Inventory exception management.
Using data science and mobile experiences to create enjoyable and meaningful work.
Problem overview
Keeping up with Inventory
In retail, products anywhere but the shelf don’t lead to sales. Inventory tracking and ordering systems are limited by problems such as spoiled food or items being misplaced. As a result, the quantity of products does not always match reality. An exception in inventory counts has occurred.
Managing Exceptions
Historically, retail workers manually adjust product counts by counting shelf sections throughout all 180,000 sqft of a store. This process is tedious, time consuming, and frankly, no one likes doing it.
How Might We Increase Productivity?
I worked with a team of data scientists to rapidly prototype a new process that relied on proprietary data models to tell retail workers exactly where inventory counts may not be correct. By mixing human interaction with data analytics, we were able to increase productivity by 42 basis points over an 8 week period in test markets.
role
Design Lead
team
1 Data Scientist
2 Product Managers
2 Engineers
1 Business Operations
As Design Lead, I conducted in-store discovery, concept creation, validation and usability research, and product design while collaborating with a small innovation team reporting directly to the VP of Store Operations.
Discovering pain points
I knew legacy processes of inventory tracking were tedious, but to understand the full extent I did it myself. To echo the sentiments of many users I spoke to: it was awful. I was led through multiple steps of a process I felt was built for machines, not humans.
Completing inventory counts took me 10 minutes. 4 minutes longer than average.
There was no sense of priority.
Cascading inventory issues forced me to give up.
Making it better with paper prototyping
Those pain points led to a deep understanding of retail worker needs as we worked with data science to create a new experience by paper prototyping in stores.
Rather than mindlessly scanning products, we were guided by meaningfully targeted exceptions.
Knowing where to look for issues decreased time in section and increased feelings of accomplishment.
We empowered ourselves to flag cascading issues for follow up, thus transforming a feeling of defeat into a productive action.
Concept creation
I created multiple approaches to a new mobile application by focusing on retail worker needs. The new experience was crafted to build trust in new data models, communicate clear instructions, and provide streamlined information for solving inventory exceptions.
Workers use a variety of methods to locate products depending on multiple factors such as tenure, mental model, and product familiarity. Adding images, information hierarchy to long numerical strings, and e-commerce product descriptions improved visual scanning of shelves. Workers found products faster and more precisely.
Initial sketches focused on bold product imagery, descriptive text, and large inputs to guide users to a single inventory exception. The goal was to increase focus and motivation in a chaotic store environment.
Higher fidelity sketches were tested in stores to refine the experience.
To verify products were correct, a combination of AR scanning features helped workers hone in on problem areas.
Various high impact stock illustrations were considered along with gamified content strategies.
Renderings of shelf images were explored to guide users. These were wildly popular, but ultimately not technically feasible.
After multiple rounds of testing, a process emerged
We realized through testing workers were focusing on shelf tags and not product images. We chose to render these tags in the app as they appear in the real world and saw improvements in time to locate exceptions. Focusing users on one product exception at a time also saw increased productivity over giving lists of multiple items per section.
A functional MVP
Users select a department to work and can favorite departments they are assigned.
Users are guided through a consistent stepped process for exception location mirroring what they see in the real world.
An augmented product scanner shows expected values and location coordinates while giving additional options in case labels are missing.
If an incorrect product is located, the data model is automatically updated while workers are directed to solve problems with additional descriptive and visual aids.
Finally, gratifying messaging leads workers into their next task.
Impact
1,920 exceptions resolved per week.
By mixing human interaction with data analytics, we were able to increase productivity by 42 basis points over an 8 week period in multiple test stores.
Customers continue to see an increase in “first time pick” (personal shoppers finding their requested item the first time) which directly relates to speed of pick up order fulfillment.
On average, time spent per shelf section dropped from 2.5 minutes to 30 seconds.
And in the words of an anonymous Reddit user: “I like [the app] a lot, so much faster than section work.”