Today’s audience choice for TC’s Startup Battlefield — plucked from the packed exhibition floor of Silicon Alley to pitch on the Disrupt SF 2014 main stage — is Donde, a Silicon Valley and Israel based ecommerce startup that wants to make it easier for shoppers to search for items based on how they look.
Co-founder Liat Zakay said the idea was sparked by the frustration of searching online for very specific items. “This experience is very, very difficult and frustrating, [I was] spending a lot of time,” she told TechCrunch. “It was clear to me that everything is going online it’s only a matter of time that a very quick and easy solution will be there. This is why I started working on Donde.”
“This technology… creates a very unique way to fetch a specific item — we created an engine that analyses items and finds the right questions to ask the user for her to create a seamless and delightful experience when she has an intent to buy,” she added.
The initial focus for Donde’s visual search app is clothes — and even more specifically dresses — but Zakay says its technology can be applied to any type of item that can be visually distinguished. So the plan is to expand the business to other product categories in time, once the UX has been fine tuned.
Donde is currently launched in private beta as an iOS app — which went live about two months ago — with hundreds of thousands of dresses in its inventory and an initial group of user testers at the University of Southern California.
The app works by letting users select what color of dress they’re looking then stepping them through a series of questions about other aspects of the dress — such as what its sleeves look like, or the shape of the hemline. These choices are displayed visually, with graphics that display silhouettes of the shapes/styles to choose from.
As the shopper selects each style point the search narrows and the app continues to refine the actual items it displays below these icons — which are being pulled in from clothes’ retailers’ websites. Users can then click on an item and hit a ‘buy’ button to purchase a dress they like the look of.
“Our technology is crawling the websites of brands and retailers and we extract the unique features of the items — either from the image or the description. We create a unique tool that does it. And then we create an ontology, a common language of all the brands, because a certain feature [in the products of one brand] is not the same as another — [for example] the red of H&M is not the same as Zara,” said Zakay.
Visual search makes plenty of sense if you’re looking for an item of clothing of a specific shape or style — say a dark green backless dress with an asymmetrical hem, three-quarter length sleeves and a peplum. Searching for such an item by chaining words together is a pretty clumsy option, and can also require specialist vocabulary to describe exactly what you mean. So it’s a hit and miss process, and often a time-sink. Donde aims to fix that.
We have had a visual retail search startup (Asap54) in TC battlefield before, back in October last year at Disrupt Europe. But Asap54’s focus was on taking photos and building a tech that could locate similar items. Donde’s interface does not require the user to have a photo to begin their search. Its system is better described as a quasi-‘build your own dress’ interface — matching whatever the user constructs with real-world items they can actually buy.
The tech underpinning Donde also factors in contextual information such as the user’s age and location as it makes decisions about the type of dresses to present to them as potential purchases. So, in other words, a teenager in Alabama is not going to be shown the same dresses as a 40-something woman in New York. It’s tailored product search that aims to be a whole lot smarter.
Inspiration for the questions based interface came from the ‘twenty questions’ game whereby players try to guess the identity of a famous person based on asking about specific attributes, according to Zakay. “We thought of why not taking that to the product world, where I need to find a product and I know what I’m looking for,” she said.
“I would need just a computer that would ask me the right questions and would lead me in seconds to finding it, using artificial intelligence of the contextual information — who you are, what’s your location, what’s your environment.”
Donde’s current business model is based on taking a commission on any affiliate sales generated by the app — it’s not processing payments itself but rather using retailer APIs — but Zakay sees potential to build a data driven business in addition to a commissions revenue stream, based on the real-time trends data the app could generate.
“We’re already are building and developing the tools for you to see in a certain spot what people are searching for, which kind of products they are interested in, and analyzing it based on the characteristics of your products that would be very beneficial for the companies to get real-time information about what your customers want,” she added. “Also in predicting the designs and real-time inventory, what is more popular.”
Donde was founded more than a year ago and has brought in a small round of angel funding thus far. The startup also spent four months in UpWest Labs, an accelerator for Israeli startups. It’s now looking to raise a seed round.
The three co-founders have a background working in Israeli military intelligence, including Zakay who spent four years in an elite tech unit in the Israeli military. “We’ve developed a lot of computer, cyber-security projects and after I graduated in computer science and economics and started working on the earlier version of Donde,” she says. “I didn’t want to be boring like all of my unit and go into cyber security because this is too regular.”
The team’s time in the Israeli army brought it into proximity with various data analysis tools — and that laid the groundwork for the technical foundation of Donde, adds Zakay.
Donde’s dev team remains in Israel but the startup is incorporated in the U.S. and has established an office in Silicon Valley — with its focus being fully on the U.S. as its initial target market.
Q: You guys built scrapers… how do you make sure that your product universe is relevant?
A: Our tech tools are scraping the data a few times per day, making sure the info is updated. It’s really important for the shopping experience
Q: Could you license the tech to existing platforms?
A: Yes it’s definitely an option. We’re currently focusing on optimizing virality and user engagement but we’ll definitely build partnerships
Q: What’s your business model?
A: On every purchase we get a commission
Q: Why is this viral or social?
A: First of all this is intuitive mobile search for every visual product that has characteristics and sold online. We chose to start with fashion vertical because you can choose to share it with friends when you need advice; this is exactly what we’re doing right now
Q: So far you have dresses?
A: Right now it’s dresses but the tech is generic for every visual product and we’re working right now on expanding it to all the categories, brands
Q: There’s some pretty good and big search companies out there, what’s the hardest thing you’re doing here?
A: First we’re focusing on the mobile experience, second we’ve built a tech that creates a very intuitive way for people to search visual products. There are a lot of products trying to do image recognition by taking a product – we have a wider proposition. It doesn’t have to be in front of you to do it. We are very focused on our tech and the way we build it. It’s unique. Proprietary algorithm. The idea of the questions
Q: You’re building taxonomies to understand different clothing types?
A: We have hundreds of features we can ask about a product. It’s all the intuitive ways that people use to describe something. We use everything we can from the website and item description… and then we use our algorithm that analyses the user to decide the questions
Q: How do you acquire users?
A: We’ve successfully proved we’re viral within sororities. We’re still working on optimizing this. At a later stage we’ll focus on user acquisition
Q: What’s the incentive not to serve the most expensive dress to the user?
A: The most important thing here is to be an objective search engine, not biased on price or partnerships with different retailers. We can monetize in a lot of different ways