Stitch Fix
Online shopping is great if you know exactly what you’re looking for, but it’s never been very good at helping consumers discover the unexpected. E-commerce democratized shopping, allowing you to do it from the comfort of your own home. But it’s inconvenient in so many ways: You’re forced to search and scroll for products, and the vast majority of what you’re seeing is not relevant to you.
We wanted to create a shopping experience based on discovery.
My Contributions
I was hired by Stitch Fix spring of 2021, soon after they had launched their Freestyle platform. This was a significant departure from their subscription model, and they were looking for a designer to lead the “front door” experience that could house and support both funnels simultaneously. The home feed fell under this umbrella, which served as the launch pad for both Fix and Freestyle features, whilst highlighting opportunities for further personalization.
During my tenureship, I’ve lead several feature projects focused on early funnel optimization, geared toward improving “stickiness” (our highest trafficked page was 270k DAU, 1.6 million MAU) with new and existing users, which is what we’ll be touching on in this case study.
project background
Stitch Fix is expanding its offering to serve clients in more ways, and more often. With the launch of Freestyle and the acceleration of our engagement strategy, clients are now presented with more content and new features than ever before. Our logged-in homepage (including the “homefeed”) plays a critical role in orienting clients and helping them get the most out of Stitch Fix. However, the current experience is not meeting clients’ needs for intuitive navigation and engaging style discovery
Less than 1/3 of visitors to homefeed click on any content presented (28% in Jan 2022)
Clients consistently submit Hotjar feedback regarding navigational confusion or frustration on homefeed
Research conducted during Onboarding 2.0 revealed that the homefeed experience is particularly confusing for new clients, who weren’t clear on the context or purpose of the content presented, or expectations on what to do next.
PURPOSE OF THE WORK
Make the first Fix experience feel fun, engaging, and effortless. Help clients understand the most important actions they need to take at the optimal times. Enable better communication between clients and SF so they’ll want to come back regularly.
Success metrics
Our North Star was Total Active Clients. How we broke that down into digestible metrics for our push was to focus on what we deemed High Value Actions. Those included product page views, Style Shuffle plays, ATC, shares, etc.
We identified and fleshed out three priorities. This case study will focus on the Shoppable Content Modules.
CHALLENGE 1
How might we help clients find inspiration, discover what’s new, and take action on the things they love?
CHALLENGE 2
How can we feed the algorithm a LOT of data so we can deliver high quality recommendations?
Preliminary Projects
Shoppable Conent Modules
We’d first explored connecting our shop to contextual content we already had or could get relatively easily.
These quick projects showed a 17-35% engagement rate (up from 2-3%).
IG Social Inspiration
An issue we observed were clients had a hard time putting outfits together. Our outfit collage module did well, but still lacked context (and fit).
We imported Instagram photos from our list of creative partners & built an image recognition software that served recommended items based on the photos, prioritizing exact matches followed by similar.
This went through a couple iterations, primarily focusing on cues for shoppability, constraining the actions to most important, and improving the image recognition results.
At launch, we saw a 30% engagment rate (up from 2%).
“Get the Look” Module
In addition to the social inspiration module, we also explored an opportunity to bring our blog articles into the discovery experience. In my search for easy-to-get content, I’d discovered a well of stylist curated media that wasn’t yet tapped by the shop space.
Using the same image recognition software we created for social inspiration, we connected it to the editorial imagery mined from the blog posts, and crafted another “shop what you see” scenario.
The added bonus with this type of module was the taxonomy - being able to put the “look” into words like “normcore”, “dark academia”, “cottage core” etc., helped our clients visually connect the dots with trends they’d previously heard of.
Measured against other outfit modules for clients interested in these particular style clusters, we saw an increase in engagement from 2-3% to 25%.
Meanwhile…
People were canceling their Fixes and quitting Stitch Fix because they didn’t feel like we’re really “getting them”.
They were dying to tell us more about themselves and needed to feel like we were listening.
STYLE SHUFFLE 2.0
Objective: Help clients find inspiration, discover what’s new, and take action on things they love.
Style Shuffle is a feature that enables clients to tell us what they like and don’t like in an easy-to-use “Tinder-esque” fashion. We originally designed it to improve the onboarding experience, which proved extremely successful. We then added it to the homefeed to see how often they would interact with it, which was much more often than any of the other modules combined.
However, that was before we launched our direct shopping capabilities. It became a primary opportunity for us to share product interest to our brand partners, which was a huge selling point, but didn’t yet extend its potential to our end users.
Around this time, I began googling “I wish Stitch Fix would…” and stumbled upon a lot of Reddit threads like this one, and this one, where clients were routinely asking void to help them identify clothing items they saw in our imagery. There was a clear opportunity for us to connect the dots.
Current Style Shuffle
Style Shuffle ratings are valuable to Stitch Fix, and many clients enjoy the current experience. We didn’t want to harm this while making improvements.
Not every item in Style Shuffle is purchasable, and when it is, it may not necessarily be available and in stock for every client. We needed to make this consistent despite this.
Style Shuffle ratings were not connected to the rest of the platform in any way, making any new features that add ‘shop’ capabilities non-trivial to build. We would essentially need to consider creating a ratings profile that would evolve with the user from the ground up.
Baseline Study
I performed a quick measure of users' attitudinal perceptions of the current Style Shuffle experience while simultaneously gathering some potential usability issues to correct.
Testing “Shoppability” Interactions
Our ultimate goal was to make it easy for users to shop items that they liked directly from the Style Shuffle module. However, I was concerned about losing valuable ratings data by 1) making it interruptive, and 2) introducing complexity.
My biggest question was then “should we allow clients to shop directly from the image card or have them wait till the end to see a round-up of their likes?”
Shop the Card
Allow the user to shop directly from the rating card. For users who might want to shop their likes *immediately*.
Questions to answer:
- Would users mind pausing their rating experience to go shop something they really like?
- What if an item isn’t buyable? How can we still -give them value? Will it confuse/frustrate them if most aren’t directly shoppable?
- Will users even see/interact with the CTA?
- Is this an easier lift? Would it get the job done with as little resources as possible?
Likes “Round-up”
A shoppable “results page” after 30 ratings. Allow them to have a seamless rating experience, and still take action on the things that they love.
Questions to answer:
- How can we get users to rate enough to show them a valuable collection of recommendations?
- How can we communicate to the user what the “candy” is so they know what to expect?
- What is the most user-friendly layout for a compelling results page? How would they want to navigate it?
Results
The “Round-Up” designs were a clear winner for several reasons.
ROUND-UP WINS (though many would like to see both together)
Users prefer NOT to be interrupted to shop while rating
When they “LOVED” an item their quick reaction was to hit “thumbs up”, not “shop this look”.
6/10 didn’t notice the shop CTA till later (half not at all). When users used it (3/10), it was more to keep vetting “maybes” or to see similar items…not bc they wanted to buy it immediately.
Nobody tried to click into the product page nor acknowledged the “bookmark” option from the shop quick view on card-level.
Question 1:
Which “Round-Up” Layout is Most effective?
What is the most user-friendly layout for a compelling results page? How would they want to navigate it?
I tested two layouts: one more aligned with a basic search results page and another that incorporated similar content modules to our discovery feed.
Search Results Page
“Magazine” Layout
Results
SEARCH RESULTS PAGE WINS
Users didn’t like having to side scroll or tap to see more, they like seeing everything laid out for easy vertical scroll.
Magazine layout felt like “just another homepage” rather than immediate results from their ratings.
All users loved the CYL outfit module. Some wished they could go deeper there, perhaps see multiple “yes ratings” in one outfit collection.
Many requested category level filter/sort
Question 2:
How can we get them to continue rating multiple packs to ensure optimal data collection?
Card View
List View
Results
WE HAVE MORE TO LEARN
Many users reported liking to see all of their options on a single page over having to side scroll to discover
A “checklist” feels better for completion than editorial cards.
Progress toward aggregated results appears to be more easily grocked with a list/thumb view over card carousel.
People read the micro-copy.
Final Thoughts
We launched the new Style Shuffle design across a smaller cohort of users, which revealed an uptick in saves, from-module ATC, and # of items rated significantly, before rolling it out to 100% a couple weeks later.
However, I still had a lot of questions leftover for future optimizations. I wanted to continue to learn about how we can optimize the rating flow to protect/grow rating # and discover best layout + navigation for the Style Shuffle results page.
I bucketed these questions above and put them in the parking lot for our team to consider.
Ultimately, we had discovered that our clients were very open to taking the time to tell us their preferences, particularly when there was a bit of feedback given to them directly after.
This enabled us to draft our future OKRs around exploring quizzes, which we are still working on today.