top of page
Finance Sector: AAI/ML for Banking Operations: 2022 - ongoing
Spearheading AI transformed workflows for cross-LoB Operations platforms
For NDA reasons I cannot share images
Image by Benjamin Davies

Project Overview

Role: Executive Director of Design for AI, CIB Operations

Team: AI Experience Design (Cross–Line of Business; Payments-first implementation), partnering with AI and Product leads.

Objectives: Develop an AI framework for accelerated product delivery. Define and scale an AI-first strategy for query and exception management in CIB operations, targeting a 30% reduction in manual Operations workload. Upskill wider CIB design team in designing for AI-first and AI augmented products. 

1. Executive Summary
In 2022, I joined the Corporate & Investment Bank (CIB) Operations with a mandate to guide the design strategy for AI and Machine Learning across Markets, Payments, and Data Operations, and to grow a team to support AI product development. One of the earliest insights was that the organisation could not meaningfully scale AI because its operational ecosystem was fragmented, inconsistent, and poorly understood.
 

This case study outlines how I:​

  • Spearheaded the first universal operations workflow for CIB Operations

  • Defined a foundational model for AI-first Operations, including automation, human governance, and feedback loops

  • Managed the end-to-end redesign of the Proof of Concept

  • Led a distributed team of designers and design managers to roll-out delivery across multiple LoBs

  • Established a framework now used to scale AI-driven operations across multiple lines of business
     

This work represents a shift from tool-based problem solving to system-level operating model design, creating the conditions for AI to function reliably across one of the world’s largest financial institutions.

2. The Strategic Challenge: No Shared View of Operations

Upon joining, I found that neither Design nor Engineering possessed a clear picture of the Operations ecosystem:

  • No inventory of existing tools

  • No understanding of overlapping workflows across LoBs

  • No clarity on which capabilities should be standardized

  • No shared taxonomy or data structures
     

The result was a sprawling landscape of hundreds of bespoke tools, each built for localized needs, but all performing similar operational tasks.
 

My strategic hypothesis

The variety of tools did not reflect diverse user needs. It reflected:

  • inconsistent data

  • inconsistent taxonomy

  • inconsistent governance

And without a unified model, AI could only ever be deployed in isolated pockets—never at scale.

3. Designing the Universal Operations Workflow

With a team of senior UX and Service designers, I led the creation of the first cross-LoB operational framework, mapping:

  • the lifecycle of a query, exception, or investigation

  • the common triage, treatment, and resolution steps

  • opportunities for predictive automation and agentic flow

  • where human oversight was essential
     

This became the foundational blueprint for:

a, AI-first capability prioritization

Which steps could be automated, predicted, or enriched through ML/LLMs.

b, Agent-driven process design

How AI should orchestrate work in sequence-autonomously when possible, with human governance where necessary.

c, Human-in-the-Loop (HITL) model

Defined roles and user experiences frameworks for validation, exception review, and model training.

d, Enterprise alignment

Provided Data, Engineering, and Operations with a shared vocabulary for how work moves across the organisation.
 

This was the first time CIB Operations had a strategy-level map that could scale AI horizontally across teams and platforms.

4. Proof of Concept: Localised Payments use case

Payments Operations was a first testing ground. Analysis revealed:

  • 25–35% of operator time was spent on triage alone

  • Duplicate effort across global teams

  • Fragmented systems with no unified audit trail

  • Significant delays caused by incomplete or missing data
     

Strategic Objective

Reimagine a payments settlements  investigation use case as an AI-first, Human-in-the-Loop system, where AI performs the majority of work, and humans intervene primarily for governance, validation, and exception handling.

5. The AI-First Paradigm Shift

Aspect

User Actions​

​

Process
 

Visibility

Human Role
 

Manual Workflow

15+ actions per case

​

Manual triage, enrichment, routing
 

Fragmented insight

Performs all steps
 

AI-First Workflow

0-2 actions per case

​

AI complete the full workflow end-to-end
 

Real-time monitoring & structured feedback loops

Supervises, validates, trains AI
 

This shift redefined the nature of operational work—from task execution to AI stewardship and judgment-based intervention.

6. Designing the AI-Driven Process Flow

My team led the definition of an AI-first process layer, where each stage is modular and can be reused across LoBs: This design enabled a consistent approach to automation, regardless of regional rules, product types, or data sources.
 

Humans intervene only when:

  • confidence scores fall below thresholds

  • ambiguity is detected

  • AI requires reinforcement for improvement


​This is the core of the HITL governance model.

7. Designing Human + AI Collaboration

a. Team Lead & Manager Workflow

A lead designer was assigned the creation of a leadership-oriented user experience enabling:

  • real-time team capacity views

  • monitoring of work distribution

  • predictive insights for staffing & rebalance

  • tools to adjust prioritization and SLAs
     

b. Building Trust & Transparency

Early testing by design researchers revealed a critical insight:

Adoption would fail if users couldn’t validate or challenge AI decisions.

To solve this, I my team developed a dual visibility model:

  • Team dashboards for oversight, intervention, and governance

  • User queues for transparency, clarity, and trust

This reduced reconciliation issues, increased adoption, and improved confidence in automation outcomes.

8. Organisational Outcomes & Enterprise Impact
Although this POC focused on Payments, the frameworks and models were designed to scale across CIB. This initiative now serves as a blueprint for AI-first operations design at JPMorgan - demonstrating how automation, human judgment, and UX clarity can co-exist at scale.
 

Key Achievements
 

Operational Impact

✅ Identified pathways to 25–35% reduction in manual triage work

✅ Streamlined workflows across global operations
 

Enterprise Strategy

✅ Introduced the first scaleable / reusable AI Ops framework

✅ Established HITL patterns for future AI workflows

✅ Improved oversight and user trust

​

Cross-LoB Adoption

✅ Prototype for Horizon 3: AI + Human collaboration at enterprise scale

✅ Payments POC became the blueprint for AI-first design xLoB
 

Design Maturity

​✅ Created enterprise-grade patterns for AI governance, model feedback, and agent orchestration

✅ Strengthened design for AI design among broader CIB team and the distributed AI design organization

© Máire Flynn 2025

bottom of page