12,000+ Financial Institutions. 200M+ Monthly Users. The Real Story Behind ChatGPT, Conversational AI in Banking and Agentic AI

ChatGPT can now understand your financial life. The next question is whether AI will eventually be able to act on it.

On May 15, 2026, OpenAI introduced a personal finance experience in ChatGPT that changed the conversation around Conversational AI in Banking. Through Plaid, eligible users in the U.S. can connect financial accounts and ask ChatGPT questions grounded in their actual financial context. Plaid provides connectivity to more than 12,000 supported banks, credit cards, brokerages and other financial institutions, while OpenAI says more than 200 million people come to ChatGPT every month for personal finance questions.

The significance is not simply that ChatGPT can now analyze transactions. It is that a general-purpose AI interface is becoming capable of understanding balances, spending patterns, subscriptions, investments, liabilities and financial goals in one conversational environment. OpenAI says its GPT-5.5 Thinking model scored 79 out of 100 on its internal personal-finance benchmark, with GPT-5.5 Pro scoring 82.5, based on evaluation involving more than 50 finance professionals.

That combination of scale, context and reasoning creates a new benchmark for how consumers may expect to interact with financial services.

But there is an important boundary.

ChatGPT can understand your finances. It cannot currently operate your finances.

The current Finances experience is read-only. It can analyze spending, review investments, identify subscriptions, compare financial trends and help users plan budgets or debt repayment. It cannot move money, pay bills, make trades, change account settings, change retirement contributions, or open and close financial accounts.

And that distinction is arguably the most important part of the entire story.

For years, Conversational AI in Banking was primarily associated with customer service. Banks deployed chatbots to answer FAQs, check application status, explain products and reduce pressure on contact centres.

The value proposition was straightforward: make it easier for customers to find information.

The new generation is different because AI can increasingly understand context and intent rather than simply match a question to an answer.

A customer might not ask, “What is my average monthly grocery expenditure?”

They might ask, “Why am I saving less this month?”

Answering that question requires the AI to connect multiple pieces of information, identify a pattern and explain the relationship between them. OpenAI's current Finances experience is designed around exactly this type of contextual interaction, including spending analysis, upcoming payments, portfolio information, budgeting and scenario planning.

This is a meaningful evolution in banking AI because the interface is no longer simply retrieving information from a bank's knowledge base. It is reasoning across a customer's financial context.

But reasoning is not execution.

A system can tell a customer that they have excess cash sitting in an account without being authorized to move that money. It can identify an upcoming bill without being allowed to pay it. It can explain an investment opportunity without being able to execute the trade.

That gap is where Agentic AI becomes strategically important.

Agentic AI is no longer confined to technology demonstrations. The 2026 Global AI in Financial Services Report from Cambridge Judge Business School found that 52% of surveyed financial-services firms were already adopting Agentic AI, while 23% had reached scaling or transforming stages and 29% remained in piloting.

The report also reveals a more important competitive divide. 57% of fintech respondents were adopting Agentic AI compared with 45% of traditional financial institutions. Looking across AI more broadly, 81% of surveyed financial-services firms were adopting AI at some level, yet only 14% considered AI transformational to organisational strategy and competitive advantage.

That 14% figure is particularly revealing.

The banking industry does not appear to have an AI adoption problem. It has an AI execution problem.

Financial institutions are experimenting with AI at scale, but relatively few have successfully converted those experiments into a transformational competitive capability. For banks, this means the next advantage will not necessarily come from having access to a better model. It will come from connecting AI to data, APIs, permissions, workflows and core banking infrastructure in a way that creates measurable business outcomes.

The easiest way to understand the difference is to follow a single customer request.

Imagine a customer says:

“I have $5,000 sitting in my account. Keep enough for my upcoming expenses and put the rest somewhere that earns a better return.”

A traditional chatbot may explain the bank's savings products.

A modern conversational AI system can potentially understand the customer's financial position and explain the available options.

An Agentic AI system could eventually take the next step. It could identify upcoming obligations, calculate an appropriate amount to retain, compare eligible products, check the customer's preferences and permissions, request authorization where necessary, and execute the approved workflow.

That is the difference between answering an intent and completing an intent.

The important point is that Agentic AI does not eliminate banking controls.

It makes those controls more important.

Financial services are fundamentally different from most consumer technology because an incorrect recommendation and an incorrect transaction have very different consequences.

If an AI incorrectly categorizes a restaurant transaction, the customer can correct it.

If an AI incorrectly transfers ₹5 lakh, the problem is fundamentally different.

This is why the future of Agentic AI in Banking depends on a control architecture around the model. Identity, authorization, transaction limits, fraud detection, policy enforcement, audit trails and human escalation all become part of the AI system.

The IMF's analysis of Agentic AI and payments frames this through three distinct layers: intent, authorization and settlement. AI can interpret an objective, but financial infrastructure still needs to determine whether the requested action is authorized and how it should safely settle.

This creates an important design principle for banks:

Do not give AI unrestricted access to banking systems. Give AI controlled access to specific banking capabilities.

In practice, that means an agent should not simply have permission to “move money.” It should have permission to perform a specific transaction, for a specific customer, within a specific limit, under a specific set of conditions, with every action recorded.

That is the infrastructure layer that will determine whether Agentic AI becomes commercially viable in financial services.

India's payment ecosystem makes this shift particularly significant.

UPI processed 24.51 billion transactions worth ₹29.82 trillion in August 2026, according to Reuters. At that scale, even a relatively small percentage of transactions becoming agent-initiated represents a substantial new payment category.

Reuters reported on September 1 that India was preparing a framework that could allow AI agents to make certain small digital payments without requiring approval for every transaction. The proposed approach is expected to include rule-based instructions, spending limits, identity checks, audit trails and a liability framework.

The significance is not that every payment will suddenly become autonomous.

It is that the payment instruction itself can become programmable.

A consumer could potentially establish a rule such as:

“Pay recurring expenses below ₹5,000 automatically.”

“Purchase this product if the price falls below ₹X.”

“Do not allow the agent to spend more than ₹10,000 in a defined period.”

The customer moves from approving every transaction individually to defining the conditions under which an agent is allowed to act.

That is a fundamental change in the relationship between customers, AI and financial infrastructure.

India is not developing this concept in isolation.

In March 2026, Santander and Mastercard announced a live end-to-end payment executed by an AI agent within a controlled banking environment. The experiment demonstrated how an AI agent could operate with predefined permissions and interact with existing payment infrastructure.

These developments matter because they move Agentic AI in Banking from a theoretical discussion toward an infrastructure question.

The industry is beginning to test what happens when an AI system is not merely allowed to read financial information, but is given carefully constrained permission to act.

That is also why themes surrounding FinovAI Summit and other 2026 financial AI discussions are increasingly relevant to bank leaders. The important conversations are moving beyond chatbot capabilities toward agent governance, infrastructure readiness, AI risk, payment execution, customer authorization and measurable ROI.

It is tempting to frame this as a battle between banks and OpenAI.

That is too simplistic.

The more important question is who owns the customer's financial interface.

For decades, the bank controlled the interface through which customers accessed financial services. The mobile banking application was effectively the front door to the customer's financial life.

Generative AI changes that relationship.

If customers increasingly begin with:

“How much did I spend this month?”

“Can I afford this purchase?”

“Which account should I use?”

“Move money so my bills are covered.”

the first interaction may happen outside the bank's application.

The bank then becomes the infrastructure behind the conversation.

That does not necessarily make banks less important. It changes where their competitive advantage sits.

A bank that owns excellent APIs, reliable data, granular permissions, strong identity infrastructure and highly automated workflows can potentially participate in AI-native financial experiences even when the customer interface belongs to another platform.

A bank with an attractive app but fragmented back-end systems may find that its strongest digital asset, the customer interface, becomes less important.

The Cambridge report provides another reason this matters. Although 81% of surveyed financial-services firms are adopting AI at some level, only 14% currently view AI as transformational to their strategy and competitive advantage.

This suggests that simply adding AI to existing processes is unlikely to be enough.

The banks that gain the most value from Agentic AI will likely redesign workflows around what agents can actually do.

Consider loan servicing.

A conventional AI assistant might answer:

“What documents do I need?”

A contextual AI assistant might know which documents the customer has already submitted.

An agentic workflow could potentially identify missing information, request it, validate available data, trigger internal checks, update the application and escalate exceptions.

The difference is not the quality of the conversation.

The difference is the percentage of the workflow that AI can complete.

That is why banking AI metrics will need to evolve.

Instead of measuring only chatbot sessions, banks should increasingly measure resolution rate, straight-through processing, workflow completion, human escalation rate, time-to-resolution, cost per completed interaction, conversion and revenue generated per AI-assisted workflow.

The most valuable banking AI may ultimately be the system that has the fewest conversations because it completes the customer's objective in the fewest steps.

The May 2026 OpenAI and Plaid integration should therefore be viewed less as a threat to banking applications and more as a signal about changing customer expectations.

200 million monthly personal-finance users. More than 12,000 supported financial institutions. Financial information becoming conversational.

Those numbers demonstrate that the interface is changing.

But the current read-only boundary demonstrates something equally important: financial intelligence and financial execution are still separate layers.

The next stage of Conversational AI in Banking will be about connecting those layers safely.

That means moving from:

Data → Answer

to:

Data → Understanding → Intent → Authorization → Action → Verification

Agentic AI is the technology layer that can potentially connect that chain.

Banking infrastructure is the control layer that makes it safe.

And the institutions that combine both could define the next generation of digital financial services.

ChatGPT + Plaid did not turn ChatGPT into a bank.

It did something arguably more important.

It showed millions of consumers what financial services could feel like when the interface understands context.

Agentic AI is now pushing the next question:

What happens when that interface can also act?

The answer will not be determined by the smartest language model alone. It will depend on identity, authorization, APIs, risk controls, payment infrastructure, governance and the ability to measure whether an AI agent actually creates business value.

For banks, the strategic race is therefore no longer simply about deploying Conversational AI in Banking.

It is about becoming agent-ready.

Because the future customer may not say:

“Open my banking app.”

They may simply say:

“Take care of it.”

And the bank that can safely turn that sentence into a completed financial outcome will have a very different competitive position from the bank that can only explain how the customer could do it themselves.

Finov AI Summit Europe 2027 focuses on the shift from AI pilots to enterprise execution because financial institutions are increasingly looking beyond experimentation. The key challenge is building AI capabilities that are secure, scalable and measurable, rather than running isolated proof-of-concepts that never reach production.

Banking leaders attending Finov AI Summit Europe 2027 can expect discussions centered on practical AI execution across banking, insurance and payments. The focus is on how financial institutions can move AI into real enterprise environments while addressing scalability, security, governance, operational readiness and measurable business outcomes.

Finov AI Summit Europe 2027 is positioned around implementation, not AI experimentation alone. Its central message, “From experimentation to execution,” reflects the growing need for financial institutions to turn AI strategies and pilots into secure, scalable capabilities that deliver measurable enterprise value.

Having AI pilots is no longer the same as having an AI-ready enterprise. Finov AI Summit Europe 2027 is relevant for banks that need to address the next stage: scaling successful pilots, connecting AI to enterprise systems, establishing governance, managing risk and proving measurable ROI from AI investments.

The most important challenge is moving from AI experimentation to reliable enterprise execution. For financial institutions, that means addressing questions around secure deployment, scalable infrastructure, governance, data readiness, workflow integration, human oversight and measurement of AI-driven business value.

Financial institutions cannot evaluate AI purely on model performance. AI systems may influence sensitive financial decisions and workflows, making security, scalability, governance and measurable outcomes essential. The broader financial-services industry is already adopting AI at scale, yet only 14% of surveyed firms in the 2026 Global AI in Financial Services Report considered AI transformational to organisational strategy and competitive advantage.

Finov AI Summit Europe 2027 is particularly relevant to the transition toward Agentic AI because financial institutions are moving beyond systems that simply answer questions toward AI that can potentially plan and execute controlled workflows. The key challenge is connecting AI capabilities with permissions, APIs, risk controls, authorization and auditability.

The summit is positioned as a cross-sector AI event for banking, insurance and payments, making it relevant to organizations dealing with similar challenges around automation, customer interactions, risk, data and enterprise AI adoption. This broader perspective can help leaders understand where AI strategies overlap across financial-services sectors.

Finov AI Summit Europe 2027 is designed for senior decision-makers across banking, insurance and payments, particularly leaders involved in AI, technology, digital transformation, innovation, data, operations and enterprise strategy. The event information highlights 150+ senior decision-makers attending in Munich.

Finov AI Summit Europe 2027 takes place on 24–25 March 2027 in Munich, Germany. The event is positioned as Europe’s AI summit for banking, insurance and payments, with a focus on turning AI experimentation into secure, scalable and measurable enterprise capability.

For financial services, “from experimentation to execution” means moving beyond AI proofs-of-concept and embedding AI into real business processes. Instead of measuring success by the number of pilots launched, institutions need to evaluate AI through outcomes such as workflow completion, operational efficiency, customer resolution, risk management and measurable financial value.

AI adoption alone does not guarantee competitive advantage. The 2026 Global AI in Financial Services Report found that 81% of surveyed financial-services firms were adopting AI at some level, while only 14% viewed AI as transformational to organisational strategy and competitive advantage. This gap makes enterprise execution a critical priority for financial-services leaders.

Financial institutions should assess whether their AI initiatives can move beyond isolated pilots into production. That includes reviewing data quality, API readiness, AI governance, authorization controls, workflow integration, human oversight, security and ROI measurement. These foundations determine whether an AI initiative can scale safely across the enterprise.

An AI pilot demonstrates that a technology can work. An enterprise AI capability demonstrates that it can work reliably at scale, integrate with existing systems, comply with governance requirements, operate securely and produce measurable business outcomes. That distinction is central to the “experimentation to execution” theme of Finov AI Summit Europe 2027.

As AI moves from experimentation into larger deployments, executives need to determine whether those investments actually improve business performance. AI ROI should increasingly be measured through metrics such as workflow completion, straight-through processing, time-to-resolution, human escalation, operating costs, conversion and revenue generated by AI-assisted processes.