IBM Watson Machine Learning Accelerator
I worked as Design Research Lead with a Product Design Lead. We were the first design team assigned to the product, which had been released four months prior.
IBM created Watson Machine Learning Accelerator (WMLA) as a comprehensive enterprise AI platform that offered dependency, security, and rapid deployment without precedent. Its key value lied in its integration of accelerated IBM Power Systems servers with popular open source deep learning frameworks and efficient AI development tools.
WMLA contained extensive command line interface support and rudimentary graphical user interface. However, WMLA was at the bleeding edge of AI innovation. Less than 5% of targeted clients had the infrastructure, human resources, or expertise to utilize the platform and its potential use cases remained unexplored. The design team was brought into this project to research client segmentation, implement initial feedback, and transform the product into a marketable, useable, user-centered experience with industry-wide adoption.
Contextual Research Process
The first step to triaging WMLA was to understand the ins and outs of deep learning frameworks and users. For several months, I worked with one other designer to interview data scientists, take courses on Python and Tensorflow, read competitor documentation, develop personas, and create prototype AI models. I synthesized most research on Mural boards to allow for collaboration with interviewees and presented findings to the design team on a rolling basis via decks and live POCs.
WMLA’s wider team of managers, developers, architects, and executives had never worked with design and required detailed reports and frequent meetings. It was both a challenge and a valuable experience to pioneer the field of Systems AI Design at IBM. At this stage, I performed mostly as a product manager: introducing the value of design, integrating client-interfacing teams with IBM support, and determining the product roadmap.
User Research Process
With less than two dozen accounts, WMLA had few users to interview. In addition, these clients often experienced months of configuration and installation complications before implementing a single use case. When I initiated the sponsor user feedback program, not a single client had deployed a trained model on WMLA.
I established rapport with frustrated clients by working as de-facto product manager/support, offering insight on their problems and connecting them with the appropriate tech sales representative. This phase of research allowed me to learn about client’s workflows, use cases, teams, and pain points. Sessions with clients were noted, recorded, and parsed to gather a unified body of data with quantifiable insights that could be shared and workshopped with the wider team.
One significant artifact from this phase was a comprehensive, end-to-end user journey map. Communication blockers among the product team prevented them from seeing the full breadth of the client experience, which I combatted by compiling detailed documentation explaining approximately 5 client journeys and hosting a workshop to combine the insights into one simplified map (which was later validated by 10 additional clients). An individual client's detailed journey map is shown below.
Insights to the complete user experience had never been shared among the product’s siloed teams.
This user journey map became the key defining artifact for the project's roadmap and led the team to adopt a multidisciplinary, agile approach.
Ongoing Research Process
After I had established a cadence to the product’s releases and a methodology for prioritizing feature improvements, we were able to make targeted design goals. I modified my interview format to show users “exclusive first looks” at lo-fi information architecture diagrams and potential marketing plays. I identified the product’s common use cases, compiled client profiles, and created sales playbooks and startup guides.
At this stage, an additional UX designer was brought on to begin work on the high-priority design needs I had identified and begun to prototype according to user research. I was moved onto another AI project. The handoff contained approximately a year’s worth of planned feature improvements and sizing for design only, as well as a blue-sky development plan for the product’s ideal trajectory.
A comparison of my client survey upon beginning work on WMLA and a year later showed a 40% increase in NPS rating. The sales team also followed an research-based plan I had developed to create a bundled “starter kit,” which I had designed to remove configuration-based pain points and reduce cost, increasing sales and industry adoption exponentially.