The Home Depot
Taming Retail Chaos
Computer Vision Platform
Enterprise UX Manager
I led UX for a new computer vision platform designed to give Home Depot's digital systems a more current view of the physical store.
01
The Problem
Home Depot’s inventory systems could never be a perfect source of truth. They depended on people and transactions to keep the digital record aligned with what was actually happening in the store.
An item might be in inventory according to the system but missing from the selling shelf. It could be sitting in a customer’s cart, misplaced somewhere in the store, set aside for pickup, stolen, or simply somewhere other than where the system expected it to be.
There was another problem: where was the additional inventory? Extra product was stored in overhead bays throughout the store. Ideally it was directly above the selling location. In reality, it could be several bays or aisles away.
Associates often knew where things were because they worked in the store every day. The system didn’t.
02
The Reality of Retail
I shadowed and interviewed store associates to understand how they reconciled what the inventory system told them with what they actually encountered in the aisles.
The system had an oversimplified understanding of inventory: not purchased, the item should still be there; purchased, the item is no longer available.
The physical store was much messier. An associate walking down an aisle could immediately recognize an empty shelf. But unless someone updated the system, that knowledge largely stayed with the person who saw it. The location of additional inventory was even more dependent on tribal knowledge: “Hey John, do you know where we keep the extra 3-inch bolts?”
The physical store was the rapidly changing source of truth for inventory. The system was always trying to catch up.
03
Designing a System That “Sees”
We explored whether computer vision could close that gap.
The solution became the Bay Capture Cart, a mobile platform equipped with cameras that associates could push through store aisles to capture current shelf and overhead conditions.
Computer vision was trained to recognize empty shelf space, read product labels, and identify where additional inventory was physically stored.
That created a new feedback loop: see the store, identify the problem, locate inventory, generate work, restock the shelf.
When an item needed replenishment, Sidekick could tell an associate not only that work needed to be done, but where additional inventory was located and what it looked like.
04
Getting the Capture Right
Computer vision introduced an unexpected UX problem: the quality of what the system saw depended on how a person pushed the cart.
Aisle width, camera position and the way an associate moved through the aisle could affect image quality and, ultimately, the accuracy of the computer vision data.
Our goal was simple: a first-time user should be able to produce high-quality data.
I designed a kiosk-style touchscreen experience that guided associates through the capture process, reducing decisions and helping them move through the aisle in a way that produced reliable imagery.
Outcome
20% faster
Inventory restock response
The value of the Bay Capture Cart wasn’t computer vision itself. It was what happened next.
By giving Home Depot’s digital systems a more accurate understanding of shelf conditions and inventory locations, the platform could turn physical conditions into work associates could act on.
More broadly, the platform created a new way for the digital system to understand what was happening inside the store — instead of depending entirely on people to tell it when reality had changed.
