Conversational Banking
Conversational Banking is the strategy of letting customers transact and get advice through natural-language interfaces — chat, voice, and messaging — rather than menus and forms. It encompasses voice assistants (Bank of America's Erica, Capital One's Eno), in-app chat, SMS-based interactions, and increasingly, agentic interactions where the assistant takes actions on the customer's behalf. KnowMBA POV: the leaders (BofA Erica with 2B+ interactions, Capital One Eno) succeeded because they treated conversational banking as a persistent product with real authority over journey redesign — not as a chatbot bolted onto an unchanged digital channel. The losers built bots to deflect calls without redesigning the underlying journeys, and ended up with chatbots that mostly say 'I don't understand, let me transfer you.'
The Trap
The trap is treating conversational banking as a deflection tool — measure success by 'calls deflected' and the bot will degrade to a search tool that gives up at the first ambiguity. Customers learn to type 'agent' immediately, defeating the purpose. The other common trap: launching a bot that can answer questions but cannot take action. A bot that can tell you your balance but cannot move money, dispute a charge, or freeze a card produces customer frustration, not value. Real conversational banking requires the bot to be a first-class action layer connected to core systems — which is an integration project, not a chatbot project.
What to Do
Build conversational banking in three layers: (1) Intent + Entity Layer — robust NLU that handles the top 50 banking intents (balance, transfer, dispute, freeze card, set up alert, etc.) with 90%+ recognition. (2) Action Layer — direct API integration with core banking, card systems, fraud, and authentication so the bot can DO things, not just answer. (3) Personalization Layer — proactive nudges based on transaction patterns ('we noticed unusual spending,' 'your bill is due in 3 days, want to schedule it?'). Measure success on (a) journey completion in-channel, (b) proactive insights delivered, (c) customer-initiated re-engagement — NOT on call deflection. Plan for 18-36 months and persistent investment; conversational banking is a product, not a project.
Formula
In Practice
Bank of America launched Erica in 2018; by 2023 BofA reported Erica had handled over 2 billion customer interactions and had 42 million users. Erica's expansion was deliberate — initial scope was simple balance and transaction queries, then bill negotiation and proactive alerts, then dispute initiation and account servicing. BofA invested heavily in the action layer (direct integration to core banking) rather than treating Erica as a deflection chatbot. Capital One's Eno (launched 2017) followed a similar architecture with a focus on proactive fraud detection and card management, including Eno's distinctive virtual card numbers feature.
Pro Tips
- 01
Measure 'actions completed' per interaction, not 'questions answered.' A bot that completes 2 actions per session is dramatically more valuable than one that answers 5 questions and refers everything to an agent.
- 02
Invest in proactive interactions. The most-loved capabilities of Erica and Eno are unprompted: 'we spotted a duplicate charge,' 'you'll be short on your bill date.' Reactive Q&A is a commodity; proactive insight is the differentiator.
- 03
Design hand-off carefully. When the bot escalates to a human, the agent should see the full transcript and the customer should not repeat themselves. Most conversational banking failures show up at this seam.
Myth vs Reality
Myth
“Conversational banking is mostly about NLP quality”
Reality
NLP recognition above ~92% is table stakes — every major vendor delivers it. The differentiation is in the action layer: how many real banking actions can the bot perform without escalation. A bot with great NLP and no actions is a search engine; a bot with adequate NLP and 50+ actions is a banking assistant.
Myth
“Customers want to chat with their bank”
Reality
Customers want their tasks done. Conversational interfaces win when they are the fastest path to action. Chat that takes longer than the app loses. The discipline is choosing when chat is faster (open-ended questions, multi-step actions) and when it isn't (simple lookups).
Try it
Run the numbers.
Pressure-test the concept against your own knowledge — answer the challenge or try the live scenario.
Knowledge Check
A bank's chatbot has 95% intent recognition accuracy but only 18% of customer journeys complete in-channel — most escalate to a human agent. What is the most likely root cause?
Industry benchmarks
Is your number good?
Calibrate against real-world tiers. Use these ranges as targets — not absolutes.
Conversational Banking In-Channel Journey Completion
% of conversations resolved without human escalation, mature deployments (24mo+ in production)Leader (Erica/Eno class)
> 60%
Strong
45-60%
Average
25-45%
Weak
10-25%
Failed Deployment
< 10%
Source: Cornerstone Advisors / Forrester Conversational Banking Reports 2022-2023
Real-world cases
Companies that lived this.
Verified narratives with the numbers that prove (or break) the concept.
Bank of America (Erica)
2018-2023
BofA launched Erica in 2018 as a voice/chat assistant inside the BofA mobile app. The team deliberately scoped initial release narrowly (balance, transaction lookup, simple alerts) and expanded capability quarterly: bill negotiation, dispute initiation, proactive fraud notifications, account servicing. By 2023, BofA reported Erica had crossed 2 billion interactions and 42 million users. The differentiator was investment in the action layer — Erica is wired directly into core banking, card systems, and fraud — rather than treating it as a deflection chatbot.
Total Interactions (by 2023)
2B+
Users (2023)
42M
Launch Year
2018
Architecture
Action-layer integrated, not deflection-only
Conversational banking succeeds when the bot can do, not just answer. BofA invested in integration depth from day one — that is the moat.
Capital One (Eno)
2017-Present
Capital One launched Eno in 2017 with a focus on proactive interactions — duplicate charge detection, unusual spending alerts, free trial conversion warnings — and a distinctive feature: virtual card numbers generated on-demand for online purchases. Eno was architected around action and proactivity from the start, not Q&A. Capital One has publicly described Eno as a product with persistent investment, not a project, and has expanded its capabilities annually.
Launch Year
2017
Distinctive Feature
Virtual card numbers
Strategic Frame
Proactive insight + action
Operating Model
Persistent product, annual capability expansion
Eno's enduring relevance comes from proactive, action-oriented design — not from chat as a primary interface.
Decision scenario
The Bot Architecture Decision
You are head of digital at a $400B-asset bank. Your team has 10 months and $12M to ship a conversational banking experience. You can buy a best-in-class chatbot vendor (3-month deploy, limited integration depth) or build with deep core banking integration (10-month deploy).
Budget
$12M
Timeline
10 months
Monthly Customer Contacts
8M
Current Agent Share
62%
Decision 1
Choose the architecture.
Buy the chatbot vendor — ship in 3 months, leave 7 months for marketing and refinementReveal
Build with deep core integration — accept the 10-month timeline; launch with action capability from day one✓ OptimalReveal
Related concepts
Keep connecting.
The concepts that orbit this one — each one sharpens the others.
Beyond the concept
Turn Conversational Banking into a live operating decision.
Use this concept as the framing layer, then move into a diagnostic if it maps directly to a current bottleneck.
Typical response time: 24h · No retainer required
Turn Conversational Banking into a live operating decision.
Use Conversational Banking as the framing layer, then move into diagnostics or advisory if this maps directly to a current business bottleneck.