Creating a Conversational Banking Experience

Scaling Conversational Banking to Resolve 40% of Customer Inquiries

Project Overview

Company: Scotiabank

Role: Product Design Lead

Team: Product (PM), Design (Product + Content), Engineering (Tech Lead, FE, QA), AI/ML (Data, ModelOps), Global Contact Centre leadership & frontline advisors

Tools & Methods: UX research, service design, conversation design, prototyping, workshops, analytics

I led the design of an AI-driven conversational banking experience within mobile. The objective was to reduce contact-centre volume while delivering a trusted, always-on support channel.

Rather than designing a chatbot feature, I structured a service layer connecting customers, AI systems, and live advisors.

---

The Problem / Challenge

Customer demand for in-app support surged while contact centres faced high volumes of repeatable inquiries.

• Customers required different levels of assistance
• Advisors needed protection from unnecessary volume
• AI required guardrails to avoid trust erosion

Constraints

• Strict compliance and financial regulations
• Variable AI confidence across intents
• Limited advisor availability

Without a structured system, chat risked frustrating customers or overwhelming support teams.

---

The Process / Approach

Research

• Call-driver and support-volume analysis
• Customer interviews and journey mapping
• Call listening sessions with frontline advisors
• Frontline advisors interviews
• Competitive audit
• Discovery workshops

I facilitated extended cross-functional workshops with design, content, AI/ML, engineering, call-centre managers, and frontline staff to identify root causes behind support demand.

These sessions uncovered recurring friction points — transaction confusion, authentication issues, and unclear product states — which directly informed system architecture.

Ideation & Iteration

• Defined intent categories and resolution types
• Mapped escalation logic between automation and humans
• Prototyped conversation flows and fallback patterns
• Validated scenarios with research and frontline staff

Key Decisions

Intent-Based Routing Over Linear Chat
Structured support around resolution outcomes instead of scripted conversations.

Escalation by Confidence
Defined AI thresholds for seamless human transfer before trust degradation.

Transparency Over Optimization
Communicated advisor availability clearly to manage expectations.

Chat as Orchestration
Redirected users to product functionality when conversation wasn’t optimal.

---

The Solution

Designed a scalable conversational framework built through sustained collaboration across design, AI/ML, content, engineering, and contact-centre operations.

• 156 structured intents and 36 guided conversation flows
• Confidence-based handoff to live advisors
• FAQ triaging and intelligent fallback handling
• Redirection to 10 in-app features
• 8 phone pathways and external resource routing
• 18 reusable conversational components
• Accessibility-first interaction patterns

The chatbot became an orchestration layer across the banking ecosystem rather than a standalone messaging feature.

---

Impact & Results

Customer Outcomes

• ~40% of inquiries resolved without human intervention
• 2M+ messages handled shortly after launch
• 70%+ response accuracy

Operational Outcomes

• Reduced repeatable contact-centre load
• Faster resolution for common banking needs
• Established scalable foundation for AI-driven support

---

What I Learned

Conversation Design is Systems Design

Designing chat required aligning AI logic, operational capacity, and customer expectations — not just writing flows.

Trust Drives Automation Acceptance

Customers tolerate automation when escalation is clear and accessible.

Transparency Reduces Friction

Setting expectations about wait times and availability prevented frustration more effectively than optimizing microcopy.

AI Needs Guardrails

Confidence thresholds and fallback patterns are critical to protect both brand and customer trust.