Due to NDA I am unable to share any images or documents during my time at Amazon.
The team that I joined at Amazon was called Central Flow, it was a brand new operations team that aimed to centralize knowledge of flow and apply best practices across all fulfillment centers. Each analyst was given a fulfillment center and would be responsible for ensuring that order flow was as smooth as possible throughout the entire shift. This required analysts to track and monitor system drop rates and worker rates across all process flows, and mitigating risk both internally and externally.
Part of our responsibility was to make staffing recommendations to the facility leads, where people were needed, and when it came down to half-time how many people to pull in or let go.
A majority of this work cumulated into two spreadsheets and one monitoring page.
After getting a sense of the data, I started to notice the estimated staffing spreadsheets weren’t matching rate proportions properly and would consistently end up with higher staff counts than necessary, even above thresholds. So I decided to do some digging on the underlying math on the spreadsheet to determine how the spreadsheet was calculating its recommendations and why.
My research found that the formulas used static assumptions on pick rate, employee loss, and rate deterioration.
With a few tweaks, I modified the spreadsheet to have analysts plug in their rates, respective to their own facilities, and it would dynamically adjust. This way an analyst would have more control over their estimates and also be able to better project and run simulations for later shifts.
This spreadsheet was sent to my leaders which was then updated to all analysts.