Guest posts are always a pleasure, but today I’m particularly delight to have AJ Shankar of Modista share his thoughts about visual exploration.
My friend Arlo and I co-founded Modista, a web site that uses computer vision algorithms and a novel user interface to enable product exploration and discovery on the web. The premise is simple: click on an item (a shoe, watch, handbag, …) and Modista will automatically show you items that look similar to it. Repeat the process and soon you’ll find yourself exploring hundreds of items in an intuitive fashion — one of which you’ll want to buy (hopefully).
Rather than focusing on Modista’s two main components, its computer vision algorithms and its structured grid display, I’d like to discuss some of the less obvious, but still important, goals we have for the site.
Goal: Take exploration cues from real life
One of our goals with Modista is to bring a bit of the real-life mall shopping experience to the web. When you go to a mall to buy a pair of shoes, you’re more likely to come out with one than if you visited an online retailer, even though the online shop has a much, much larger selection, often at lower prices to boot. Why?
We think a major component of the answer is the ease of exploration. It’s trivial to walk around a shoe store and notice something that strikes your fancy.
(Some other components — like the ability to try on a shoe for size or feel its texture — are tricky, and we leave it up to our retailer partners to approximate solutions to them, such as fast, free shipping, and a liberal return policy. Others, such as the desire to justify the time investment of going to the mall with a purchase, are even further out there.)
Obviously our grid-based visual similarity interface is meant to encourage the exploration process. But we also take a critical look at smaller interface issues: What’s disruptive? What distracts from the experience? For instance, we really dislike scrolling and page load wait times: when was the last time you had to wait for a rack of shoes to “load” at your local mall? So when we wrote the server for Modista, we made sure it could operate with extremely low latency, and when we designed the website, we used AJAX for the page transitions, and made sure that we displayed just as many items as would fit on the browser screen without any scrolling. Try resizing your browser to see how it works.
At the same time, we don’t shy away from the strengths of the web. You can still see all 160,000+ women’s shoes we catalog, and our algorithms sift through every one of them to find you the best matches. The mall has nothing on the web when it comes to inventory.
Goal: Encourage positive actions
Here’s a typical scenario at an online retailer’s website: a customer types in some search terms, and is faced with 20 pages of results, with 50 or more items per page. The user scrolls through the first page, finds nothing she wants to buy, and clicks “Next”. This process repeats several more times before the user gives up.
What has the retailer learned from this interaction?
Each “Next” click by the user is a negative action: the customer is saying “I don’t like any of these 50 items”. So after ten clicks, the retailer knows that the customer doesn’t like 500 of the 50,000 items it carries. If the retailer’s eventual goal is to find something that the customer does like, it’s a frustrating place to be.
The Modista interface encourages positive actions. The default behavior in a Modista browsing session is a click on an item. The click says, “Even if this isn’t the perfect item for me, I like it more than everything else I’m seeing here.” After ten such clicks, we have a much better understanding of what the user is looking for than if she just clicked “Next” ten times. Better yet, the user feels like she is making progress: the click is taking her somewhere better. We like that.
Goal: Be memorable
Modista is a portal site that aggregates inventories from lots of retailers. In that sense, we’re in the same space as comparison shopping engines (CSEs).
There’s a well-known problem with CSEs: they’re not memorable. Here’s a sample consumer thought process: “I need to buy a DVD player. Go to Google, type in some keywords… click on a link to [your CSE of choice]. Page loads, click on an offer at Amazon, buy the player. Great!”
Sounds good so far. But then the next time the consumer wants to buy an item, she thinks, “Well, what did I do last time? I went to Google… and bought the item at Amazon.” The CSE experience is so short and bland that it is left entirely out of the equation.
This is partially by design. CSEs get money for every click lead they generate, so they’re often constructed to get users clicking out quickly and frequently. For instance, not displaying size information next to a shoe may increase revenue, since the user will then have to click out to see if the shoe is in her size.
We decided to try a different approach. We aim to present all the information a user might want on Modista itself, and try our best to normalize product data (size, widths, multiple views, etc.) from different retailers to streamline the user experience. We’re perfectly happy if a user spends all day at Modista, and only clicks out once — to actually buy an item.
It turns out that this approach has been working pretty well. Many people find using Modista to be a little exciting and addictive, and we think that the continuity of experience plays a major part in this. Users spend an average of 15 minutes on Modista, and returning users spend more than 20. We hope those 15-20 minutes will help them think of Modista the next time they want to buy some shoes or a handbag.
We’re always working on ways to improve Modista. Naturally, a main sources of information is our users — whether by unprompted feedback, site metrics (we track pretty much everything a user does), or videotaping sessions — so I’d love to hear your thoughts on how we can further improve the site!
The Noisy Channel 
What a fantastic and concise explanation of the difference between the way people interact with conventional web search engines and the way they seek information through online communities like Facebook, LinkedIn and Twitter.
Today I even had the opportunity to experience this difference first- hand: I used social search to replace my headphones. I first went to Hunch, which did a solid job of pointing me to a suitable pair of headphones. In fact, I ultimately purchased the headphones recommended as the 3rd of 53 possible choices. After getting recommendations from Hunch, I turned to Twitter, where I received a flurry of advice, as well as requests to share what I learned. Given that advice, I turned to a few ecommerce sites to follow up on a few candidates. And ultimately I bought a pair of headphones that will hopefully arrive before my current ones fall apart.
So, is Twitter a search engine after all? I still say no, but it certainly facilitates social search.