AI Won’t Deliver ROI Until Banks Redesign How Work Gets Done
By Nicolás Kaplun, CEO of the Financial Services AI Studio at Globant
Simple Subscribe
Subscribe Now!
AI investment across retail banking is accelerating — and so are expectations.
Boards expect measurable productivity gains. Regulators expect tighter oversight. Competitors, both fintechs and large incumbents, are racing to operationalize AI beyond pilots.
Yet, inside many institutions, measurable results remain uneven. In some cases, they’re invisible. Copilots are deployed, and proofs of concept are celebrated, but enterprise-wide ROI is elusive. Productivity improvements are hard to quantify. Risk leaders raise valid concerns. Business units struggle to scale experimentation into repeatable value.
Key Insight: AI does not stall because the models are immature. It stalls because banks have not redesigned work, accountability and governance to support hybrid execution.
The institutions that lead in the next era of retail banking will not simply deploy AI tools. They will deliberately rebuild their operating models around a human–AI hybrid workforce — and measure that transformation in business terms.
Need to Know:
- AI pilots struggle when layered onto unchanged legacy workflows.
- Clear separation of execution, judgment and accountability accelerates enterprise adoption.
- A new leadership capability — the agent manager — is emerging as an essential capability.
- Legacy infrastructure and fragmented data increase risk before value materializes.
- Workforce transformation must show up in cycle time, cost-to-serve and throughput metrics.
AI Fails When It’s Layered onto Legacy Workflows
Across retail banking operations — underwriting, fraud, AML, onboarding, disputes and servicing — a significant share of daily work requires interpreting signals across multiple data sources, policies and customer contexts.
AI is particularly valuable in these gray areas, where it can synthesize information and provide insights that help humans make better decisions.
It can assist by:
- Summarizing documentation
- Surfacing patterns or anomalies across transactions
- Drafting communications
- Classifying and organizing cases
- Prioritizing alerts based on multiple signals
Humans, however, excel at contextual judgment, relationship management, regulatory interpretation and accountability.
The mistake many institutions make is deploying AI without redefining ownership, escalation and accountability.
Why It Matters: Without clear coordination around roles and responsibilities, AI adoption becomes fragmented across teams and individuals. Employees experiment with AI in isolated tasks, risk teams raise concerns about oversight, and executives struggle to translate scattered productivity gains into measurable ROI. Coordination, not just experimentation, is what turns AI usage into enterprise impact.
Banks that scale successfully separate responsibilities by design.
Leaders should:
- Map high-volume workflows and explicitly separate execution tasks from judgment-based decisions.
- Explicitly define what AI owns and what humans own.
- Maintain clear human accountability for regulated decisions.
- Tie compensation and performance KPIs directly to AI-enabled productivity gains.
AI can handle repetitive execution at machine speed while helping surface insights for more complex decisions. Humans remain responsible for contextual judgment, relationships and accountability.
The Rise of the Agent Manager
As AI systems evolve from assistive tools to semi-autonomous agents capable of initiating actions within defined guardrails, the way institutions manage and coordinate work must evolve as well.
A new capability is emerging inside forward-looking financial institutions: the agent manager.
Agent managers are not necessarily engineers or data scientists. They are operations leaders, product managers, compliance officers and risk professionals responsible for coordinating digital and human work across critical processes.
Their responsibilities include:
- Defining agent permissions and scope.
- Monitoring performance and error rates.
- Managing escalation pathways.
- Ensuring explainability and audit readiness.
- Continuously optimizing performance.
Key Insight: Without clear coordination around how humans and AI interact, scaling AI can trigger governance anxiety, slow deployment and inflate perceived risk. Effective agent management is not only about oversight, but about enabling structured collaboration — where humans review, challenge and refine AI-generated outputs to arrive at stronger decisions.
When this collaborative supervision is institutionalized, banks gain both speed and control simultaneously.
To operationalize agent management, institutions should:
- Assign a named executive owner for every AI-enabled workflow.
- Create clear override and escalation protocols.
- Establish performance dashboards tracking AI output quality.
- Integrate AI governance into existing risk committees.
Scaling AI is not purely a technical transformation. It’s a governance and leadership transformation.
Modernization Is the Hidden Multiplier
AI exposes infrastructure weaknesses faster than most institutions can remediate them.
Established banks often operate on legacy cores and fragmented data architectures optimized for stability rather than agility. Data access may not be real time. Integration layers may be brittle. Compliance frameworks may assume deterministic systems rather than probabilistic outputs.
Fintechs, while digitally native, often scale rapidly on layered architectures that were not originally designed for enterprise-grade governance or regulatory scrutiny at scale.
Why It Matters: Without clean data flows, resilient APIs and adaptable controls, AI increases operational and compliance risk before it increases productivity.
Modernization is not optional. It is the prerequisite for scalable AI.
Banks serious about enterprise transformation must first address the technical debt embedded in legacy systems and fragmented data architectures. This requires deliberate modernization across core workflows:
- Audit data integrity and real-time availability across priority workflows.
- Strengthen API integration between core systems and AI services.
- Redesign control frameworks to address probabilistic outputs.
- Embed AI directly into operational systems rather than isolating it inside tools.
Modernization unlocks non-linear productivity — increasing throughput without proportional headcount growth. That’s where structural advantage emerges.
The traditional growth model in banking has been largely linear: more volume requires more people. Institutions that modernize can break that equation — increasing output, responsiveness and innovation velocity without proportional cost expansion.
Measure Workforce Transformation in Business Terms
Upskilling initiatives are often measured in certifications or training completions. That does not prove transformation.
Key Insight: If AI does not reduce cycle times, lower cost-to-serve or improve risk outcomes, it’s not transformation — it’s experimentation.
Banks should measure AI-driven workforce change across three business dimensions:
1. Capability Adoption
- Are employees AI-literate in role-specific ways?
- Are agent managers formally trained and accountable?
- Is AI integrated into daily workflow tools?
2. Workflow Transformation
- How have output quality, resolution speed and decision consistency improved across priority processes?
- Has case resolution time decreased?
- Have manual error rates declined?
- Has fraud detection accuracy improved?
3. Business Performance
- Has time-to-market accelerated?
- Has cost-to-serve decreased?
- Has throughput increased without proportional hiring?
- Have customer satisfaction,retention and lifetime value improved?
When workforce capability links directly to measurable outcomes, AI investment becomes defensible at the board level and sustainable under regulatory scrutiny.
Reframe AI as Capacity Creation, Not Cost Reduction
Cultural framing determines adoption speed.
When AI is positioned primarily as headcount reduction, resistance intensifies and adoption slows. Employees interpret it as a threat rather than an enabler.
Forward-looking institutions position AI as capacity creation and revenue expansion.
Why It Matters: When AI accelerates parts of the workflow, it frees up human capacity that can be redirected toward higher-value activities that drive growth and resilience.
Winning institutions reinvest AI capacity into:
- Deepening advisory relationships and helping clients improve their financial health.
- Expanding personalized financial guidance.
- Strengthening fraud and compliance oversight.
- Accelerating product development cycles.
- Improving customer response times.
The competitive divide forming in retail banking will not be defined by who experiments with AI first. It will be defined by who redesigns work fastest and responsibly.
The Bottom Line
AI alone will not determine which banks win the next decade. Operating model redesign will.
If organizations want to create operating leverage competitors cannot easily replicate, they must:
- Clarify human–AI accountability.
- Institutionalize agent management.
- Modernize infrastructure and address technical debt.
- Measure transformation through ROI metrics (cycle time, cost-to-serve, throughput and risk outcomes).
Consider a common banking workflow such as loan origination. AI can rapidly analyze financial documents, summarize borrower information and surface risk signals, while human experts challenge those insights, apply contextual judgment and make the final decision. When this collaboration is coordinated across the workflow, approval times can shrink, decision quality can improve and teams can handle greater volume without proportional staffing increases.
Those that treat AI as incremental automation risk remaining stuck in pilot mode, while institutions that redesign how humans and AI work together will reset productivity expectations across the industry.
The technology is ready. The question is whether your operating model is.
