The App Is Not Disappearing. It Is Becoming an Agent.

81 per cent of financial services firms are now adopting AI at some level, and 40 per cent report advanced adoption, according to the Cambridge Centre for Alternative Finance's 2026 Global AI in Financial Services Report, which surveyed 628 institutions, vendors, and regulators across 151 jurisdictions. The debate inside most of those institutions is no longer whether AI belongs in the customer journey. It is what the app itself is supposed to become. The honest answer is not that the app disappears. It is that the app becomes an agent.

What Is an Agentic Financial App?

An agentic financial app is a banking, lending, insurance, or wealth app in which an embedded AI agent understands the customer's live context, detects friction as it happens, guides the next decision, and takes authorised action inside existing workflows, rather than simply displaying screens and waiting for the customer to navigate them correctly.

Key Takeaways

The shift, and what it does not mean:

• Financial apps are not being abandoned. Cambridge's 2026 research puts AI adoption at 81 per cent across surveyed institutions, and the shift is happening inside existing digital channels, not away from them.
• The real cost most BFSI leaders are underpricing is friction: KYC drop-off, mandate failures, loan and credit card abandonment, policy renewal lapses, and SIP setup abandonment all leak revenue that never shows up as a clean loss line.
• Static walkthroughs, FAQ chatbots, and delayed outbound calling all treat friction as something to explain after it happens. An in-app agent is built to catch it while it is happening.
• The core banking and record systems are not going anywhere. What is becoming agentic is the workflow and interface layer that sits between the customer and those systems.
• Governance is not a constraint on this shift. According to the World Economic Forum's 2026 AI Playbook for Financial Services, trust, governance, and human oversight are what determine whether institutions can scale AI past the pilot stage at all.

The Friction BFSI Leaders Are Still Underpricing

Every BFSI product leader already tracks conversion funnels. Far fewer treat the gap between customer intent and completed journey as a single, addressable revenue line, rather than a dozen separate problems owned by a dozen separate teams.

• KYC and onboarding abandonment: a customer gives up during document upload, video verification, or a form field asking for information they do not have on hand.
• Mandate and payment failures: an e-mandate or auto-debit setup fails silently, and the customer only discovers it when a payment bounces weeks later.
• Loan and credit card drop-off, concentrated at document collection and eligibility clarification, exactly the steps a static screen cannot self-serve.
• Policy purchase and renewal friction: a policyholder intends to renew, opens the app, cannot locate the right screen, and closes it.
• First-investment and SIP abandonment: a customer is asked to make five sequential decisions (fund, amount, date, mandate, KYC) with no guidance between them.
• Feature discovery failure: a capability the institution already built goes unused because nothing tells the customer it exists at the moment it would matter.
• Context loss when a customer moves between the app, a support call, and WhatsApp, re-explaining the same problem to each one.

None of these are new problems. What has changed is that they are now measurable, and increasingly viewed as an operating layer problem rather than a UX polish problem.

What Global Banking Leaders Are Already Saying About Agentic AI

The clearest signal that this is a strategic shift, not a vendor pitch, is how senior BFSI leaders are already describing it.

Marianne Lake, JPMorganChase's CEO of Consumer and Community Banking, has been notably measured about how fast agentic commerce moves from experimentation to production, telling investors the bank sees a longer road to broad adoption even as it moves ahead with agents for specific tasks, including a consumer-facing travel agent JPMorganChase aims to pilot (according to Banking Dive, June 2026). The bank's own investor materials describe AI already contributing to a 20 per cent increase in private banking gross sales, with room to expand banker client coverage by as much as 50 per cent (JPMorganChase investor materials, June 2026).

DBS Group CEO Tan Su Shan has described DBS's preparation for agent-to-agent banking, and the bank reported its AI and data initiatives generated roughly 1 billion Singapore dollars in economic value in 2025. On an earnings call, she said DBS was seeing "the green shoots of AI working in generating ideation that generates fees" (CNBC, November 2025). DBS's Gen AI assistant, DBS Joy, now lets corporate users move beyond retrieving information to executing everyday banking tasks in one conversation, serving more than 10 million customers across three markets (DBS Newsroom). Some coverage of DBS's own vision has framed a future where the standalone app plays a shrinking role as agent-to-agent transactions grow. RevRag AI reads the same evidence differently: DBS's most concrete result so far, AI-assisted relationship managers working inside existing banking workflows, is itself an in-app agent story. The more defensible prediction is not that the app goes away. It is that the app becomes the surface the agent operates through.

At Standard Chartered, Allan Song, Head of Data and Digital for Financing and Securities Services, has written that agentic AI "represents a new operating model, moving beyond task automation and into autonomous execution with embedded governance," letting institutions redeploy human expertise toward higher-value decisions (Standard Chartered, April 2026). Embedded governance, not autonomy for its own sake, is the operative phrase.

In India, Ashok Chandra, MD and CEO of Punjab National Bank, used the bank's own inaugural AI Summit 2026 to set out PNB's view that as the bank moves toward generative and agentic AI, its responsibility as a public sector bank is to adopt it with transparency, accountability, and auditability, in line with the RBI's AI framework (India TV News, August 2026). That same emphasis on governed autonomy runs through the World Economic Forum's 2026 AI Playbook for Financial Services, developed with Accenture from 18 months of roundtables with more than 150 leaders across over 100 institutions: as firms move past pilots, success depends on combining speed with security and keeping human judgement at the core (World Economic Forum, June 2026).

Why Static Walkthroughs, Chatbots, and Call Queues Fall Short

A static in-app walkthrough assumes every customer hits friction in the same place, in the same order, for the same reason. Most do not, and a walkthrough built for last quarter's drop-off point says nothing about the one a customer is stuck on today.

An FAQ-style chatbot answers a question the customer already knew how to ask. It cannot see which screen the customer is on, whether they have already failed an action twice, or whether a mandate registration is the actual blocker behind the question. It resolves the query. It does not resolve the journey.

Delayed outbound calling catches drop-off after the moment has passed. A customer who abandoned a loan application at 11 a.m. and gets a callback at 4 p.m. has usually already decided something else, often with a competitor. And each time a problem crosses from the app to a call to WhatsApp, the customer re-explains it from zero.

Traditional App vs Chatbot vs In-App AI Agent

Across five dimensions, the difference is structural, not cosmetic:

• Awareness of context: traditional app, none beyond the current screen. Chatbot, whatever the customer types. In-app AI agent, live journey state: screen, action history, hesitation, failed attempts.
• Response to friction: traditional app, silent until the customer gives up. Chatbot, waits to be asked. In-app AI agent, detects hesitation, inactivity, or a failed action, and intervenes proactively.
• What it can do: traditional app, displays forms. Chatbot, answers FAQ-style questions. In-app AI agent, guides a decision and takes an authorised action inside the workflow.
• Continuity across channels: traditional app, none. Chatbot, typically resets per channel. In-app AI agent, shares context across in-app, voice, and WhatsApp while the app stays the system of record.
• Accountability: traditional app, no decision to account for. Chatbot, logs a transcript. In-app AI agent, permissioned and audited actions with a defined human handoff for exceptions.

What an Agentic Financial App Actually Does

An agentic financial app does four things a static one cannot. It understands the exact screen and journey the customer is in, not a generic session. It detects friction as a signal, not an eventual support ticket: hesitation on a field, a repeated failed tap, an inactive session at a decision point. It responds with guidance plus an authorised action inside the existing workflow, not a deflection to an FAQ article. And when a case genuinely needs a person, an exception, a dispute, a judgement call outside its permissions, it hands off with full context attached, rather than making the customer start over.

Stated plainly: an agent automates effort, never accountability. Every action should be permissioned, logged, and traceable to a specific decision point, the same discipline Ashok Chandra and the WEF's Playbook describe at the institutional level, applied at the level of a single interaction.

RevRag AI builds in-app AI agents on exactly this model for BFSI institutions: agents that read the customer's live context inside the app, connect to existing core banking, loan management, policy, and CRM systems rather than requiring a rebuild, share context across in-app, voice, and WhatsApp touchpoints while keeping the app itself the central, permanent surface, and hand off cleanly to a human for exceptions, with full audit trails for product, risk, and compliance teams.

Six Predictions for the Agentic Financial App Era

Where the app-to-agent shift is headed next:

• Navigation gives way to intent. Within two to three years, a leading BFSI app's top journeys (onboarding, a mandate registration, a policy renewal) get completed by stating what the customer wants, not by locating a menu item.
• One app, several coordinated agents. Apps will run multiple specialised in-app agents (onboarding, collections, cross-sell) coordinated through a single governed layer enforcing consistent permissions and audit standards across all of them.
• Metrics move from clicks to completion. Product dashboards report resolution rate, time-to-outcome, and recovered revenue as primary metrics, with page views and session length as secondary signals.
• Product, support, and operations converge at the point of friction. The split where product owns the app, support owns the queue, and operations owns the back office compresses around whichever team owns the friction point, since the agent surfaces the same signal to all three at once.
• Core systems stay. The interface layer becomes agentic. Core banking, policy administration, and loan management systems remain the system of record. What changes is the layer between the customer and those systems.
• Human effort concentrates in judgement, not repetition. Relationship managers, underwriters, and support staff spend a growing share of time on empathy, negotiation, and exceptions, the categories an agent is deliberately not permissioned to resolve alone.

What This Means for BFSI Product and Risk Leaders

The practical decision most BFSI institutions face is not whether to adopt agentic AI. Cambridge's 2026 data already puts adoption at 81 per cent. It is whether the agent layer gets built with the app as its foundation, permissioned and audited from the first release, or bolted on afterward as a chatbot that cannot see what the customer is doing.

Institutions treating this as an extension of the app they have already built, not a separate channel competing with it, are the ones reporting measurable results: JPMorganChase's gains in private banking sales, DBS's roughly 1 billion Singapore dollar AI value in 2025, and PNB's own public commitment to auditability from the outset. The rest will spend the next two years rebuilding what they should have built once.

Frequently Asked Questions About Agentic Financial Apps

What is the difference between an in-app AI agent and a banking chatbot?

A chatbot answers questions a customer types, using a fixed script or a retrieval system, and has no visibility into what the customer is doing on screen. An in-app AI agent has live journey context: the screen, the action history, and any friction signals, and can take an authorised action inside the workflow, not just provide an answer.

Will agentic AI replace the need for a banking or insurance app?

No. Core banking, policy, and loan systems remain the system of record, and the app remains the customer's primary point of engagement. What changes is that the app's interface and workflow layer becomes agentic, proactive, conversational, and action-taking, rather than the app being replaced by another channel.

How does an in-app AI agent detect friction before a customer abandons a journey?

It reads signals a static screen cannot: hesitation on a specific field, a repeated failed tap, an inactive session at a decision point, or a stalled multi-step process, and responds with guidance or an authorised action before the customer gives up, rather than waiting for a support ticket after the fact.

What is agent-to-agent banking?

Agent-to-agent banking describes a future state where an AI agent representing a customer and an AI agent representing an institution interact directly to complete a transaction, a model DBS Group has publicly said it is preparing for. It describes a new transaction channel, not the removal of the underlying banking app or the institution's system of record.

How do BFSI institutions keep agentic AI accountable and auditable?

By treating every agent action as permissioned, logged, and traceable to a decision point, with a defined handoff to a human for exceptions or disputes, the same principle Punjab National Bank has tied to its own agentic AI rollout, and what the World Economic Forum's 2026 AI Playbook identifies as the deciding factor in scaling AI past pilots.

What BFSI workflows benefit most from an in-app AI agent first?

Workflows with high abandonment and a known customer intent benefit fastest: KYC and onboarding completion, mandate and payment setup, loan and credit card application recovery, policy renewal, and first-investment or SIP setup, since each has a clear, measurable drop-off point an agent can catch in real time.

Does adding an in-app AI agent mean fewer human relationship managers or support staff?

No. It changes what they spend time on. Agents absorb high-volume, well-defined interactions, freeing relationship managers and support staff to focus on exceptions, negotiation, and judgement calls that genuinely require a person, the shift JPMorganChase, DBS, and Standard Chartered are each describing in their own commentary.

How is an agentic financial app different from adding generative AI features to an existing app?

Generative AI features typically answer or summarise within a single screen. An agentic financial app goes further: it maintains context across the customer's full journey, detects friction proactively rather than waiting to be asked, and takes an authorised action inside an existing workflow rather than only generating a response.