31% of Consumers Are Already Asking Conversational & Agentic AI About Money. Is Your Bank Losing the First Financial Conversation?

The biggest threat from conversational AI in banking is not that customers will stop using banks. It is that they may stop asking banks first. For decades, financial institutions controlled the most valuable moment in the customer journey: the moment a person had a financial question.

Need to understand a transaction? Open the banking app.

Looking for a credit card? Visit the bank.

Want to compare a loan? Search the bank's products.

Need financial guidance? Speak to the bank.

That model is beginning to change.

According to Forrester's September 2, 2026 research, nearly one in three consumers, 31% across the US, Canada and the UK, now use conversational AI for at least some personal finance questions. More importantly for banks, 24% of those consumers use AI tools outside their bank's ecosystem.

Those two numbers create a much bigger story than AI adoption. They suggest that a growing share of consumers are beginning their financial journeys somewhere the bank does not control. And that changes the competitive question.

It is no longer simply:

"How good is our banking chatbot?"

It is:

"When our customer has their next financial question, will they ask us, or will they ask an AI first?"

The 31% figure is easy to interpret as another consumer AI adoption statistic.

That would undersell it.

A financial question is often the beginning of a commercial journey.

"Can I afford a new car?" can become an auto-loan conversation.

"How much should I save every month?" can become an investment conversation.

"Is there a better credit card for me?" can become a product-switching conversation.

"Why are my expenses increasing?" can become a financial-health conversation.

If these questions begin inside the bank, the institution has context. It knows the customer's relationship, products, balances, transaction history and potentially their eligibility for additional services.

If the same question begins with a third-party AI assistant, the bank may enter the journey later.

This is what makes Forrester's finding about the 24% using AI outside their bank ecosystem particularly important. Forrester describes this as the risk of interface-level disintermediation, where third-party AI becomes the gateway through which consumers access financial information and potentially financial services.

The bank does not necessarily lose the customer.

It can lose the moment of influence.

And in financial services, that moment can influence what product gets considered, which provider gets shortlisted and what the customer expects before they ever reach the bank.

Digital banking changed where customers performed banking. Conversational banking could change how customers discover what banking they need in the first place. The distinction matters. A digital banking experience typically asks customers to understand the bank's structure.

They navigate:

Accounts → Payments → Cards → Loans → Investments → Support

Conversational AI reverses that model.

The customer starts with an outcome:

"I want to reduce my monthly expenses."

"I need a loan for a new car."

"I have extra cash. What should I do with it?"

"Why did my spending increase this month?"

The system then determines which information, products or workflows may be relevant.

This is why the future of conversational banking is not simply about making chatbots sound more human. It is about making financial services intent-driven.

Forrester's research on conversational banking argues that banks need to move beyond isolated chatbot deployments and build the technology, data, governance and operating foundations required to scale conversational experiences.

That is a very different proposition from adding another chat window to a mobile banking application.

The consumer side of the equation is only half the story. The financial-services industry is simultaneously increasing its investment in AI.

The 2026 Global AI in Financial Services Report, produced by the Cambridge Centre for Alternative Finance at Cambridge Judge Business School with partners including the BIS, IMF and World Economic Forum, surveyed 628 organisations across 151 jurisdictions, including 203 fintechs, 149 financial incumbents, 146 AI vendors and 130 regulators.

Its findings are significant.

81% of financial-services industry respondents are adopting AI at some level.

40% are already at advanced AI adoption stages, defined as scaling or transforming.

Yet only 14% currently view AI as transformational to organisational strategy and competitive advantage.

That creates one of the clearest signals in the 2026 banking AI market:

AI adoption is moving faster than AI transformation.

Banks are deploying AI.

But many have not yet redesigned the underlying business around it.

That distinction becomes even more important when looking specifically at agentic AI.

The same Cambridge research found that 52% of financial-services industry respondents are already actively adopting agentic AI, meaning they are piloting, scaling or transforming with the technology.

Of that group, 23% are already at scaling or transforming stages, while 29% remain in the pilot stage. Fintechs are ahead of traditional financial institutions, with 57% reporting agentic AI adoption compared with 45% among traditional financial institutions.

Looking forward, 81% of industry respondents expect agentic AI to be meaningfully achieved by 2030.

This is where the story becomes strategically interesting. The industry is not simply building better conversational interfaces. It is beginning to build systems that can reason across workflows, use tools and potentially perform actions.

That creates a progression:

Chatbot → Conversational AI → Contextual AI → Agentic AI

The first three stages primarily improve the customer's interaction with information. The fourth changes what AI can potentially do.

Consider a simple banking scenario. A customer asks:

"I have ₹1 lakh sitting in my account. What should I do with it?"

A traditional chatbot can explain savings products. Conversational AI can understand the question in natural language and provide contextual guidance. A more advanced system could consider the customer's financial information, goals and existing relationship.

Agentic AI introduces another layer. It can potentially determine the sequence of actions required to achieve an objective, interact with approved systems, request authorisation and execute permitted steps. That is the difference between answering a question and pursuing an outcome.

This distinction is critical because it explains why the future of conversational AI in banking cannot be measured only by chatbot containment. If AI eventually becomes capable of completing financial workflows, the relevant question becomes:

How much of the customer's objective can the system safely complete?

A financial AI system can be extremely intelligent and still be unsuitable for autonomous execution. That is because banking has a characteristic that many other industries do not:

Every meaningful action has consequences.

An incorrect recommendation can be corrected.

An incorrect payment can create a financial loss.

An incorrect credit decision can affect a person's access to finance.

An incorrect investment action can create regulatory and fiduciary consequences.

That means agentic AI requires a second layer around the intelligence itself.

The system needs to know:

Who is the customer?

What is the customer asking for?

What is the agent allowed to do?

What requires explicit consent?

What transaction limits apply?

Which compliance rules must be checked?

When must a human intervene?

How is every action recorded?

This is why agentic AI in banking is fundamentally an architecture and governance problem, not simply an LLM problem.

The Cambridge report reinforces the scale of that challenge. While AI adoption is widespread, 55% of industry respondents said measuring AI deployment value is difficult, rising to 76% among large financial institutions.

The implication is important: Banks are not only trying to determine whether AI works. They are trying to determine whether AI creates enough measurable value to justify its operational, security and regulatory complexity.

There is another finding from the Cambridge research that should receive more attention.

Data availability and quality were identified as a leading constraint on AI adoption, cited by 40% of industry respondents and 34% of fintech respondents.

This matters enormously for conversational banking. A bank can deploy the latest foundation model and still deliver a poor experience if the AI cannot reliably access accurate customer, product and transaction information.

Imagine asking:

"How much can I safely spend this month?"

The AI needs more than a language model.

It may need:

Current account balances.

Pending transactions.

Upcoming bills.

Credit obligations.

Recurring expenses.

Savings goals.

Product information.

Customer permissions.

Risk policies.

The quality of the answer depends on the quality, freshness and accessibility of those underlying data sources.

This is why the next banking AI architecture will increasingly resemble a connected intelligence layer rather than a chatbot sitting on top of a knowledge base.

The OpenAI and Plaid partnership provides one of the clearest examples of this transition.

On May 15, 2026, OpenAI launched its personal finance experience in ChatGPT. Eligible users in the US can connect financial accounts through Plaid and ask questions grounded in their financial context. OpenAI says the experience supports connections to more than 12,000 financial institutions through Plaid.

Plaid also says more than 200 million people use ChatGPT each month for personal-finance-related questions.

The important point is not that ChatGPT has become a bank. It has not.

The current Finances experience can help users analyze spending, bills, subscriptions, net worth and investments, but it cannot move money, pay bills, make trades, change account settings or open and close financial accounts.

That limitation actually makes the product more strategically interesting. It demonstrates the first layer of the transition:

AI can sit between the customer and financial information without owning the underlying financial institution.

The next question is what happens when AI also gains controlled access to financial actions.

That is the point at which conversational AI starts becoming agentic.

The industry will naturally focus on the headline:

31% of consumers use conversational AI for personal finance questions.

But for bank executives, the more strategically important number may be 24%.

Why?

Because it tells banks where those conversations are happening. If consumers use AI inside the bank's ecosystem, the institution retains the interface, customer context and relationship. If they use third-party AI, the institution may become one data source or service provider inside someone else's interface.

That changes the economics of customer engagement. The traditional model is:

Bank owns interface → Bank owns conversation → Bank owns journey

The emerging model could become:

AI owns interface → AI mediates conversation → Bank provides data, products and execution

That does not automatically mean banks lose. It means banks need to determine what they must own and what they must make interoperable.

If the conversational interface becomes commoditized, banks need another source of differentiation.

That source could be trusted.

A third-party AI might understand a customer's request.

But the bank controls the regulated financial account. The bank controls transaction infrastructure. The bank controls identity and authentication. The bank manages risk and compliance. The bank owns the contractual relationship.

The opportunity is therefore not necessarily to beat every AI company at conversation. It is to make the bank's infrastructure sufficiently intelligent, secure and interoperable that it can participate in AI-driven financial journeys. This creates a new strategic equation:

AI interface + bank data + permission layer + banking APIs + execution = intelligent financial service

The bank does not have to own every layer. But it needs to be capable of operating across them.

The difference between conversational AI and agentic AI becomes clearer when we stop talking about technology and start talking about business outcomes. Imagine a customer saying:

"I want to reduce my monthly debt payments."

A chatbot can explain debt-management options.

Conversational AI can analyze the customer's financial situation and explain potential strategies. An agentic system could potentially identify relevant obligations, compare permitted options, initiate an approved workflow and track progress. Now consider:

"I need a new credit card with better travel benefits."

The AI could potentially understand spending patterns, identify eligible products, explain tradeoffs, complete parts of the application journey and hand off where regulated decisions require human or institutional intervention.

Or:

"Make sure my rent and utility bills are covered every month."

An agent could potentially monitor the relevant account, identify upcoming obligations and execute permitted payments within customer-defined rules. The value is no longer measured in how naturally the AI speaks. It is measured in:

How much work does the customer no longer have to do?

That is the commercial promise of agentic AI.

This is where the industry's enthusiasm needs a reality check. Consumer adoption of financial AI does not equal consumer willingness to delegate financial decisions.

Forrester's September 2026 research makes this distinction explicit. Consumers are increasingly comfortable using AI to learn about financial topics, compare products, evaluate options and monitor their finances. Confidence falls when AI moves toward making financial decisions or acting on their behalf.

That creates a critical design principle for banks:

The future is unlikely to be fully autonomous banking. It is more likely to be controlled by autonomy.

Customers should be able to define boundaries. An agent might be permitted to monitor spending but not make payments. It might be permitted to pay bills below a certain amount but require confirmation above it. It might be allowed to move money between a customer's own accounts but not send money to a new beneficiary.

It might be allowed to prepare a loan application but not make the final credit decision. This is where permissions become a product capability.

This may be one of the most important changes brought by agentic AI. Today, customers interact with banking applications through screens. Tomorrow, they may increasingly interact through rules and intent. Instead of manually approving every small action, a customer could establish a policy:

"You can pay recurring bills up to ₹10,000 without asking me each time."

"You can move money between my savings and current account when my balance falls below ₹20,000."

"Never invest more than 5% of my available cash without confirmation."

The AI then operates inside those boundaries. This is a fundamentally different interaction model. The customer is not giving the AI unlimited autonomy. They are giving it conditional authority.

That distinction could become one of the defining principles of agentic AI in financial services.

India's digital payments ecosystem provides a useful illustration of where controlled AI execution could eventually go.

UPI processed 24.51 billion transactions worth ₹29.82 trillion in August 2026, according to Reuters. India is preparing a framework for agentic payments that could allow AI agents to conduct certain low-value payments without requiring individual approval for every transaction, with proposed controls including spending limits, identity checks, rule-based instructions and audit trails.

The significance is not that AI will suddenly control UPI.

The significance is that payment infrastructure is beginning to accommodate a world where the entity initiating a transaction may be an AI agent acting within customer-defined permissions.

That is a major conceptual shift.

Today:

Customer → Payment

Potentially tomorrow:

Customer → Rule → Agent → Authorization → Payment

The customer remains in control.

But the customer no longer needs to manually initiate every permitted action.

This shift also changes the metrics banks should use to evaluate conversational AI and agentic AI.

A chatbot can look successful because it handles millions of conversations.

But conversation volume does not necessarily equal business value.

A bank should increasingly ask:

What percentage of customer intents are successfully resolved?

What percentage of workflows are completed without human intervention?

How long does it take from intent to outcome?

How often does the AI escalate?

What is the cost per completed outcome?

How much revenue is generated or retained through AI-assisted journeys?

How many processes move through straight-through processing?

How often does the AI make an error requiring remediation?

What is the risk-adjusted ROI?

This distinction is especially important because Cambridge found that 76% of large financial institutions struggle to measure AI deployment value.

The industry therefore needs to move from:

AI adoption metrics

to:

AI outcome metrics.

The winning banking AI system may not be the one that handles the most conversations.

It may be the one that completes the most valuable customer outcomes safely.

The strategic response should not be "build a better chatbot."

Banks should start by mapping the customer intents that matter commercially.

Which questions lead to product discovery?

Which conversations lead to applications?

Which workflows generate the highest servicing costs?

Which journeys require multiple systems?

Which actions can safely be automated?

Which actions require explicit approval?

Which data must be available in real time?

Which APIs need to be exposed?

Which decisions need human oversight?

This creates an agent-readiness roadmap.

The first stage is making data accessible.

The second is making context available.

The third is connecting conversational AI to governed workflows.

The fourth is introducing granular permissions.

The fifth is enabling controlled execution.

The sixth is continuously measuring outcomes, risk and customer trust.

This is a much more practical way to approach agentic AI than attempting to make every banking process autonomous at once.

The themes surrounding Finov AI Summit Europe and the wider financial technology ecosystem increasingly point toward a question that is more consequential than chatbot adoption:

Who owns the financial relationship when AI becomes the customer's primary interface?

That question connects conversational AI, financial discovery, personalisation, open banking, embedded finance and agentic AI.

It also creates three possible futures for banks.

In the first, banks successfully build trusted conversational experiences and retain the customer relationship.

In the second, third-party AI platforms become the primary interface while banks remain the underlying financial infrastructure.

In the third, banks become deeply interoperable with multiple AI interfaces and compete based on the quality of their products, data, execution capabilities and trust infrastructure rather than simply the quality of their mobile application.

The outcome is not predetermined.

But the 31% and 24% Forrester figures suggest the transition has already started.

The financial-services industry has spent years trying to own the digital experience.

The next battle may be about something deeper:

Who understands what the customer actually wants?

If the customer says:

"I want to save more."

The winning financial institution will not simply show a savings account.

It will understand the customer's context.

If the customer says:

"I want to reduce my debt."

It will understand which actions could help.

If the customer says:

"Make sure I don't miss my bills."

It will understand the rules under which it can act.

That is where conversational AI evolves into agentic AI.

The technology moves from:

Answering → Understanding → Recommending → Planning → Acting

And the bank's role moves from:

Interface → Data provider → Product provider → Execution layer

The institutions that prepare only for the first transition may find themselves playing catch-up when the second arrives.

Forrester's latest research gives banks a number they should not ignore:

31% of consumers across the US, Canada and UK already use conversational AI for at least some personal finance questions.

Then comes the number that should make banking executives uncomfortable:

24% of those users are doing it outside their bank's ecosystem.

At the same time, Cambridge's 2026 global research shows that 81% of financial-services firms are adopting AI at some level, while 52% are already actively adopting agentic AI. Yet only 14% see AI as transformational to organisational strategy and competitive advantage, and 76% of large financial institutions struggle to measure AI value.

Put those numbers together and the picture becomes clear. Consumers are moving toward AI faster than banks are transforming around it. The immediate threat is not that AI replaces banks.

It is that AI becomes the place where customers discover, compare, question and decide before the bank ever enters the conversation. The longer-term opportunity is even bigger. If conversational AI owns the interaction and agentic AI eventually owns parts of the workflow, banks need to own what AI cannot easily replicate:

trusted financial data, regulated products, identity, permissions, execution, security and accountability.

That is why the future of conversational banking will not be decided by who builds the most impressive chatbot. It will be decided by who can connect:

Customer intent + trusted data + AI intelligence + permission + execution + measurable outcomes.

The bank that can do that will not lose the customer simply because the conversation started somewhere else. It will become the infrastructure that makes the conversation useful. And that may be the most important shift in banking AI heading into 2027.