Oct 2018 – Nov 2019
I worked for one of the largest, most well-known banks in the U.S. (which prefers not to be named here, though I'm sure you can guess). So it makes sense that it's got many different customer segments who want a high level of service and assistance. How do we deliver that level of service? Well, one way is with highly trained chatbots, armed with personalized context, that can quickly help customers at any time of day, wherever they are.
I worked on a chatbot platform with the goal of enabling any line of business at the bank to create a custom conversational bot. The team had previously launched bot pilots, so this was a tested idea.
I joined the project as the second designer, but soon became the sole designer, and then the main full-time designer when another joined me part-time. I worked closely with product managers, engineers, and QA throughout.
Web app optimized for desktop
Since I joined when the project was already in full swing, my main task in the beginning was to get up to speed as quickly as possible.
Once I was fully immersed and working independently, I met often with key collaborators - particularly the product manager and the front-end lead - to discuss the next tasks that had already been roughly laid out in previous sprint planning. Quick iteration was key here.
Overall, the process that I experienced doesn't strictly adhere to the "ideal" design process - but that's just reality sometimes. It wasn't just design having to be more flexible about process; the whole team had to exercise their judgment along the way. The good news is that I became much more observant of processes that worked and didn't work, and much quicker to make suggestions.
Here's a summary of what the process looked like for me:
I learned how conversations are structured and created, as well as the goals and existing processes of the project. Since I'd never done conversational design before, there was a bit of learning curve.
While we had a high-level roadmap and preexisting architecture to work from, many of the details still had to be hashed out. I created mockups based on discussions and in-progress requirements, which went to critique with the core team - and we repeated this loop until everything was ready.
Our users were internal teams, so we had more access to them than we might have otherwise. We built relationships with them, both informal and formal (so as to limit unintended bias). Since the platform and the work of creating a bot is very complex, having ongoing relationships with the same users was extremely advantageous - we would have had a much harder time recruiting the right users from the general population.
We worked in an agile structure, but due to the size of the project and the team, I had to take the impact of even seemingly small changes very seriously. Luckily, I had close collaborators who were more than willing to fill me in.
Because I inherited the project as the sole designer a couple months in, I thought a lot about design's role within the project. I'm continuing to work on ways to improve on this front - including but not limited to a design system, component library, and aligning design versions with development versions. I think that with any project, one of design's responsibilities is to build a sustainable structure that allows designers to be creative, but still efficiently deal with all of the constraints.
Imagine a conversation in your head, and there are just about a million ways it could go. Even if you limit it to a narrow topic of interest, there are so many different ways of expressing the same idea and getting to the same conclusion.
Managing this expansive tree of conversation is the challenge of one group of our users, the conversational designers. And once we have lots of data on how people like to speak to the bots, it's the bot optimizers' job to use that data to further enhance the bots' capabilities.
As a designer on this project, my goal was to help both groups of users do their jobs.
We've released the newest generation of the platform and successfully migrated some use cases over from the previous generation. Hats off to the team for adding such complex new features while not losing the reliability that such a tool demands. We're gathering lots of feedback from our users and continuing to develop features and tweaks that would help make their lives easier as they fully immerse themselves in the tool.