Bravura Solutions · Principal UX Designer
2020-2021
Stanza A.I.
Bravura Solutions' Head of Innovation asked me and the company's data scientist to find a new product in the overlap between our two disciplines. The result was Stanza, an algorithm that reads unstructured communications - incoming emails, PDF dealing forms - and extracts their intent.
The algorithm only worked if humans could train it fast. My job was the interface that let teams verify and correct thousands of machine classifications at speed, so that Stanza's accuracy climbed to the point where human oversight was no longer needed. I ran the workshop that framed the problem, built the personas, wireframed, guerrilla-tested and took it through to developer handover. I also named the product.
Stanza was built into a working prototype, taken to market in fintech, adopted by its first client and demonstrated to the Bravura board.
The brief
My data scientist colleague had been working on an algorithm that he believed could work out the intent of incoming communications, such as emails, with sufficient training by a human.
Stanza would vastly reduce the need for laborious and time-consuming manual processes to manage incoming communications for a large business, giving huge cost savings and accelerating data processing.
The main problem was getting Stanza’s accuracy up to a place where it could be trusted to run without any human oversight. Stanza needed to learn how to do its job as quickly as possible.
My role
I worked closely with not only our resident data scientist, but Bravura’s Head of Innovations, developers and Product Managers at various stages of the project.
Skills used:
Workshop
Ideation
Wireframing
Prototyping
Guerrilla user testing
Visual design
Handover to development.
The problem
My main task was to create an interface that could be used by teams of individuals to train our algorithm. Auto generated results would be verified and corrected by humans, which would lead to an increasingly accurate result from the application, until the human training was no longer required.
To achieve this level of confidence, a large number of results would need to be checked at speed, so an efficient user-centred interface for the people checking its results was essential to the application’s success.
However, my very first task was to give this algorithm a name. Bravura’s main software product was called ‘Sonata’, a classical music term, so I decided to follow the theme and called this new product Stanza, meaning ‘a related group of lines in a poem or song; a verse’, which also linked to its goal of finding meaning in words.
Insight and research
Workshop
I ran a workshop with my data scientist colleague and a senior Business Consultant, to understand what the problems were that needed solving and start the process of white-boarding potential solutions.
It was the first time the data scientist had ever worked with a UX designer, so I started by talking through my approach to UX problem-solving, we wrote down words to describe the desired outcomes and followed that up with a description of what Stanza’s training interface should achieve.
The key takeaways were:
Enable the human user to work fast and efficiently, volume helps Stanza become more accurate.
Make the interface intuitive and easy to use
Ensure all the key pieces of information are collected and corrected.
User flow
The user flow was initially established during the workshop and further developed as more information and feedback was gathered.
Stakeholder interviews
Stakeholder interviews were conducted with the Head of Innovations and developers to understand business goals, key pieces of information needed, and technical limitations.
User personas
I created two user personas based on the information gathered:
Our primary user is a junior member of a large financial institution, happy to do laborious repetitive work
Our secondary user was the manager of the team. This user would probably not be going through any emails or PDFs themselves, but assigning work to the team and keeping an eye on how well Stanza’s algorithm is performing.
Early iterations
After the insight and research part of the process, I proceeded to these early low fidelity wireframes. We guerrilla tested these within Bravura and swiftly moved on to further iterations.
Outcome
Senior management at Bravura responded well to Stanza, viewing it as an important new micro-service for future revenue generation. The initial design concepts were built into a fully functional prototype and the product was taken to market in fintech. It was adopted by its first client, and demonstrated to the Bravura board.
Folks,
A massive thank you!
This morning we gave the Bravura board an update on our product and technology strategy. This was supposed to be a face to face session here in London, but given the current situation we did the meeting over Skype.
One of the things we did in the session was a Stanza demo, and the board were very impressed, which was great.
I love the way it pulls together our expertise in UX, web design and build, and of course AI into a product that really shows what great stuff we can do, as well as underpinning how much benefit we can get by adopting a modular microservices strategy.
So a big thank you to all involved in bringing Stanza to life.
Kind regards,
Nick Parsons
Chief Operating Officer (Global)
