It’s a bit like opening your LinkedIn timeline now. Every post reads the same. Someone’s AI platform just revolutionized banking. Someone else’s banking software just became AI-native. The language is almost interchangeable: modern stack, pre-trained models, faster implementation, industry-leading accuracy.
Scroll long enough and you start wondering if everyone attended the same webinar on “How to Position Your Product in the AI Era.” The pitches sound sophisticated from both directions. What they rarely mention is what happens after the demo, when these claims meet actual banking operations. When you’re processing real customer money under regulatory oversight. When Diwali bonus season hits and transaction volumes triple overnight, and your AI needs to distinguish between normal seasonal spikes and actual fraud patterns. When a customer who’s been with you for 15 years suddenly looks like a new acquisition because their account migrated from a different product code and your engagement engine doesn’t recognize the relationship history.
This is where experience stops being a vendor claim and becomes what really keeps systems working. And it’s where the conversation needs to shift from AI capability to something more fundamental: giving banks the power of cognition to actually deliver value to their customers. Not just automation. Not just pattern matching. Cognition that understands context, learns from patterns specific to banking, and makes intelligent decisions that serve customers better.
Cognitive Banking
Cognitive banking, as Clayfin defines it, is the practice of empowering banks with intelligence capabilities specifically designed to help them deliver greater value to their customers — not AI features adapted from other industries, but cognition built for the realities of financial services. It rests on three capabilities.
Contextual understanding: the ability to interpret customer behaviour not in isolation, but within the full context of their financial life, relationship history, and life stage.
Regulatory reasoning: the ability to operate within the boundaries of banking compliance not as a constraint bolted on after the fact, but as a design principle embedded in how the platform thinks.
Adaptive intelligence: the ability to learn from patterns specific to banking – the signals that distinguish financial stress from routine cash management, seasonal volume spikes from fraud, and a loyal long-tenure customer from a new acquisition. Together, these capabilities give banks the cognitive power to move from reactive service to proactive value creation for the customers they serve.
What You Learn by Watching What Breaks
Banking expertise isn’t simply about knowing what the industry should do in theory. It’s more like pattern recognition from watching what breaks in practice, repeatedly, across different contexts.
We’ve seen how customer behaviour shifts during major regulatory changes. Take the account aggregator framework rollout. On paper, it enabled data sharing across institutions. In practice, it changed how customers think about their financial relationships. Their mental model of “my bank knows about my money” became “multiple services can see pieces of my financial life.” Some customers embraced the convenience immediately. Others needed months to understand what they were consenting to and why it mattered. The intelligence layer needs to understand both the technical capability and this psychological shift, because the same data-sharing action means completely different things to different customer segments.
Merger integrations teach you even more. When two banks combine systems, it’s never just a technical migration. Customer expectations developed over years at one bank collide with different service cultures at another. Legacy product structures don’t map cleanly to new offerings. Someone who was a premium customer at Bank A might be just another account holder in Bank B’s segmentation logic. The intelligence layer needs to understand both histories to serve the combined customer base without making people feel like they’ve lost something in the transition.
- Satyanarayana Raju, MD & CEO of Canara Bank, describes how this plays out in their implementation. They built 50 AI/ML models specifically for upselling, cross-selling, and churn prediction. Not generic recommendation engines adapted to banking. Banking-specific models that understand product interdependencies, regulatory constraints, and customer lifecycle timing within their operating environment. That specificity matters.
This is what cognitive banking really looks like in practice. It’s not about having more data or faster processing. It’s about having the cognitive ability to understand what different patterns mean in different contexts and make decisions that create value for customers.
Why Banking Context Can’t Be Prompt-Engineered
Generic AI platforms are genuinely sophisticated at pattern matching. Banking requires understanding which patterns genuinely matter and which ones are noise. That’s a different skill. That’s where cognition becomes essential.
Take regulatory interpretation. The RBI’s digital lending guidelines don’t just affect disclosure text, rather they shape how you structure nudges, when you can surface credit offers, and which customer signals you’re allowed to act on. A platform that doesn’t understand these boundaries can generate recommendations that look intelligent in a demo but create compliance violations in production. Cognitive banking means understanding not just the rules, but how they define what intelligent engagement looks like within those constraints.
Product construction logic works the same way. Cross-selling a credit card to a savings account customer has different risk, reward, and regulatory dynamics than cross-selling a personal loan — different qualification logic, different timing windows, different oversight. This isn’t something you bolt onto a horizontal AI platform after the fact. It needs to be embedded in how the platform thinks about customer relationships from the start.
Customer lifecycle timing matters more than most vendors realise. Low balance means something completely different for a salary account versus a trading account versus a business account. The intelligence layer needs to know when it signals financial stress versus routine cash management because getting that wrong doesn’t just cost a cross-sell opportunity. It costs trust. Cognitive banking means having the reasoning capability to make these distinctions and act on them appropriately.
Ratan Kumar Kesh, ED & COO of Bandhan Bank, framed their recent technology partnership with explicit targets: “reduce cost by 20% and improve efficiency by about 20%.” Those aren’t vanity metrics. They’re outcomes that require banking-specific intelligence, not generic workflow automation. The platform needs to understand their operating context well enough to deliver measurable efficiency within their regulatory framework.
Production vs. Demo
Demos show AI making perfect recommendations in clean scenarios. Production is where the messy reality lives.
Real banking operations deal with incomplete data from legacy migrations, channel switches where context gets lost between systems built by different vendors in different decades, and address formats the legacy system can’t parse so your credit scoring model fails before it starts. Account aggregator data might show significant balances elsewhere, but your system can’t act on it because consent windows expired during a maintenance window.
When Jana Small Finance Bank won the IBSi Digital Banking Award, it wasn’t for pilot metrics. It was for production outcomes at scale, real turnaround time reductions processing actual customer requests under regulatory oversight, with all the edge cases that implies.
The difference between platforms built for banking versus platforms adapted to it shows up in how they handle these situations. Banking-specific platforms treat compliance, data quality, and system constraints as design requirements. Adapted platforms treat them as exceptions to handle later and that “later” tends to arrive during implementation, not the sales cycle.
Mohamed Amiri Al Muhairi, CEO of Ajman Bank, emphasizes this when discussing their digital evolution. Banks operating under Islamic banking principles face an additional layer of complexity. They need solutions that understand Shariah compliance alongside regional regulatory requirements and customer expectations specific to their market. Generic platforms require extensive customization to handle this. Platforms built with cognitive banking principles embed that context from the start.
Experience as Competitive Moat
AI capability is commoditising quickly. You can access sophisticated models through APIs, train on massive datasets, build impressive demos in weeks.
Banking expertise under regulatory oversight doesn’t commoditise the same way. It comes from watching what breaks across transformation cycles and from understanding how regulatory regimes shape what intelligence can recommend, which patterns predict churn in one market but signal satisfaction in another, how customer expectations shifted through COVID-driven digital adoption, the account aggregator rollout, and changing frameworks across markets.
After two decades across India, the Middle East, Southeast Asia, and Bangladesh, we’ve learned something that sounds simple but isn’t: the hard problems aren’t AI problems. They’re banking problems that require cognitive capability to solve.
This is what cognitive banking delivers. Not AI features added to banking software, or banking use cases bolted onto AI platforms but cognition applied specifically to help banks deliver value to their customers. Understanding context across channels and time. Learning from patterns specific to financial relationships. Making intelligent decisions within regulatory constraints. Reasoning through edge cases no training dataset fully captures.
The platforms winning enterprise deployments aren’t the newest or flashiest. They’re the ones that understand what works when you’re processing real customer money at scale under regulatory oversight. That understanding comes from production experience accumulated across markets, regulatory regimes, and transformation cycles from building cognitive capability specifically for banking, not adapting generic AI to financial services after the fact.
The LinkedIn timeline will keep insisting everything is AI-powered now. Some of it is. Some of it is marketing. Banks that have been through enough technology cycles can usually tell the difference. They’re looking for cognitive banking that delivers value to customers not just AI claims that sound good in vendor pitches.
