By the end of this year, m -commerce-to-wallet company Paytm will have all its category pages made by machines. Paytm wants to respond to each customer who visits its website looking for fashion wear or sports gear or iPhone covers in a personalised way. The website will offer customer choices based on past usage and social media posts. For example, if you recently went to Goa on a holiday and posted photos of the trip on Facebook, you might get ‘beach-themed’ iPhone covers. The idea is to hook customers with what they prefer. Given that the choice of iPhone covers run into several hundreds, a customer visiting an online marketplace might not have the patience to browse through all the pages and options.
But by throwing up just what he desires, the website might be able to coax him into buying. Online marketplaces like Paytm call this conversion rates the number of visitors who end up buying stuff. Helping them bump up conversion rates are algorithms. Algorithms are responsible for customers browsing for goods being greeted with shopping recommendations. Algorithms decide what to display for online marketplaces.
They keep track of what customer are browsing and buying. “The goal is to improve conversion rates and help the industry become profitable,” says Vijay Shekhar Sharma, founder, Paytm. How does it work? Internet merchants are swamped with mind-boggling flow of data for example, Paytm has about 30 lakh visitors every day with about 3 million page views daily. Algorithms help it crunch data on customer preferences and increase sales. “Algorithms are the base for everything online shopping, shipping, packaging, payments, price points etc,” says Sandeep Aggarwal, founder, Shopclues. com, an e-commerce marketplace. The importance of algorithms becomes stark looking at the current online marketplace conversion rates. It is at less than 3% compared with that of offline retail at 22-25%.
Algorithms will also underpin the future of ecommerce companies. There was a time when these companies could live with that poor statistic. Not anymore. They are stacking up $150-200 million in losses every month, throwing good money at customer acquisitions and deep discounts. Profitability was not a priority. But now they face a funding squeeze and pressure from investors to show profits. Pragya Singh, vice-president, retail, Technopak a retail consultancy, says the focus until now was on topline growth. “In the last few months it’s about how to come out of deep discounting and show profits.” Flipkart has been downgraded twice in the last four months by investors Morgan Stanley and T Rowe Price.
In March, the Department of Industrial Promotion and Policy, the nodal agency for investments, while allowing 100% FDI in pure marketplaces banned deep discounts, predatory pricing and ‘big billion sales’. With no room for manoeuvring prices to attract buyers, the route to achieve better conversion and reduce losses is big data analysis and algorithms. Praveen Bhadada, partner and practice head, Zinnov, a Bengaluru-based management consulting firm, sees the reliance on algorithms as the second wave of ecommerce in India. “The first wave was about getting the model right, getting people used to the idea of shopping online. Now, a sizeable customer base is there (about 55-60 million internet users shop online) and in the second wave companies are using algorithms to improve profitability,” says Bhadada.
Data as a Weapon: At any given time, there are 3 to 4 million visitors online. They spend an average of seven minutes viewing 8-10 pages. By the end of the day, about 15 million records are generated. ComScore data for February for all etailers shows 52.98 million unique visitors, 4.42 billion page views and about 55 minutes a visitor a month. The minutes spent on e-shopping leave a trail and clues that companies want to dive into.
What was the shopper looking for? What are his previous purchases? What device did he use? How many times has he visited the website? “We have to use this basic data what did a person do for strategic advantage. So, if a user has not logged in for 3-4 days the listing might be stale and the algorithm refreshes it. If a customer does a lot of cancellations, the cash on delivery option for him is automatically disabled (the customer might be doing it just for fun),” says Aggarwal. Generating traffic is not the problem for etailers. Getting customers to buy is. “We are super ambitious about using data to help a person find what he is looking for.
This will increase conversion rate and improve profitability,” says Rajiv Mangla, CTO, Snapdeal. “We want to detect patterns in user behaviour to improve conversion.” A number of companies are already using algorithms to improve conversion rates. Ugam Solutions is a Bengaluru based data analytics company whose clients include leading ecommerce platforms such as eBay, LG and Staples. The company analyses data for clients and offers signalswhat inventory to carry, what models are trending, what are users searching for and what competition is carrying.
Say a marketplace wants to dominate luxury watches segment, should it carry the whole inventory from Rolex to Rado or focus on brands like Breitling or Chopard which have the more likelihood of sales. Mihir Kittur, co-founder & CEO, Ugam Solutions, says India is a growth market where the belt has tightened. To be sure, companies are looking at data with renewed interest.
Saurabh Vashishtha, vice-president Paytm says his company “stores everything”. “There’s a huge push to dynamic content from static a year back.” So if six months back all visitors saw the similar content on each category page, now Paytm has a better idea and displays content based on what the algorithm picks up. Deepali Tamhane, senior director, product management, Flipkart, says the company is working towards achieving the next level of personalisation. “We want to provide our users with what they want, even before they know they want it, of course with their consent to use their data.”
Finding the Sweet Spot: Adds Bhadada, “in the small window the user logs in the goal should be to understand what she wants and not carpet bomb. At furniture e-tailer Pepperfry, if a popular product, like the Disney almirah for kids becomes too common, more people may not buy.
“There comes a point when it should be moved out. That point is not determined by humans but machine learning software,” says Sanjay Netrabile, CTO, Pepperfry. Pricing is another aspect where algorithms become handy. Consultancy PricewaterhouseCooper’s (PwC) leader data and analytics Sudipta Ghosh says humans decide on pricing products for offline retailers. In the online world, with millions of simultaneous transactions, this decision is taken by data analytics. “If price point is too low people might perceive it as too cheap to buy and abandon purchase. This point is determined by algorithm,” says Ghosh.
According to Bhadada, all types of data is useful and outside the platform as well, in logistics, shipping, warehouses 5-10% can be saved if data is correct. Algorithms help a logistics firm to decide on the best delivery route. Most companies use Hadoop hbase (server software) to analyse big data, machine learning tools like R & Paython, which use data to create business models and web traffic data analytics from Alexa, Google Analytics or Adobe’s Omniture. Besides the big data analytics tools, inhouse teams write codes for specific outcomes.
Snapdeal has a 25 people data engineering teamwhich essentially determines what kind of data to collect and a 25 people data science team which analyses the data collected and tweaks the algorithm. At a broad level, it could be to push cricket memorabilia or IPL gear in the current season and at micro level, it could mean wooing a Kolkata Knight Riders fan with a KKR T-shirt, a taste picked up from Facebook. “We have to create that intelligence in conversion; else it could misfire,” says Mangla of Snapdeal. For example the goal could be to maximise sales.
So “shoppers see products from sellers whose returns are lower. The software could also note that certain brand of shirts at a price point of `800 are selling fast, but all sellers are not getting orders. It could determine the reason as poor quality of the catalogue and alert the seller”, says Vashishtha. He says conversion rate has gone up 50% in the last six months due to better intelligence. Adds Aggarwal of Shopclues, “Data analytics is science and delivers better return on investment than any other system like marketing or advertising.”
Shopclues transactions have improved ten times in the last 12 months thanks to algorithms compared with the traditional approach of mass advertising. Pepperfy uses a sorting algorithm that detects a potential shopper in Mumbai or Delhi who have different needs based on the character of the cities they reside in. The former gets to see contemporary styles and space saving furniture while the latter gets options in solid wood, with less concern on space saving designs. “The goal is to get to know the sweet spot,” says Netrabile.
Kittur of Ugam Solutions believes data analytics can lead to 3-7% improvement in bottom line and at least 40% improvement in conversion rates in the short term. At present there are 50-60 million online shoppers and 400 million internet users. With rising internet users, more shoppers are expected to come online. Karthik Bettadapara, CEO, Dataweave, says, “The industry is shifting from blind discounting to targeted analysis.
Now funding is tight, marketplaces have to be smart about spending money.” Dataweave, funded by Google honcho Rajan Anandan, Blume Venture and others, looks at even external data to create intelligencelike what is the competition selling, at what price points, what products are people buying and so on. As companies like Paytm move to completely automated systems, the goal is to be like Uberdo real time analytics to predict where demand is and multiply the chance of success. Sharma of Paytm says Uber has among the best data science teams in the business. “We would eventually like to do it real timemeet a buyer’s demand at almost all times.” Wes Hopkins Authentic JerseyShare This