banking-assistant

What's the challenge?

▸ CONTEXT

The bank had been working on a chatbot for a while, before it was deprioritized. By the time the Conversational Banking team brought the project back to life, the team was mostly developers and the interface was mostly GUI interactions.

I was called in to advice the team on designing user experiences that made sense for a virtual assistant.

▸ OUR CHALLENGE

How can I integrate a conversation design approach into a dev-driven chatbot experience?

▸ ROLE

I acted as consultant, external to the project, advising UI designers and the product owner on conversation design (CxD).

▸ PROJECT DURATION

February - November 2022

Context

GUI vs CUI

I was first asked take a look at the components and a diagram of the dialogues. The feedback loops were making up almost 60% of the happy path, the language was robotic and it the interaction was mostly through graphic UI components in a chat window.

I documented my initial feedback, trying my best to educate on the difference between designing for conversational UIs instead of graphic UIs. This feedback was well received, getting the team to modify around 70% of the experience.

Original flows - The interactions prototyped on the chat window were predominantly graphic and form-like.

Feedback report - After my feedback on content, flow, components and overall best practices, the team acknowledge how designing for conversational user interfaces (CUIs) is different.

The product team didn't know there was design beyond graphic components. The experience was lead by business and development, putting the product's quality at risk.

▸ PROBLEMS DETECTED

> No value in having conversational channel if all components made more sense in a GUI website.
‍
> Over-complicated texts and flows that conflicted with the channel's best practices.
‍
> Siloed work: there was no communication with other bank areas that could later become obstacles.

Not enough hands

At this point the product team thought their UI designers could cover everything design-related: naming the bot, designing an avatar, the dialogues, the graphic assets. The problem was that their UI designers were external to the company.

I needed to convince them to bring in the bank's voice from other areas to avoid complications. So I pulled in marketing, institutional branding, and our design system and content teams from the bank's core UX area.

Having these conversations allowed us to notice potential blockers, their do's and their don'ts sooner rather than later. After that, the team agreed on having recurring meetings for CxD and UI design feedback with our bank's design team.

Before - The conversational banking product team was working with their own designers, with little to no contact with other areas.

After - We built a direct relationship with the design system and content teams. I remained as the CxD advocate. They now have brand validations early in their process.

Identity

Another major detail: we needed a name and personality for the bot. After some convincing, it became more clear to the team that the bot needs an identity to function and to scale.

Also, the name of the bot had to make sense. However, the product team suggested a list names that had little to no resonance or relation with the bank and picked their winner.

I talked them into hitting the 'restart button' and then facilitated an ideation workshop. I used previous research as context and important aspects of bot naming as triggers. The results were way better.

Voice and tone - We used the content team's voice and tone guidelines as a base for what our bot sounded like in different situations.

Personality design - Using canvases and the previous UX research, the designers and I managed to settle on a personality that made sense with the bank and the users' expectations. This style guide will drive their IVRs redesign as well.

Naming - I taught workshop participants the importance of having names that are easy to pronounce, remember and relate to the brand. We brought to the table dozens of name ideas, from which I curated the final 5 options to bounce off with legal and marketing.

Ready?

Here's the result

We went from this:

To this:

Cleaner, clearer, consistent and on-brand.

Outcomes + results

Significantly better conversations

Weekly feedback sessions allowed the bank's core UX team to be in constant communication with the chatbot's UI designers. Although not every battle was won, we managed to define better standards for component usage and dialogues.

Component guidelines - This was mainly the UI designer's work. My guidance was around the logic of the interactions and states of graphic components to fit this particular context.

Chit-chat - The product team came up with a set of training phrases for the bot, but I found many intents unrealistic or irrelevant. I dropped the number of intents from 44 to 16 by removing and merging. There's still plenty to do there, though.

Voice and tone - We built sample dialogues around three different tones for the bank's voice for starters: positive, neutral and negative moments. We now have content reviews and the conversation doesn't feel jumpy anymore.

This bot is currently under development. Unfortunately, we're not able to collect any insights until released production.

THE IMPACT

> Our design effort laid the groundwork for the incremental stages. Although there are still opportunity areas, this Q&A stage of the bot is solid enough to broaden the content and start planning for more complex dialogues.
‍
> The product and business team now acknowledges the importance of three disciplines (content, interaction and conversation design) and keeps us involved in the quality assurance reviews.
‍

Takeaways

▸ THINGS WE DID RIGHT

> Justifying all of my choices with theory and examples. This helped the team gain credibility on my insights.

> Challenging the team when the decision was arbitrary with data and a plan to reframe it.
‍

▸ THINGS WE CAN IMPROVE

> Making sure we're involved in the post-production reviews, not only pre-implementation.

> Test the bot with real users before implementation. That's one battle I lost. :(

> Balancing the amount of training phrases.

More projects

Conversational journaling experience

Team engagement analysis

Chatbot for banking

Ready?

Let's talk