Financial Services Chatbot MVP

 

There were 7 business days left until our scheduled launch. After months of project turmoil, the product owner had been replaced with a more junior analyst, completely new to the project. Because of widespread interest and buy-in across departments, the list of desired capabilities for MVP had been in constant flux since the project’s birth. With the deadline looming, leadership confirmed 12 topics for the bot to cover.

 
 

Although users will be able to type freely, product requirements mandated that all topics be reachable using only ‘suggested action’ buttons. Users must also always have quick access to a live agent. Technical limitations allow us to display exactly 3 buttons with every bot response. None of these buttons may exceed 25 characters.

 
 

Required

MVP

Topics

 

Find my Access Number
Find my account numbers
Activate my card
View products & rates
Update my contact info
Open a new account
Set up direct deposit
Set up automatic payments
What’s Zelle?
Get started with Zelle
What’s my routing number?
Apply for a loan

For the past 12 months, I had collaborated with 3 different analysts, each tasked with a separate set of ‘intents’ for the bot to address. Synthesizing my work with each of the analysts, it became clear that we needed to address a number of scenarios that were vital to the experience:

  • greetings for new and returning users

  • opportunities to transfer to a live agent

  • guidance when live agents aren’t available

  • a variety of “I don’t understand” responses

  • explanations the bot’s functionality

  • assurance of security in conversing with the bot

  • farewell messages, appropriate to the nature of the call’s termination

I had already written text covering these scenarios, but now the task at hand was fitting them all together. Because of the wide variety of stakeholders, the topics that were prioritized for launch weren’t all directly related to one another. Still, everything needed to connect logically.

I proposed a system of categorization, using menus and a branching content structure. We added 2 category buttons, and by updating certain responses to imply connections between the topics of direct deposit, account numbers & routing number, I proposed a cyclical flow that met all requirements while remaining within our technical limitations.

What would I do differently?

The description above is not meant to cast blame toward the project’s leadership or any of its stakeholders. Nor is it meant to suggest that I, with the support of 3 noble business analysts, saved the project just in time for release. Truly, this was a project where almost everything seemed to go wrong.

As for my involvement:

  1. This is not how AI-enabled chatbots are supposed to work. I would have insisted on this point, even if it meant delaying the MVP release.

  2. More user research and validation testing throughout the whole design process. Always.

  3. Sterner insistence on the values of SAFe Agile practices, even if it led to more awkward, uncomfortable, and even hostile project meetings. There were many of these as it was.

  4. I would have asked to move off the project. Social dynamics during meetings and between project team members separately reached a level of volatility I cringe to think back on. My fascination with AI and its mechanics chained me to the project. A few months after I moved on from this organization, I learned the project had been canceled completely.

Next

On the Border