Artificial Intelligence is paving the way for opportunities in retailing unimaginable a few years ago. The subtle presence of AI tools working behind the scenes are helping them make better purchase decisions. So much so that it is predicted that in 5 years, 85% of consumers’ relationships with a company will be managed without even interacting with a human being. Online vendors are using AI to promote sales by improving customer experience. One of the ways that this is happening is by improving customer personalisation. Until recently, this would have amounted to little more than seeing your first name appear when you logged into your account. But now, AI provides styles of personalisation that can benefit the consumer and vendor.
These are some of the ways that this is happening:
Recommender systems
Recommender systems have been used in retail businesses for more than a decade but have matured significantly in recent years. Machine learning, a branch of AI, has provided ways for vendors to gain insights from customer behaviour. It’s been used in retailing since the late 1990s to analyse large amounts of customer data – such as customer purchasing history – and from doing so, gain insights for decision making. This use of machine learning predated e-commerce on the Web as the data was usually from in-house databases of customer transactions. The AI technique that was used enabled vendors to make decisions that could improve sales –it was known as “data mining”. One of the best-known examples of this was the unexpected identification of the correlation between beer and diaper purchases in a retail chain in 1998. This discovery led the store chain to speculate that this might be due to young fathers who needed to make a trip to the store to purchase diapers. The trip, they believed, would prompt the young fathers to reward themselves with beer purchases whilst there. As a consequence, the chain decided to position these items so that they were next to each other in the store and sales rose as a result of this decision. Many other examples of the benefits of data mining followed.
Nowadays, a high proportion of sales are online. This means that colossal amounts of customer data are available on the Web. This combined with improvements in machine learning techniques has led to a rise in the development and use of recommender systems. These systems use data analysis techniques to find and recommend items that are likely to match and appeal to a consumer. The pioneers of these systems were Amazon and Netflix but they are now also being used with providers of services, such as the social networking site LinkedIn – which can recommend people to connect with. Recommender systems help consumers to find products and services that they may like – such as books, films, and music. They recommend personalised content based on purchases made or through ratings given by consumers – both implicitly through past purchases and explicitly through customer ratings on Web sites.
There are two main techniques used for implementing recommender systems. They are content-based and collaborative filtering. Content-based, as the name suggests, makes a recommendation based upon the content of the item. That is, they use product item features only. As a simple example, a purchaser of a Beethoven 5th symphony CD, may also like Beethoven’s 7th symphony CD since they are a similar musical genre and the same composer. Collaborative filtering, on the other hand, looks at correlation preferences for groups of users and, based on their similarities, a customer’s taste could be predicted and recommended. Using the same example, if many purchasers of a Beethoven symphony CD also purchased and liked a Haydn symphony CD in significant numbers, then it may be recommended. Correlated items would be listed and filtered to produce the highest-ranked fits. Collaborative filtering can use large amounts of data and, like the diaper and beer example quoted earlier, can reveal completely unexpected insights. Recommendation systems based upon customer purchases are now very common – and quite effective because they continuously learn and refine their recommendations so that eventually their suggestions coincide closely with the consumer’s taste.
Improving search for customers
Online customer interactions are increasingly driven by AI. This is not surprising because customer needs vary hugely. One of the ways that this is happening is through customer search product queries. Google is the leader in search and, over the years, have improved their main search engine using AI. Google uses something called “knowledge graph”. It collects an array of facts of world knowledge gathered from a variety of Web sources to present as summarised content. This enhances the search experience for the user by presenting the knowledge in an infobox next to the search results. This knowledge is automatically extracted from Web pages. However, many vendors use on-site search engines whose purpose is to know what products the consumer wants to buy. This can be difficult because their search engines have to deal with fuzzy queries from customers – such as “do you have a battery that fits a Toyota Auris car?” or “do you have a Duke Ellington album that features Louie Bellson on drums?”. Amazon, who account for 54% of all online product searches, is developing a search engine that can learn to predict the context from their customers’ search queries as well as recognize synonyms, and so on. But from the first query, there might be more information needed before a match could be found, such as the type of car, engine size, fuel type, and so on. Chatbots (see next section) are also now being used to handle this type of query.
Chatbots for customer interactions
Chatbots are automated online help assistants that can enable human sales representatives to concentrate on high-value interactions. Chatbots, are AI programs that engage in conversations with humans to help solve problems, usually via natural language text messages. They work by receiving query’s or requests from users and then writing a response in natural language. A dialogue can then ensue, rather like a conversation between two people, that can continue between parties until one side or the other terminates the conversation.
In the first generation chatbots that were used a decade ago, the responses that they generated were quite limited because they were programmed to identify keywords like “sale”, “discount”, “returns” and so on, and then use “canned text” scripts to respond. They were without learning capabilities and therefore seen as providing little more than novelty value. Nowadays chatbots can understand the user intent better and extract better responses from the system. These responses contain generic text or information stored on a database that is contextualized for the requester, or a follow up question so that the chatbot response can clarify the requester intentions. Chatbots can nowadays learn by considering each conversation it has with the customers – so that they can improve future responses. For instance, they could learn to identify synonyms or colloquialisms that customers use, or they could try to learn to identify customer intentions – even if those intentions are written or spelt incorrectly. They can assist in sales by providing information on everything, such as product specification, explanations of product usage, and more.
Online chat help has become ubiquitous in recent years using human operators. But augmentation, or even replacement, with chatbots, is now a common trend. AI has changed everything by allowing companies to use chatbots and virtual assistants to answer common customer service questions. Natural language using spoken voice can replace textual dialogue. This has enabled their use in call centres where customers can perform more automated tasks, such as helping to reset a password. As these technologies become more mainstream, privacy and security issues will need to be addressed, because whilst chatbot conversations may be conducted anonymously other data collected for machine learning might be considered sensitive.
Intelligent agents
Intelligent agents are AI programs that can be built to perform useful tasks on behalf of the customer. They are proactive in the sense that they become active following some change in a business environment. For example, a customer who visited a product Web page might have been alienated by the price of the product. An intelligent agent might activate if the price falls in the future and alert the user. This may then trigger a sale. The intelligent agent could interact with the customer could be via a pop-up window notification on a laptop, or perhaps via a banner and sounds on a smartphone, depending on their preferences. Spoken voice delivery could also be used, such as via the Apple iPhone spoken voice assistant Siri. Intelligent agents can also learn and adapt to user preferences by using machine learning. This is important because a customer’s taste may change over time. Customer transactions can be facilitated by using spoken voice intelligent assistants, such as Alexa and Siri, and can even link up with retailers to enable customers to order products by voice.
Email personalisation
Personalisation of email content is an important factor in communication between customer and vendor. Customers rate personalisation highly. In the past, vendors have used hand-crafted email composition for marketing products. This would have been time consuming and an expensive use of employee time. But now AI is using machine learning algorithms to record customer browsing behaviour and therefore understand their content preferences. This knowledge helps the algorithm to create personalized emails.
AI services for vendors
Vendors wanting to incorporate AI tools in their Web sites would not need in-house expertise to develop them. A range of development support services are available for implementing personalized content and chatbots. For example, Azure provides support for developing agents, knowledge mining, and more. They also offer support for developing chatbots. Pandorabots, also specialise in chatbot services for a range of businesses.
Conclusions
There are, as this article shows, potentially many benefits for vendors who invest in AI. Computers don’t draw salaries, get tired, or go on sick leave. And chatbots are not going to get upset by angry customers taking it out on them. Consumers are more likely to feel valued from the personalised experience that they can get from the AI tools described in this article. Furthermore, in an age of availability, customers want immediate answers, so having access to constant support at any time on any day is essential. Vendors embracing AI are more likely to make that happen.
Keith Darlington
Comments on this publication