The next time you shop on fashion website Myntra, you might end up choosing a t-shirt designed completely by a software - the pattern, colour and texture-without any intervention from a human designer. And you would not realise it. The first set of these t-shirts went on sale four days ago. This counts as a significant leap for Artificial Intelligence in e-commerce.
For customers, buying online might seem simple--click, pay and collect. But it's a different ballgame for e-tailers. Behind the scenes, from the warehouses to the websites, artificial intelligence plays a huge role in automating processes.
Online retailers are employing AI to solve complex problems and make online shopping a smoother experience. This could involve getting software to understand and process voice queries, recommend products based on a person's buying history, or forecast demand.
So what are the big names doing?
"In terms of industry trends, people are going towards fast fashion. (Moda) Rapido does fast fashion in an intelligent way," said Ambarish Kenghe, chief product officer at Myntra, a Flipkart unit and India's largest online fashion retailer.
The Moda Rapido clothing label began as a project in 2015, with Myntra using AI to process fashion data and predict trends. The company's human designers incorporated the inputs into their designs. The new AI-designed t-shirts are folded into this label unmarked, so Myntra can genuinely test how well these sell when pitted against shirts designed by humans.
"Till now, designers could look at statistics (for inputs). But you need to scale. We are limited by the bandwidth of designers.The next step is, how about the computer generating the design and us curating it,"Kenghe said. "It is a gold mine. Our machines will get better on designing and we will also get data."
This is not a one-off experiment.E-commerce, which has a treasure trove of data collected over the last few years is ripe for disruption from AI. Companies are betting big on AI and pouring in funds to push the boundaries of what can be done with data.
"We are applying AI to a number of problems such as speech recognition, natural language understanding, question answering, dialogue systems, product recommendations, product search, forecasting future product demand, etc,"said Rajeev Rastogi, director, machine learning, at Amazon.
An example of how AI is used in recommendations could be this: if you started your search on a retailer's website with, say, a white shirt with blue polka dots, and your next search is for an shirt with a similar collar and cuff style, the algorithm understands what is motivating you.
"We start with personalization - it is key. If you have enough and more collection, clutter is an issue. How do you (a customer) get to the product that you want? We are trying to figure it out. We want to give you precisely what you are looking for,"said Ajit Narayanan, chief technology officer, Myntra.
A related focus area for AI is recommending the right sizes as this can vary across brands. "We have pretty high return rates across many categories because people think that sizes are the same across brands and across geographies. So, trying to make recommendations with appropriate size is another problem that we are working on. Say, a size 6 in Reebok might be 7 in Nike, and so on," Rastogi said in an earlier interview with ET.
Myntra uses data intelligence to also decide which payment gateway is the best for a transaction.
"Minute to minute there is a difference. If you are going from, say, a HDFC Bank card to a certain gateway at a certain time, the payment success rate may be different than for the same gateway and for the same card at a different time, based on the load. This is learning over a period of time,"said Kenghe. "Recently, during the Chennai cyclone, one of the gateways had an outage. The system realised this and auto-routed all transactions away from the gateway. Elsewhere, humans were trying to figure out what happened.
Support from AI specialists
A number of independent AI-focused startups are also working on automating manually intensive tasks in ecommerce.Take cataloging. If not done properly, searching for the right product becomes cumbersome and shoppers might log out.
"Catalogues are (usually) tagged manually. One person can tag 2,000 to 10,000 images. The problem is, it is inconsistent.This affects product discovery. We do automatic tagging (for e-commerce clients) and reduce 90% of human intervention,"said Ashwini Asokan, chief executive of Chennai-based AI startup Mad Street Den. "We can tag 30,000 images in, say, two hours."
Mad Street Den also offers a host of other services such as sending personalised emails to their clients' customers, automating warehouse operations and providing analysis and forecasting.
Gurugram-based Staqu works on generating digital tags that make searching for a product online easier. "We provide a software development kit that can be integrated into an affiliate partner's website or app. Then the site or app will become empowered by image search. It will recognise the product and start making tags for that,"said Atul Rai, cofounder of Staqu, which counts Paytm and Yepme among clients. Staqu is a part of IBM's Global Entrepreneurship Program.
The other big use of AI is to provide business intelligence. Bengaluru-based Stylumia informs their fashion retailer clients on the latest design trends.
"We deliver insights using computer vision, meaning visual intelligence,"said CEO Ganesh Subramanian. "Say, for example, (how do you exactly describe a) dark blue stripe shirt. Now, dark blue is subjective.You cannot translate dark blue, so we pull information from the Net and we show it visually."
In product delivery, algorithms are being used to clean up and automate the process.
Bengaluru-based Locus is enabling logistics for companies using AI. "We use machine learning to convert (vaguely described) addresses into valid (recognizable) addresses. There are pin code errors, spelling mistakes, missing localities.Machine learning is critical in logistics.We even do demand predictions and predict returns,"said Nishith Rastogi, chief executive of Locus, whose customers include Quikr, Delhivery, Lenskart and Urban Ladder.
Myntra is trying to use AI to predict for customers the exact time of product delivery. "The exact time is very important to us. However, it is not straightforward. It depends on what time somebody placed an order, what was happening in the rest of the supply chain at that time, what was its capacity. It is a complicated thing to solve but we threw this (challenge) to the machine,"said Kenghe.
"(The machine) learnt over a period of time. It learnt what happens on weekends, what happens on weekdays, and which warehouse to which pin code is (a product) going to, and what the product is and what size it is. It figured these out with some supervision and came up with (more accurate delivery) dates. I do not think we have perfected it, but it is a big deal for us."
The next big challenge
One of Myntra's AI projects is to come up with a fashion assistant that can talk in common language and recommend what to wear for various occasions. But "conversational flows are difficult to solve.This is very early. It will not see the light of the day very soon. The assistant's first use would be for support, say (for a user to ask) where is my order, (or instruct) cancel order,"said Kenghe.
The world over, conversational bots are the next big thing. Technology giants like Google and Amazon are pushing forward research on artificial intelligence. "As we see (customer care) agents responding (to buyers), the machine can learn from it.The next stage is, a customer can say `I am going to Goa' and the assistant will figure out that Goa means beach and give a list of things (to take along),"Kenghe said.
While speech is one crucial area in AI research, vision is another. Mad Street Den is trying to use AI in warehouses to monitor processes. "Using computer vision, there is no need for multiple photoshoots of products. This avoids duplication and you are saving money for the customer...almost 16-25% savings on the operational side. We can then start seeing who is walking into the warehouse, how many came in, efficiency, analytics, etc. We are opening up the scale of operations,"said Asokan.
Any opportunity to improve efficiency and cut cost is of supreme importance in ecommerce, said Partha Talukdar, assistant professor at Bengaluru's Indian Institute of Science, where he heads the Machine and Language Learning Lab (MALL), whose mission is to give a "worldview"to machines.
"Companies like Amazon are doing automation wherever they can... right to the point of using robots for warehouse management and delivery through drones. AI and ML are extremely important because of the potential. There are a lot of diverse experiments going on (in ecommerce). We will certainly see a lot of innovative tech from this domain."
For customers, buying online might seem simple--click, pay and collect. But it's a different ballgame for e-tailers. Behind the scenes, from the warehouses to the websites, artificial intelligence plays a huge role in automating processes.
Online retailers are employing AI to solve complex problems and make online shopping a smoother experience. This could involve getting software to understand and process voice queries, recommend products based on a person's buying history, or forecast demand.
So what are the big names doing?
"In terms of industry trends, people are going towards fast fashion. (Moda) Rapido does fast fashion in an intelligent way," said Ambarish Kenghe, chief product officer at Myntra, a Flipkart unit and India's largest online fashion retailer.
The Moda Rapido clothing label began as a project in 2015, with Myntra using AI to process fashion data and predict trends. The company's human designers incorporated the inputs into their designs. The new AI-designed t-shirts are folded into this label unmarked, so Myntra can genuinely test how well these sell when pitted against shirts designed by humans.
"Till now, designers could look at statistics (for inputs). But you need to scale. We are limited by the bandwidth of designers.The next step is, how about the computer generating the design and us curating it,"Kenghe said. "It is a gold mine. Our machines will get better on designing and we will also get data."
This is not a one-off experiment.E-commerce, which has a treasure trove of data collected over the last few years is ripe for disruption from AI. Companies are betting big on AI and pouring in funds to push the boundaries of what can be done with data.
"We are applying AI to a number of problems such as speech recognition, natural language understanding, question answering, dialogue systems, product recommendations, product search, forecasting future product demand, etc,"said Rajeev Rastogi, director, machine learning, at Amazon.
An example of how AI is used in recommendations could be this: if you started your search on a retailer's website with, say, a white shirt with blue polka dots, and your next search is for an shirt with a similar collar and cuff style, the algorithm understands what is motivating you.
"We start with personalization - it is key. If you have enough and more collection, clutter is an issue. How do you (a customer) get to the product that you want? We are trying to figure it out. We want to give you precisely what you are looking for,"said Ajit Narayanan, chief technology officer, Myntra.
A related focus area for AI is recommending the right sizes as this can vary across brands. "We have pretty high return rates across many categories because people think that sizes are the same across brands and across geographies. So, trying to make recommendations with appropriate size is another problem that we are working on. Say, a size 6 in Reebok might be 7 in Nike, and so on," Rastogi said in an earlier interview with ET.
Myntra uses data intelligence to also decide which payment gateway is the best for a transaction.
"Minute to minute there is a difference. If you are going from, say, a HDFC Bank card to a certain gateway at a certain time, the payment success rate may be different than for the same gateway and for the same card at a different time, based on the load. This is learning over a period of time,"said Kenghe. "Recently, during the Chennai cyclone, one of the gateways had an outage. The system realised this and auto-routed all transactions away from the gateway. Elsewhere, humans were trying to figure out what happened.
Support from AI specialists
A number of independent AI-focused startups are also working on automating manually intensive tasks in ecommerce.Take cataloging. If not done properly, searching for the right product becomes cumbersome and shoppers might log out.
"Catalogues are (usually) tagged manually. One person can tag 2,000 to 10,000 images. The problem is, it is inconsistent.This affects product discovery. We do automatic tagging (for e-commerce clients) and reduce 90% of human intervention,"said Ashwini Asokan, chief executive of Chennai-based AI startup Mad Street Den. "We can tag 30,000 images in, say, two hours."
Mad Street Den also offers a host of other services such as sending personalised emails to their clients' customers, automating warehouse operations and providing analysis and forecasting.
Gurugram-based Staqu works on generating digital tags that make searching for a product online easier. "We provide a software development kit that can be integrated into an affiliate partner's website or app. Then the site or app will become empowered by image search. It will recognise the product and start making tags for that,"said Atul Rai, cofounder of Staqu, which counts Paytm and Yepme among clients. Staqu is a part of IBM's Global Entrepreneurship Program.
The other big use of AI is to provide business intelligence. Bengaluru-based Stylumia informs their fashion retailer clients on the latest design trends.
"We deliver insights using computer vision, meaning visual intelligence,"said CEO Ganesh Subramanian. "Say, for example, (how do you exactly describe a) dark blue stripe shirt. Now, dark blue is subjective.You cannot translate dark blue, so we pull information from the Net and we show it visually."
In product delivery, algorithms are being used to clean up and automate the process.
Bengaluru-based Locus is enabling logistics for companies using AI. "We use machine learning to convert (vaguely described) addresses into valid (recognizable) addresses. There are pin code errors, spelling mistakes, missing localities.Machine learning is critical in logistics.We even do demand predictions and predict returns,"said Nishith Rastogi, chief executive of Locus, whose customers include Quikr, Delhivery, Lenskart and Urban Ladder.
Myntra is trying to use AI to predict for customers the exact time of product delivery. "The exact time is very important to us. However, it is not straightforward. It depends on what time somebody placed an order, what was happening in the rest of the supply chain at that time, what was its capacity. It is a complicated thing to solve but we threw this (challenge) to the machine,"said Kenghe.
"(The machine) learnt over a period of time. It learnt what happens on weekends, what happens on weekdays, and which warehouse to which pin code is (a product) going to, and what the product is and what size it is. It figured these out with some supervision and came up with (more accurate delivery) dates. I do not think we have perfected it, but it is a big deal for us."
The next big challenge
One of Myntra's AI projects is to come up with a fashion assistant that can talk in common language and recommend what to wear for various occasions. But "conversational flows are difficult to solve.This is very early. It will not see the light of the day very soon. The assistant's first use would be for support, say (for a user to ask) where is my order, (or instruct) cancel order,"said Kenghe.
The world over, conversational bots are the next big thing. Technology giants like Google and Amazon are pushing forward research on artificial intelligence. "As we see (customer care) agents responding (to buyers), the machine can learn from it.The next stage is, a customer can say `I am going to Goa' and the assistant will figure out that Goa means beach and give a list of things (to take along),"Kenghe said.
While speech is one crucial area in AI research, vision is another. Mad Street Den is trying to use AI in warehouses to monitor processes. "Using computer vision, there is no need for multiple photoshoots of products. This avoids duplication and you are saving money for the customer...almost 16-25% savings on the operational side. We can then start seeing who is walking into the warehouse, how many came in, efficiency, analytics, etc. We are opening up the scale of operations,"said Asokan.
Any opportunity to improve efficiency and cut cost is of supreme importance in ecommerce, said Partha Talukdar, assistant professor at Bengaluru's Indian Institute of Science, where he heads the Machine and Language Learning Lab (MALL), whose mission is to give a "worldview"to machines.
"Companies like Amazon are doing automation wherever they can... right to the point of using robots for warehouse management and delivery through drones. AI and ML are extremely important because of the potential. There are a lot of diverse experiments going on (in ecommerce). We will certainly see a lot of innovative tech from this domain."
0 comments:
Post a Comment