Make the Case for a Banking Virtual Assistant: Who Wins and How?

A mid-sized retail bank wants to put a conversational assistant inside its mobile app and online banking site. Before anyone talks architecture, the product lead asks you to make the case: why is this worth building, and for whom?

You are expected to reason about two audiences separately:

1. The bank - operations, support, compliance, product and revenue teams.
2. The account holders - people checking balances, disputing charges, asking about fees, or trying to move money at 11 p.m. on a Sunday.

What to deliver

• A short list of concrete benefits for each audience. For every benefit, explain the mechanism that creates it (not just "it's more convenient") and the metric you would watch to prove it.
• The risks that are specific to a bank: the assistant states a wrong balance or due date, someone impersonates the account holder, sensitive data leaks into logs or a third-party model, regulators view an answer as financial advice, or the customer gets stuck in a loop with no human.
• A recommendation on scope: which requests the assistant should own, which it should assist with, and which must always go to a person.
• How you would decide, three months after launch, whether to expand it, keep it as is, or roll it back.

Assumptions you may make (state them)

• Around 4 million active digital customers; the contact center receives roughly 900k calls and 300k chats a month.
• About 60% of contacts are routine: balance, recent transactions, card lock/unlock, payment due date, branch hours, password reset.
• The assistant can call internal account APIs after the user is authenticated in the app.
• Human agents exist and can receive a handoff with conversation context.

Clarifying questions worth asking

• Is the first release informational only, or can it take actions (lock a card, schedule a payment)?
• Is the goal primarily cost reduction, customer satisfaction, or reaching customers on new channels?
• Are there populations (older customers, non-native speakers, customers in hardship) the bank especially wants to serve well?

Expect the interviewer to push on measurement: a bot that blocks people from reaching a human can make "containment" look excellent while satisfaction collapses.