Designing a settings console that control machine learning algorithm
Premise
Software: Energy asset bidding software driven by a machine learning algorithm

Users: Energy asset managers, energy traders (think stock markets)
User Behavior: Traders weren’t using the price bids generated by the software
Problems:
Customer (who was beta-testing) stated the lack of control over the algorithm as a core issue in the RFP
One of this team’s OKRs was entry into the California market and this was the first client
Objective
"We built them an autonomous aircraft but without controls for the pilot"
My task: Design a trade settings page that allows users to schedule setting changes, and adjust the bidding algorithm according to risks, resources, etc.
Process
Existing Resources?
Existing page of a “sister” product created by another Designer Design impact: starting with hand sketches vs high fidelity prototypes

What are the settings?
Sessions with Data Scientists and internal trading experts to understand how settings influence the algorithm Design impact: virtual knobs vs entry fields vs other methods

Technical Constraints
Sessions with Developers to understand when and how user submissions impact forecasting Design impact: creating a time/date picker that clearly indicates what ”asap” vs “schedule”

Design Review
Presentation and walkthroughs with Product Manager, Backend, Frontend, Data Science to kick start development

Results & Next Steps
Outcome
Traders had enough control to use the generated bids
Next Steps
Partner with Platform Analytics to further understand users and iterate on design
A/B testing of placement of components and settings knobs
Consider scalability issues such as: archive log of setting changes, adding filters to types of settings, consider grouping settings in tabs