
Sam's Club
How might we design one forecasting tool that works for Demand Planners, the model that learns from them, and the engineers who build it?
One tool, three sets of needs
Demand Workbench helps Demand Planners predict customer demand so each club holds the right inventory at the right time. Their existing tools forecast only from historical trends and upcoming promotions. Planners made up the difference with their own expertise and a large number of manual adjustments, called overrides.
We were merging two legacy tools and adding a model that learns from planners' overrides, so one design had to work for three stakeholes:
Demand Planners who to do their daily work faster and to trust the tool with their decisions.
Data science who needed planners to enter overrides in a form the model could learn from, which the team called "feeding the beast."
Engineering who needed a scope it could build for an MVP, including a model that could not return a recalculated forecast right away.
This Product and Engineering team had not worked with UX designers before, and the project began as a list of requirements to check off. One requirement called for intensive configuration up front. Research into planners' daily work, followed by continuous feedback and iteration, reduced that initial load to the essentials and added repeat and bulk operations to make the rest more efficient.
At a glance
Company
Sam's Club
Industry
Supply Chain
Timeline
Mar to Sept 2023
Role
Associate Creative Director, working hands-on as a product designer, paired with a principal UX designer. I produced designs, facilitated design feedback sessions and synthesized the feedback to guide the next iteration.
Team
Two argo designers with Sam's Club business, product management, data science and engineering colleagues
What I did
Built research into a process everyone could see
We set out five phases: alignment, discovery, detail design, design support, and validate and improve. Planners had a named place in three of them: interviews in discovery, validation during design sprints, and verification of the built product. Putting user contact on the plan made it an expectation for Product and Engineering, not a favor.
Learned how planners work before designing.
Through interviews and walkthroughs of the legacy tools, we found three needs: managing changes in bulk, trusting that a decision had saved while the system updated, and seeing trends a table could not show. We also saw how much local knowledge goes into a forecast. For example, when a region hosts the Super Bowl, planners double down on grill items.
Outcomes
The UX team took what seemed like an overwhelming ask and turned it into an intuitive, user-friendly interface. Their ability to distill complex needs into a seamless experience has been invaluable in helping us ‘feed the beast’.





