the need for complex human-rule configuration, and even software coding, to implement and maintain them. This has limited not only the impact, volume and complexity of decision experiences but also the number of contexts in which they can be applied. It’s not uncommon, for example, for large banks to have teams of 100 or more working on experience configuration, testing and management. 

Going forward, we expect the rise of agentic AI capabilities to rapidly bypass the implementation phase of such experiences and allow human-in-the-loop review and testing for compliance and stability, and for ongoing review and maintenance. It will be facilitated by agents that will: learn to configure and code experiences based on domain-specific prompting by a human subject matter expert; learn to interrogate configurations and code for errors and misconfigurations; and honor regulatory constraints that are preconfigured or expressed by human operators. Today, CurinosCopilot, for example, delivers compliant executive-ready insights in near-real time from forecasting and pricing—thereby streamlining how pricing strategy is developed, communicated and deployed across teams.  

In the banking space, such agents can be informed by the same deterministic or stochastic methodologies that are in place today in industry-leading solutions. Because they could be subject to the same model risk management review and interpretability requirements already common in the industry, we foresee banking to be able to adopt such capabilities rapidly – within the next two to three years.  

Similarly, in the components of the decision intelligence workflow pertaining to decision content and creation, agents can be used to create, review, judge and provide guardrails for content. These can include brand and product styling, tone-of-voice and value proposition constraints, again with humans in the loop for final review and compliance approval. The outcome is a step-function change in the volume and impact of high-quality decision contexts and experiences that can be managed and acted upon simultaneously within the bank. 

Impact #3: Decision intelligence systems that combine generation and learning agents to orchestrate and execute compliant decision-making will move long-term business outcomes.  

The next frontier for decision intelligence is the emergence of autonomous decision-making systems that integrate generative AI with reinforcement learning to optimize long-range business outcomes. This is an area that Curinos has pioneered, and with our most-recently issued U.S. patent introduces a novel capability: it enables AI agents to make decisions that maximize expected value over extended time horizons, while operating within a framework of human-defined guardrails. These guardrails ensure that decisions remain contextually appropriate, both aligned with brand and compliant with regulatory and ethical standards. 

This paradigm shift will make possible dynamic, adaptive decision orchestration. Coupled with reinforcement learning, generative models continuously learn from the outcomes of their actions – adjusting targeting, timing, channel selection and content generation to maximize customer value over time. As is the case with Curinos’ Amplero Personalization Optimizer, these agents are not merely executing predefined workflows; they’re actively shaping them by learning which combinations of actions yield the highest engagement, conversion or financial impact across diverse market conditions and customer segments that are dynamically learned rather than statically predefined.  

The implications for banking are profound. Imagine a system that autonomously tailors product pricing and incentive offers, timing of outreach and messaging style for each customer, learning from every interaction to improve future decisions. These agents can simulate long-term customer journeys, optimizing for metrics such as lifetime value, retention and cross-sell effectiveness. Critically important is that human oversight remains central. Domain experts define the strategic boundaries – such as acceptable targeting criteria, tone of voice and compliance constraints – within which the agents operate.  This is critical in the setting where customer-segment growth widens focus beyond the contemporary emphasis on “mass affluent” and into identifying high-future-value customers in the “mass market,” exponentially increasing the audience targeting, offering and engagement channel complexity to generate impactful experiences. 

In the setting of engagement and lifecycle marketing in digital channels, using Amplero, one bank realized 23% growth in incremental new deposit accounts in less than six months, and more than $900M in incremental new balances (Figure 1). Fully 45% of that balance growth was new to bank, so the communications not only improved customer and balance retention but also attracted significant new-to-bank deposits.