Better Recommendations
Nuts, Bolts, and a Better Shopping Experience
Improving Recommendations
Sometimes, shopping should be as seamless as Oreos and milk—or in this case, bolts finding their perfect nuts and washers. But in the vast world of e-commerce, our fasteners category wasn’t quite hitting the mark. It’s an area we identified as having strong growth potential, so I came armed with a vision to make compatibility and convenience the name of the game.


A Quick Word From Our Sponsors Me
I wanted to include this project because I’m incredibly proud of the outcomes. But it isn’t an overly exciting project, and I think that’s a good thing. It’s important as a Design Leader to find fast and easy improvements that have a big impact with very little lift from the team. And we achieved that with stellar results with this project. So don’t get too excited thinking you’ll see some revolutionary design changes. This isn’t it. This is about identifying something we were already doing—but poorly—and designing a solution to do it better.
After all, it’s just recommending products related to what you’re already looking at. But with a scientific and data-driven approach, we were able to have a major impact on sales and profits.
The Final Results
Results
Start at the End
The new feature rolled out as a pilot across 400,000 carefully selected SKUs. We wanted to get the results as quickly as possible, so we compiled a list of products with solid product data that also had frequent sales every month. This would allow us to get comparable results more quickly than if we had rolled it out across the entire product line. We gathered the proper data to benchmark against and launched our test for compatibility.
The results? Beyond expectations:
Notable Outcomes for the Business
36% increase in AOV (average order value)
82% overall increase in Average Items Per Order
44% increase in profits per order placed
We had high expectations, but even we were impressed with these results. The majority of our fastener orders tended to include only 1 or 2 items. But after our Fits With feature launch, the number of items per order nearly doubled. And almost every order included a second product from our recommendations list. We knew we had a winner with this strategy.


Introduction
Project Background
How We Got Here
So let's talk about how we got here. Picture this: you’re shopping for a 3/4"-20 bolt, and our system suggests an unrelated #8 deck screw or a random 1/2"-32 nut. Helpful? Not exactly (if you know anything about fasteners).
Customers were left piecing together their orders manually, running multiple searches for nuts and washers that should have been recommended alongside their bolts in the first place. It was inefficient, frustrating, and a missed opportunity for better sales. I identified the issue while watching some customers interact on our site, raised it to the organization, and recommended a hypothesis for how we might fix it and increase sales at the same time.
The Plan: Right Product, Right Time
Design can solve a lot of problems, but it can't fix everything alone. To get data from our internal systems out into the front-end of the website requires multiple teams and a lot of collaboration. Luckily, that's where I thrive. *Points at myself with both thumbs*
We huddled with the merchandising team to craft a system of compatibility criteria ranked by importance: Must match, should match, and can match. By analyzing this compatibility data, we revamped our recommendation engine to suggest only items that truly matched the product being viewed—nuts and washers compatible in size, grade, and material at a minimum. We repositioned the recommendation carousel closer to the “Add to Cart” button, ensuring it caught the user’s eye when they were in the midst of their purchasing decision.

Getting to MVP
Final Product
Getting On With It
Like all good projects, it starts with a solid plan and clearly defined problems.
Researching the Pain Points: Customer told us loud and clear—irrelevant recommendations were a headache, and many customers would simply abandoned the site due to frustration. In a focus group study, we saw users wrongly assume we didn’t have the product they needed, validating a strong customer need.
Prototyping Solutions: We explored several designs, but ultimately focused on quick, impactful fixes. What components do we already have? What design elements can we reuse to minimize new tech debt to maintain?
User Testing: Moderated and unmoderated tests confirmed our hypothesis with real Zoro customers. In 100% of the tests, users were able to find compatible products almost immediately, and unanimously preferred the new approach.
Conclusion
The Road Ahead
The Conclusion
Making high quality recommendations is more science than art. But thoughtful cross-team collaboration with Data Science and Merchandising Teams allowed us to create actionable rules for what we display with each product. The rest was me using existing site components and structures to craft an intuitive experience that made sense to the customer that we could build quickly and efficiently.
We launched it as a proof-of-concept, but the results got us buy-in from leadership to expand this effort out across a greater number of categories. What began as an effort to sell more fasteners transformed into a robust system that’s improving how our users shop, in every category, every day.