Mention artificial intelligence (AI) and most of us think of driverless cars, virtual-assistant chatbots and computers that are better at playing the board game Go than people.
While these manifestations of AI technology capture imaginations and headlines, there’s a less visible AI revolution underway that is having a more immediate impact on everyday lives than futuristic robot butlers.
Behind the shopping websites and e-wallets at Alibaba Group, embedded into the digital infrastructure powering the world’s largest retail commerce company, AI is redefining the shopping experience for millions of China’s online shoppers and merchants. To users, some of these enhancements may seem subtle: product search results that have become uncannily precise; prescient search engines that draw attention to goods and services consumers might not have even realised they wanted; virtual storefronts that display information tailored to individual shoppers based on their unique characteristics and preferences; customer-service chatbots that can resolve consumer complaints without human intervention: orders that are delivered in hours instead of days.
Together these improvements to the “customer journey” — from product discovery to purchase decision to product delivery to after-sales service — have significant implications. While many technology companies are pursuing flashy AI-enabled products and services, Alibaba is spreading artificial intelligence throughout its core e-commerce operations, making it a cornerstone technology to speed the arrival of what company officials envision to be the future of not only e-commerce, but all commerce.
Later we’ll explain in more detail the role AI plays when people shop on Alibaba websites. First, some background.
AI and Alibaba
In implementing AI, Alibaba is applying numerous complex machine-learning methods and disciplines, such as online machine learning, transfer learning, deep learning, artificial neural networks and reinforcement learning. The latter mimics the way the human brain learns through experience. Simply put, the computer “learns” to recognise patterns in data that lead to right and wrong answers through an iterative process, similar to the way a child learns by trial-and-error to throw a ball.
This technology depends on the creation of software algorithms that enable computers to reach correct conclusions by being fed huge numbers of examples, building up a base of “experiences” that can be applied to recognise patterns and solve problems in new situations.
These learning machines become smarter and more accurate as they are given more and more data. This means Alibaba, with its vast computer resources and access to online data from hundreds of millions of consumers, is well positioned to introduce practical, real-world AI applications into everyday use. The company over the last 17 years has built up one of the world’s largest distributed-computing networks, allowing it to marshall the computational horsepower necessary for AI. Alibaba’s cloud computing subsidiary, one of the world’s top five public cloud providers, has developed an operating system called Apsara that organises data centers into a computational engine that can process more than 175,000 transactions a second.
In addition, Alibaba’s websites — including market-leading C2C site Taobao Marketplace and leading B2C site Tmall.com — as well as mobile apps have some 500 million active users and garner millions of visits a day. This offers a wellspring of big data on consumer behavior—shopping habits, payment and credit history, demographics, search preferences, social networks, personal interests and other information — that is fodder for learning machines.
“Large-scale computing and data are the father and mother of artificial intelligence,” explained Alibaba Executive Chairman Jack Ma at a recent internal company technology summit.
“Today, worldwide, companies that have the resources and platforms to truly develop artificial intelligence technologies are fewer than five. Successful companies must have both data and computing capacity, and also believe in high demand for AI. Alibaba is definitely one of the top three.”
Ma stressed that the field is still in its infancy. But according to a recent report from Goldman Sachs, machine learning is one of several inter-related technologies that are already having an impact on online retailing. Alibaba’s “big data, cloud services and the coming of age of machine learning technology should continue to deliver a personalised shopping experience to consumers and targeted marketing solutions to brands and merchants,” the Goldman report stated.
Here’s a closer look at how Alibaba is making AI part of the user experience:
Smart Product Search and Recommendation
Some search and recommendation engines use only historical data on what you’ve bought in the past to determine what products you might be wanting today. This is why, if you’ve just purchased, say, a wristwatch, you’re likely to be served up nothing but product recommendations for wristwatches —sometimes for the exact model you just bought.
To make search and recommendation more relevant and holistic, Alibaba has developed software it calls the “E-commerce Brain.” This system uses real-time online data to build models for predicting what consumers want, models that through AI are constantly updated for each individual to reflect not only past and recent purchases but also a range of online activities such as browsing, bookmarking, commenting and other actions.
Crunched by algorithms, this data allows the E-commerce Brain to determine correlations between content consumption and purchasing behavior. This generates a wider range of recommendations for not only products that consumers have shown an interest in previously, but also for related products and other information. The Brain can home in on a consumer’s predilections for different categories of products, price ranges, brands, product specifications and other key parameters.
“In the past, the operations team needed to select and determine patterns manually,” explained Zhao Binqiang, a former senior algorithm expert at the Alibaba Search Engine Division who now leads product recommendation for Alibaba’s digital marketing unit.
“Today, the job of matching patterns that was done by humans has been given to deep learning. In this way, we can apply the same set of algorithms in different promotions and scenarios, and standardise the process of recommending a wide range of things, covering stores, products, content, reviews and live broadcasts.”
Here’s a simple example of how this works in the real world: Taobao’s mobile app offers a feature called Taobao Headlines, a consumer information platform that serves up stories and articles to users according to their interests. Consider the case of a mother of a child with serious skin allergies whose doctor recommends she use soft cotton gauze towels for her baby to minimize irritation. She buys the towels through Mobile Taobao.
The next time she opens the app, Taobao Headlines has recommended an article offering advice on dietary supplements for infants with allergies. The mother had not searched for this information—the E-commerce Brain algorithm had already learned that many women who buy cotton gauze towels are also browsing information on child allergies. After reading the article, the mother might then become interested in purchasing dietary supplements for her child.
What makes Alibaba’s AI especially powerful is that data is integrated not just from Alibaba shopping websites but throughout the company’s ecosystem of owned and affiliated internet assets, including transaction and credit data from e-payments service Alipay, location data from location-based services provider AutoNavi, search data from mobile browser UCWeb, and content consumption data from video site Youku.
This not only results in a broader dataset of behavioural information for better recommendations. It also means Alibaba can recommend offline local services—not just products—in real-time.
Smart Customer Service
Most people have played around with AI-driven virtual assistants like Apple’s Siri, Amazon’s Alexa or Google Home. At Alibaba, chatbots are all business. If you call Alibaba with a question or a complaint, more likely than not you’ll be connected not to a human customer service agent, but to a computer system called Ali Xiaomi (Ali Assistant).
Ali Xiaomi, which handles both spoken and written queries, focuses on e-commerce services, acting as customer-service rep and personal shopping assistant. In addition to providing answers to frequently asked questions and answering questions about specific transactions such as delivery status, the chatbot can help users find products when provided with a text or voice description or even a photo, returning a list of recommendations that users can filter by brand, color and other characteristics.
Capable of accessing personal information about individual callers and continuously upgrading its capabilities by analysing millions of customer-service interactions, Ali Xiaomi can also help users with more general assistance, such as topping up phone credits, checking weather and purchasing airline tickets.
The chatbot’s combination of AI and speech-recognition technology has already paid dividends in efficiency and lower costs compared with human-based customer service—and may ultimately replace call-center staff altogether. Alibaba says Ali Xiaomi is capable of handling up to 95 per cent of customer-service enquiries. The technology is proving so successful, a text-based, customizable version of the chatbot is being offered free to all merchants on Alibaba platforms so they can automate their own customer service departments.
Online merchants are always striving to better engage with customers, thereby increasing click-through rates and ultimately conversion rates (sales). Alibaba is helping them achieve this through highly targeted, AI-assisted marketing. Merchants selling on Tmall and Taobao are able to personalize their virtual storefronts for individual visitors, offering real-time, tailored product recommendations based on purchase histories, age, gender, geographic locations and a host of other data points.
“You can think of Taobao and Tmall as a virtual mall with billions of products,” said Wei Hu, director of data technology in Alibaba’s Merchant Service Business Unit. “The challenge is how to select a few products to display based on the extensive shopping behavior over the entire site. In many cases a store is new to customers, so there is no previous shopping behavior. Our challenge is to decide which products to display to random store visitors and boost the conversion rate.”
The solution: teach AI algorithms to make individual-specific predictions, matching products with purchaser and ranking them according to which items the visitor is most likely to be interested in. “If the customer has shopped in the store before, we will match products purchased previously with new products in the pool,” said Wei. “If the customer has not shopped in the store before but shopped elsewhere, we will match products purchased elsewhere with new products in the pool.” This all happens in the blink of an eye, and algorithms are continuously improving as new data is fed into the system.
Webpage personalisation has been applied beyond storefronts. During last year’s 11.11 Global Shopping Festival, Alibaba’s mega 24-hour online sale, Alibaba for the first time achieved personalization for all of its China retail marketplaces, from the Taobao and Tmall homepages to special promotions pages to product details pages. During the 11.11 sale, some 6.7 billion personalized shopping pages were generated by more than 230,000 merchants on Taobao and Tmall—striking evidence of the scalability of Alibaba’s AI technology.
The payoff: personalized landing pages had a 20-per-cent higher conversion rate during 11.11 compared with non-personalised pages, according to Alibaba.
Smart Supply Chain
To upgrade supply chain management, Alibaba has developed what it calls the Ali Smart Supply Chain (ASSC), which applies AI to help online and offline merchants forecast product demand; prepare, replenish and allocate inventory for optimal turnover; determine the right products to offer; and choose appropriate pricing strategies.
The overall aim is to enable quick business response to shifting consumer tastes based on trends sniffed out of transactional data, allowing merchants to balance supply with demand and more efficiently coordinate the flow of merchandise–hence reducing costs and lost sales due to out-of-stock items. The system also helps synchronise and coordinate the activities of service providers to better manage product design, manufacturing, inventory and delivery.
ASSC uses an AI predictive model to forecast demand for newly introduced products as well as demand for different types of products during large promotional events. The model processes a range of historical and real-time data, including seasonal and regional variations as well as consumer preferences and behaviors.
During Alibaba’s 11.11 Global Shopping Festival last year, ASSC’s auto procurement and allocation system optimized merchandise availability by directing inventory to warehouses throughout China, increasing turnover by taking into account regional demand forecasts. Many customers received their orders the next day despite the huge volume of deliveries because Alibaba was able to predict regional demand for certain types of products and place inventory in the closest warehouses in advance.
Cainiao Network, Alibaba’s logistics affiliate, was founded in 2013 with a group of shipping and delivery companies to create a logistics information platform that links a network of partners, warehouses and distribution centers. As of August 2016, Cainiao was processing data for 70 per cent of parcel deliveries in China, an average of 42 million a day, while connecting more than 90 domestic and international partners.
These feats are made possible by Cainiao’s big data and smart technologies that power an intelligent warehouse management system, digital waybill system, and computerized parcel sorting and dispatch at distribution centers.
Cainiao’s engineers apply a number of cutting-edge AI technologies to speed parcel delivery. For example, the company is using GIS (Geographic Information System) and AI techniques to train a computer model to determine the fastest and the most cost-effective delivery routes in a variety of complex road networks, including both rural villages and crowded urban areas. Cainiao recently announced a partnership with major Chinese automakers to manufacture 1 million green-energy delivery vehicles equipped with Cainiao’s AI technology, which helps to cut vehicle use by 10 per cent and travel distances by 30 per cent.
Cainiao also uses AI technology to predict the size of boxes that should be used to efficiently pack orders consisting of items of various sizes and weights. The company said its solution reduces the use of packing materials by more than 10 per cent.
Alibaba is also working on AI technology that is not all about e-commerce. Alibaba Cloud, the company’s cloud-computing subsidiary, is developing an all-purpose “brain” called ET that correctly predicted the winner of a popular reality TV show and helps improve traffic flow in the city of Hangzhou, China.
Meanwhile, Alibaba’s fintech affiliate Ant Financial has deployed AI-enabled face-recognition technology for making secure electronic payments that was recently mentioned by MIT Technology Review in its “10 Breakthrough Technologies” list for 2017. Alibaba established an AI lab in mid-2016 that has a growing team of about 100 people, most of them AI developers and data scientists who are working with companies such as Mattel on smart toys and other smart products. The company also established an internal “Institute of Data Science and Technology” with a mission to significantly improve the core technologies of Alibaba Group in speech, image, video, machine learning, and natural language processing.
“We should not be worried about how much the internet is impacting traditional business,” Ma said recently. “Rather, we have to use internet and AI to our advantage … Machines can do what people can’t. We must make machines our best partner, rather than letting them replace us.”