For years, large banks have had an advantage in customer analytics because of their scale, technology budgets and access to sophisticated data infrastructure. A new Celent case study involving Union Savings Bank suggests that advantage may be narrowing. The Connecticut community bank is using predictive AI to identify customers at risk of leaving, recommend retention actions and uncover opportunities to deepen relationships—without replacing its core banking system or building a new data lake.
The most valuable application of AI in retail banking may not be a chatbot.
It may be knowing which customer is about to move their money—and understanding what the bank should do before that happens.
That is the approach highlighted in a new Celent Solution Brief examining how Union Savings Bank (USB) uses NGDATA’s AI Hub and Intelligent Engagement Platform to improve customer retention, protect deposits and increase Customer Lifetime Value.
The case is particularly relevant for community and regional banks.
These institutions often have years or decades of customer and transaction data, but turning that information into real-time intelligence can be difficult when it is spread across core banking systems, account databases and other applications.
USB’s strategy takes a different route. Instead of replacing its existing infrastructure, the $3.3 billion community bank is using NGDATA’s technology to activate information already available across its systems and apply predictive models to customer relationships.
The goal is straightforward: identify customers showing signs of attrition, determine why they may be at risk and recommend an action while there is still time to influence the relationship.
That moves AI from analytics into decision-making.
From churn prediction to next-best action
Traditional customer analytics can tell a bank that a customer is likely to leave.
That is useful, but incomplete.
A retention team still needs to know what to do next.
USB’s implementation combines an attrition model with next-best-product recommendations and targeted customer journeys. When a risk signal appears, the bank can initiate an outreach strategy designed around the customer’s circumstances rather than sending the same retention offer to every account holder.
According to the Celent brief, USB achieved a 74% churn catch rate, while its next-best-product recommendations reached 79% Top-3 accuracy for retention.
The bank also reported 90% retention in targeted Certificate of Deposit renewal campaigns.
Those are results reported in the Celent/NGDATA case study rather than independent benchmarks for the banking industry, so they should be viewed as evidence from a specific deployment rather than a general expectation for AI-powered retention.
Still, the underlying strategy aligns with a broader direction in banking.
McKinsey has identified machine learning-driven churn prediction and personalized deposit offers as important tools for banks seeking to protect deposits. In one banking example examined by the consultancy, personalized term-deposit campaigns more than doubled additional funds brought in compared with untargeted campaigns at the same average cost of funds.
The economic logic is becoming more compelling.
Deposits are not passive assets. Customers can move them rapidly when rates change, promotional periods expire or competing institutions offer more attractive products.
And AI may make that behavior even more fluid.
McKinsey estimates that $23 trillion of the world’s $70 trillion in consumer deposits sits in checking accounts earning very little interest. As consumers increasingly use AI agents to compare financial products and optimize their money, banks could face greater pressure to defend customer relationships.
For community banks, that makes customer intelligence a strategic capability rather than simply a marketing function.
The data-lake problem
One of the more interesting elements of USB’s deployment is what it does not require.
According to NGDATA and the Celent case study, USB did not have to replace its core banking platform or build a separate enterprise data lake to deploy the customer-intelligence capabilities.
That matters because large-scale data modernization can become a multiyear undertaking.
Banks have accumulated information across deposits, transactions, products, interactions and customer accounts. But moving all of that data into a new architecture before delivering any business value can create a significant barrier to AI adoption.
NGDATA’s approach is to put its Intelligent Engagement Platform within the bank’s existing environment and use AI Hub to apply predictive models to activated customer data.
The architecture effectively separates data activation from core replacement.
That could make AI adoption more accessible to smaller institutions that do not have the technology budgets or engineering teams of global banks.
Gartner’s 2026 research on banking customer experience identifies siloed data and legacy-system limitations as major obstacles to anticipating customer needs and delivering personalized experiences. The research recommends technology strategies that illuminate customer journeys and support personalization.
USB’s case study is therefore relevant to a much larger banking technology problem.
The question is not simply whether banks possess customer data.
Most do.
The question is whether they can turn that data into an actionable decision quickly enough to affect a customer’s behavior.
Explainability becomes important
There is another element that becomes increasingly important as AI moves closer to customer decisions: explainability.
A churn score by itself is difficult for a relationship manager to act on.
If the system can identify the factors contributing to the prediction, however, a banker has more context for deciding whether an intervention makes sense.
NGDATA says AI Hub provides explanations of the factors behind individual churn predictions in plain language.
That distinction matters in banking because customer retention decisions can affect pricing, product recommendations and the treatment of individual customers.
AI systems that simply generate scores can create a black-box problem. Systems that provide context around those predictions can give employees a basis for reviewing and challenging the recommendation.
McKinsey has similarly emphasized that AI-powered banking decision systems need explainability and strong model governance as AI increasingly influences pricing, credit and other customer decisions.
Why deposits are becoming an AI battleground
Deposit retention may become one of the most important tests of AI-powered personalization.
Customers historically tolerated considerable inertia. Moving money between banks required effort, research and a willingness to change established relationships.
Generative AI and financial agents could reduce that friction.
A customer may increasingly ask an AI system to monitor rates, compare accounts and recommend where idle cash should go. McKinsey warns that such agentic behavior could disrupt deposits by reducing the inertia that historically benefited incumbent banks.
That changes the competitive equation.
Banks can no longer assume that a customer relationship will remain stable simply because the customer has used the institution for years.
The bank needs to understand changing behavior and respond with relevant products, pricing and communication.
That is where predictive customer intelligence can become valuable.
Competing with bigger banks through data
The most important implication of the USB example may be its relevance to the broader community-banking market.
Smaller banks do not necessarily need to replicate the technology architecture of JPMorgan Chase, Bank of America or other global institutions to compete on personalization.
They need to extract more value from the information they already possess.
That creates an emerging category between conventional CRM and full-scale core modernization: AI-driven customer intelligence layered onto existing banking infrastructure.
NGDATA is competing in that space alongside broader customer-engagement, decisioning and personalization platforms. Larger technology providers—including Salesforce, Microsoft, Adobe and cloud AI platforms—also offer components that can support customer analytics and personalization, while specialized banking technology vendors focus on decisioning, customer data and engagement.
The competitive question will ultimately come down to implementation.
Can a bank deploy predictive models quickly? Can relationship managers understand the recommendations? Can the system connect predictions to actual customer journeys? And can the bank measure whether those interventions protect deposits or increase product relationships?
USB’s case suggests that the answer can be yes without beginning with a wholesale replacement of core infrastructure.
That could be particularly important as AI investment shifts from experimentation toward measurable business outcomes.
Gartner forecasts worldwide spending on AI models and platforms will reach $64 billion in 2026, up 63.4% from 2025, while noting that enterprises are increasingly scrutinizing AI spending for cost, performance and measurable value.
For banks, that means the next AI investment case is unlikely to be built around novelty.
It will be built around numbers such as deposits retained, products added, customers saved and lifetime value increased.
USB’s approach illustrates that shift.
The real promise of banking AI is not simply predicting what a customer might do.
It is giving the institution enough time—and enough context—to do something useful about it.
Market Landscape
Retail banking AI is moving toward decision intelligence and personalized engagement, with customer data becoming the foundation for both retention and revenue growth.
Gartner’s 2026 banking research identifies AI use cases across customer experience, revenue growth, operational efficiency and risk management, while its customer-experience research highlights unified data and journey orchestration as important components of personalization.
The market can broadly be divided into several layers:
- Customer data and intelligence: consolidating and activating information about accounts, products, transactions and interactions.
- Predictive analytics: identifying churn, propensity to buy, deposit behavior and other customer signals.
- Next-best-action decisioning: recommending products, offers or interventions based on individual circumstances.
- Engagement orchestration: delivering those actions through digital channels, relationship managers or targeted campaigns.
- Core banking infrastructure: the systems that ultimately hold accounts, transactions and product relationships.
The strategic advantage comes from connecting those layers.
A predictive model that never triggers an action has limited commercial value. Likewise, a personalized campaign is less effective if the bank does not understand which customers need it.
That is why AI-powered banking is increasingly becoming a closed loop: observe behavior → predict risk or opportunity → recommend action → engage → measure outcome → improve the model.
For community and regional banks, the ability to build that loop without replacing the core may prove particularly attractive.
Top Insights
- Community banks can use AI without wholesale core replacement, activating existing customer and transaction data to generate predictive insights and targeted engagement.
- Churn prediction becomes more valuable when paired with action, allowing banks to identify at-risk customers and determine relevant retention or product strategies.
- Deposit retention is becoming an AI battleground, as personalization and emerging financial agents make customers more sensitive to rates, products and competing offers.
- Explainable predictions matter in banking, giving relationship teams context behind AI recommendations instead of forcing employees to act on opaque risk scores.
- The strongest AI business cases are becoming measurable, shifting attention from experimentation toward deposits protected, products added, customers retained and lifetime value.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
