The Battle for the Bank Customer. Why banks can still win in the AI era – and what it will take.
An article written by Udi Ziv, Chief Executive Officer of Personetics
On May 15, 2026, OpenAI flipped a switch that sent a chill through every bank boardroom in the world. ChatGPT Pro subscribers in the United States could now connect their bank accounts through Plaid and ask the platform about balances, transactions, investments, and liabilities, across more than 12,000 institutions, including Chase and Schwab. The financial press wrote the headline that banks had been dreading for years: ChatGPT is now your bank app.
The stakes are real. LLMs are fast becoming the default interface through which people access information and search for answers in every facet of their lives. ChatGPT is approaching one billion weekly active users. No single bank comes close to that reach. A bank that becomes an invisible utility inside a customer’s AI workflow – essentially becoming a dumb pipe – does not necessarily lose the account. But in addition to becoming indistinguishable from every other bank, it risks losing something far more valuable: the intelligence layer that shapes the conversation, guides decisions, and influences behavior. In an AI-first world, whoever understands the customer best determines what happens next.
For an industry that has owned its customer relationships for hundreds of years, the prospect of becoming a faceless, commoditized provider of financial products – is not a road bump. It is an existential moment.
The Chatbot is not the Strategy
The instinctive response from most large banks has been to fight fire with fire. Build a chatbot. Add a prompt box to the app. Launch an AI assistant with a name and personality. The mistake is believing that the chatbot itself is the strategy, and not just a feature. As conversational interfaces become ubiquitous, differentiation will not emanate from standalone chatbots. It will come from the intelligence, context, and action layer underneath it. Banks that don’t grasp the distinction will lose the battle for the bank customer.
In most cases, these AI features are launched as standalone initiatives, disconnected from the intelligence layer and how the bank engages with its customers: the structured experiences, the contextual nudges, and the personalized guidance built over years. The result is often a fragmented experience. A customer asks the standalone LLM, “Am I ready to start investing?” based on single-threaded transaction information, the LLM may say yes. However, the bank knows the customer is building an emergency fund and is only halfway to their savings target. Without that context, the recommendation to invest may conflict with the customer’s current financial priorities.
Additionally, replicating the conversational interface of ChatGPT or Gemini will prove to be a losing strategy. The external LLMs have a head start measured in years, and a user base measured in hundreds of millions. A bank-built chatbot that attempts to replicate that experience will not pull customers away from their LLM of choice. It will remind them of the capability gap – and drive them deeper into it. Most banks are already surrendering. They just don’t know it yet. But the banks that recognize what is actually at stake have a genuine opportunity to win – not by competing with LLMs on their turf, but by doing something LLMs fundamentally cannot. From our work partnering with the largest banking institutions across the world, we see that banks growing share of wallet and solidifying primacy are the ones leveraging context to deliver personalized insights that the customers really value – with chatbots fully integrated into the customer’s overall relationship context.
LLMs lack the Data Depth that Banks Possess
The ChatGPT-Plaid integration delivers four categories of data: balances, transactions, investments, liabilities. Read-only. But inside the bank sits a completely different picture: behavioral signals, device fingerprints, login geolocation, dispute history, internal fraud scores, card controls, customer service transcripts, loan officer notes, wire beneficiary details, joint account permissions, beneficiary designations and on and on. None of this crosses the aggregator wire by design, and the regulatory framework explicitly says it does not have to. The bank knows who the customer actually pays each month, how they behave across every channel, and what that behavior signals about the customer’s financial future. The bank that fails to act on that advantage is handing the relationship to someone else.
Context that compounds over time
ChatGPT sees a grocery spend that jumped 40% last quarter. The bank sees the same data and knows the customer recently added a dependent, has a savings rate that’s been declining for six months, and is three months away from a CD maturity. ChatGPT knows what happened. The bank knows what it means – and unlike any LLM, it can act on it. It can originate a HELOC, execute a fiduciary trade, underwrite a credit decision, file a SAR, or move money on a customer’s behalf. These capabilities are created by statute and by the customer’s pre-existing relationship with the institution – neither of which is available for purchase. Consumers may trust AI to explain their finances. They do not trust AI to control their finances. That trust, built over generations, is the deepest moat of all.
This context is built from financial behavior, preferences, goals, and intent – not just financial data. A customer moving from reactive to proactive money management will exhibit patterns: fewer last-minute transfers, fewer low-balance situations, more consistent goal contributions. Where LLMs see transactions, the bank can see trajectory, and recommend a next best action based on it. That understanding compounds across every interaction and channel in a way no read-only transaction feed can reconstruct.
The banks that win the AI era will not be the ones who built the best chatbot. They will be the ones who embed AI across the enterprise and deploy it to serve customers personalized experiences at scale. AI and agentic capabilities need to be woven into every touchpoint: proactive alerts that feel genuinely personal, guidance timed to the right moment, actions suggested and completed in a single seamless flow. That same intelligence needs to travel with the customer – available to the branch banker, the relationship manager, and the call center agent the next time that customer shows up. The structured engagement experiences banks have mastered are supercharged, not replaced, by AI. When conversational AI is built in isolation, it becomes just another channel. When it is built as part of the intelligence layer across all channels, it transforms the relationship.
A bank that deploys its full data advantage does not feel like a utility. It feels like a financial partner that genuinely understands each customer – and can act on that understanding in ways no external AI ever will.
Be the destination, not the API
Some banks are already doing this – deploying AI not as a feature but as the connective tissue across every customer touchpoint – and the distance between them and everyone else is growing. The opportunity for all banks is there for the taking, but the window is closing. The major LLMs are advancing rapidly and adding millions of active users monthly. While banks maintain their unique advantages of data and trust, they would be best served to spend this time building a compelling, intelligent engagement destination that will position themselves to thrive in the agentic era.
The ChatGPT-Plaid integration was a wake-up call. For banks that move slowly, it is also the beginning of the end. Banks have the data, license, and trust. To win, they need to invest now in building an intelligence layer that enables them to leverage these resources to unlock the full power of AI. But rest assured – the customer isn’t waiting for their longtime bank to figure this out. The bank that shows up first – with real intelligence, real context, real action – wins the relationship.
