The problem

Furniture is a high-consideration purchase, people need to understand financing, availability, delivery, and protection plans before they'll commit. At Dufresne, those decisions were scattered across the journey, so customers hit a wall of uncertainty right when they were closest to buying.

Two things kept surfacing in complaints and support logs: customer support was getting pulled in constantly to explain warranty and protection-plan options that the interface should have answered on its own, and shipping and financing options were confusing enough that cart abandonment was flagged as a real problem long before I got involved.

Business goals we were designing against

• Increase service visibility, make warranty and financing options easy to discover, not something guests had to hunt for or call in about.
• Increase paying customers, enrich the journey enough to convert more browsing guests into buyers.
• Reduce cart click-throughs, cut the number of clicks a guest needed to select shipping, warranties, and payment options.

How I got there

Before proposing a solution, we validated the problem in three phases: a competitive analysis benchmarking Wayfair, Home Depot, and Best Buy on how they surface services at the point of purchase; a voice-of-guest survey of 50 customers to test our assumptions against real behavior; and an ideation workshop that pulled in fulfilment, tech, and guest care, not just design, to make sure whatever we proposed was actually buildable.

I mapped the full path from product discovery through post-purchase service to find exactly where questions turned into blockers, then sorted the decisions into what was essential in the moment and what could wait.

Customers weren't asking for more information. They needed the right information, in plain language, exactly when the decision got consequential.

Key decisions

The solution

The redesign surfaced a protection plan directly on the product page with a clear result of the research behind it, and let guests add or remove a protection plan right from the cart, no phone call required.

What changed

• Online sales increased 15%, with cart abandonment dropping to 35% within two months
• 46 of 50 guests said the added service and financing info increased their confidence in deciding
• 38 of 50 guests said the mini-cart protection plan prompt encouraged them to learn more about it
• A noticeable drop in customer support calls asking about protection plans mid-purchase

These numbers came from the team's own reporting rather than a single dashboard I owned, so I'd want to re-verify the exact measurement window before quoting it precisely in an interview, but directionally, this is one of the clearer wins across my case studies, and it's part of why the reusable panel pattern became the foundation for later commerce work.

What I'd carry forward

The strongest move here wasn't adding a feature, it was sequencing information that already existed around how customers actually make the decision, and making operational complexity feel calm instead of overwhelming, even while working inside a fairly rigid e-commerce platform.

The problem

Campaign images did their job, they created desire. But guests couldn't easily tell what the model was actually wearing, so the handoff from "I want that" to "I found that" was broken. One guest, quoted directly in our research: "I want to replicate this look but I don't know how to buy the outfit the model is wearing in this banner. Help me find the products so I can bring my inspiration to reality." People were leaving the campaign entirely to go search for something they'd already decided they liked.

How I got there

I partnered with research on three tracks that ran in parallel rather than sequentially, so we had both breadth and depth before committing to an interaction model:

One insight reshaped the whole direction: we'd designed two competing entry points into the same interaction, a dot hotspot and a separate garment icon. Guests recognized and clicked the dots immediately. The icon barely got used, and having both active at once didn't add a second path so much as introduce doubt about which one was "correct."

Product hotspots were understood immediately. A second, competing icon suggested a second path and just added doubt, one strong invitation beat two weaker ones.

Key decisions

Designing the tooltip as a real component, not a one-off

Because this had never existed in the pattern library before, I worked directly with the accessibility council and the design systems team to define font weight, sizing, and how price and sale-price information stacked, including how it should behave when a product went on sale mid-campaign.

Building it to WCAG, not just past it

We mapped a full keyboard focus order for the carousel before development started: focus stays on the arrow buttons rather than jumping into the carousel automatically, and using the pagination dots moves focus to the first item on that slide. Small decisions, but the kind that are invisible if you get them right and disqualifying if you don't.

What shipped

A three-part system, a progressive-disclosure CTA, accessible hotspots, and responsive tooltips, that kept the campaign image visually dominant while making every featured product discoverable and purchasable.

What changed

• 33.6% increase in campaign-image engagement on homepage
• 2 in 5 engaged guests purchased one or two featured items
• $4.4M web + $2M app in incremental revenue

These are the numbers the team reported after launch, I don't have the exact measurement window in front of me anymore, but the direction held consistently across reporting periods.

What I'd carry forward

Accessibility has to sit inside the interaction model, not get bolted on at the end. Bringing in assistive-technology users early led to a simpler experience for everyone, not just a compliant one, and building a new pattern from scratch meant getting buy-in from the design systems team was as much a part of the job as the interaction design itself.

The problem

The existing outfit module was visible but barely used, 2.35% engagement on web, 1.6% in app. Products were showing up as a row of AI-generated recommendations, but the interface did nothing to help a guest picture a coherent, head-to-toe look. Guests saw the outfit, but didn't engage further, either because they weren't inspired by the pairing or because they had a hard time visualizing it as an actual outfit rather than a list.

What guests actually needed

Research surfaced three distinct needs, each backed by direct guest quotes:

• Versatile outfits that solve more than one problem, "I need a variety of product suggestions that will help me pair what I already have in different ways so I can build a more stylish and versatile wardrobe for my lifestyle."
• Style inspiration and education, "I need idea for putting outfits for different events and activities together so that I can buy the right looks and educate myself on style and how to accessorize."
• Confidence for the occasion, "I need to put together an outfit appropriate for the season, so that I can be comfortable and ready for my activity."

How I got there

I led a brainstorming workshop to generate concepts, then a sorting exercise to surface the strongest direction. The ideas grouped into five distinct experience types before we converged on a grid-based layout, chosen because it prioritized efficiency, clarity, consistency, and scalability over the alternatives.

Guests wanted an outfit idea, not a row of related products. The relationship between the seed item and every recommendation had to be obvious at a glance.

Details that mattered more than they look like they should

• A fixed seed item. Keeping the seed item static while guests swiped through outfit options reduced repetition and kept orientation clear.
• A "current item" tag. Guests genuinely got confused about which piece was the main one they were already looking at, a small label fixed it.
• Truncated product names, no price at this level. Given price and size are one click away, we prioritized giving guests inspiration first, not a full spec sheet, to reduce clutter.

Designing for a model that doesn't always get it right

This wasn't styling done by a person, the flat-lay outfit images were generated by pulling real product photography together based on what the machine learning model paired with the seed item. That changes the design problem. A human stylist can be trusted by default; a model has to earn it, and the interface has to hold up on the days the model's pairing is a little off.

• Designing for inconsistency, not the ideal case. Most component design assumes reasonably predictable content. Here, the layout rules had to survive an odd color pairing, a mismatched formality level, or a recommendation that just didn't quite work, without ever looking broken.
• No explanation, by design. We deliberately didn't add a "why we picked this" explanation layer, at this stage it would have drawn more attention to the model's reasoning than the guest cared about. The bet was that a strong visual result would build more trust than a caption ever could, and the engagement numbers suggest that held up.
• Honest failure states mattered as much as the ideal composition. If the model couldn't generate a confident pairing, the component needed to degrade gracefully, fall back to a simpler recommendation, rather than force a bad output onto the guest.

I don't have visibility into the model itself, that sat with the ML team, not design, so my role was entirely about the guest-facing consequences of imperfect AI output: what the interface owes the guest when the underlying system is probabilistic rather than deterministic.

Key decisions

What shipped

A structured flat-lay grid that establishes the seed item first, then builds out a complete look, with an accessible carousel and modular content slots so future recommendation strategies can plug in without a redesign.

What changed

• Engagement moved from 2.35% (web) / 1.6% (app) to 5%
• More multi-item purchases started directly in the component

I don't have the exact evaluation window handy, I'd want to confirm the dates before quoting this in an interview, but the lift was consistent across the reporting I saw.

What I'd carry forward

Designing around machine-generated content means designing for inconsistency. Clear layout rules, resilient states, and honest empty states mattered just as much as the ideal-case composition, and future-proofing a component is a design decision, not just an engineering one.

The problem

Guests didn't see lululemon as a destination for head-to-toe looks, they bought single items from the franchise or category they already loved, then went elsewhere for outfit inspiration and styling advice. Single-product journeys made it hard for guests to build a complete look, and just as easy for different teams to ship disconnected outfitting features that didn't add up to a coherent experience.

Dealing with ambiguity

Strategy work with this much ambiguity can go off the rails fast, so before anything else I partnered with our Product Manager to set milestones, define roles, and put a system in place to keep senior leadership looped in on progress every two weeks. We broke the guest base into four archetypes to shape which research questions we'd prioritize:

Outfitting isn't one component. It's a connected behavior that starts with inspiration and needs continuity all the way through purchase.

Generating and testing ideas

I led a brainstorming workshop that produced five distinct concept directions, Collage View, Grid View, Vary Combinations, Mannequin, and Shop the Image, then ran a sorting exercise with the team to weigh each against guest impact and feasibility before taking a shortlist to senior leadership for a vote on which opportunity areas to prioritize.

Key decisions

What it enabled

• Aligned product, merchandising, research, and engineering around one shared model
• Directly became the foundation for Shop the Look and Outfit Inspiration

Both of those projects have their own metrics, documented on their own pages, I'm not attributing their numbers to this one, since this was the framework, not the feature.

What I'd carry forward

A useful strategy is specific enough to guide real decisions and flexible enough to survive a changing roadmap. Tying the principles to concrete components, not just a slide of values, is what made this one actionable instead of aspirational. Influencing stakeholders to break from the norm took real persuading, and I learned that this kind of ambiguous, org-wide work is never a one-and-done process.