Visual search start-ups help companies enable their users to discover products online, based on photos of objects in the real world
Bengaluru: When graduate student Anupama Pasumarthy shops online, she says she is always disappointed by the recommendations.
“It’s tough finding clothes (which are always too large) or shoes (which are too small) in my size, and I don’t find stuff that’s similar. I used to shop online but these days I just go to the store when I want to buy something," the 22-year-old says.
Tech-savvy millennials like Pasumarthy are the demographic that most fashion retailers target, but the problems she faces are all too familiar for anyone who shops online. To help retailers overcome this, start-ups like Stylumia Intelligence Technology Pvt Ltd, which offer artificial intelligence-based solutions for smart visual recommendations, have started to take off.
There are multiple start-ups globally trying to crack visual search. Visual search start-ups help companies enable their users to discover products online, based on photos of objects in the real world. In India, companies like iLenze (which raised $500,000 in funding last year) and SnapShopr (which raised an undisclosed amount of angel funding) offer visual search platforms.
Chennai-based Mad Street Den, which raised $1.5 million in 2015, also offers visual search, but its most used offering is a visual-recommendation engine, which sifts through catalogue data to show relevant recommendations to users.
With e-commerce booming in India, Singapore-based Visenze, whose visual search offering is used by companies like Flipkart, is setting up operations in India to cater to the demand.
Many visual search companies cater to multiple verticals, and have so far concentrated on consumer applications.
Started by former chief operating officer of Myntra, Ganesh Subramanian, and machine learning scientist Ram Prakash, who developed Quillpad, the first machine learning based language input for Indian languages, Stylumia is different. It focuses only on fashion, and using the same core technology, it is looking to help both consumers and businesses make data-driven decisions.
“We are developing a technology which takes natural images, videos, be it Bollywood videos or TV serials, whatever influences fashion, and decipher and extract fashion elements from that," says Prakash.
The start-up then hopes to use this derived intelligence in two ways – one, to make smarter recommendations to consumers browsing for products and two, to give suggestions to fashion buyers and retailers what to buy and make, based on real world consumer-purchasing data.
Right now, Prakash says that decisions at fashion companies are made based on some analytics, but intelligence based on visual cues is missing.
“They look at the patterns and say this is doing well because this is a red colour T-shirt with a contrast collar, what they cannot do right now is look at the same red colour T-shirts with contrast collars which are not doing well. They do not have a way to see all the relevant data together. That’s another problem that we are trying to solve," says Prakash.
Stylumia is set to launch its product in the first week of April. It currently has partnerships with retailers (which they it does not want to disclose before the product launch), says chief executive officer Subramanian. For now, it is using a team of four engineers to capture and label data but hope to automate this process very soon.
“There’s a lot of interest among both online and offline retailers in this space. Our aim is to provide the most accurate prediction of demand and our consistency will improve as we work with more retailers and brands and get more and more data," says Subramanian.
Despite the progress in technology, unless there is an overhaul on the supply-chain side, real impact is difficult to create, says Devangshu Dutta, chief executive officer, Third Eyesight, a New Delhi-based consulting firm.
“Large retailers today are planning several months in advance and are structured in such a way that by and large it takes them several months to respond to any particular trend. Till you can address that, data is just data," he said.