Using conversational AI to help consumers make better choices

With a growing number of consumers gravitating towards online shopping, e-commerce businesses find themselves confronting the joint challenges of acquiring and retaining customers. Brands are trying to meet these challenges  by incorporating more human elements in an effort to delight the customer and build long-term brand loyalty. A key part of every e-commerce interaction is the product search, so we see a strong desire among online stores  to deliver a product search experience that is more reflective of best-in-class real-world experiences, in which an intelligent, courteous, and helpful store assistant is available at the right time to guide the shopper’s search.

As seen in an earlier article, marketplaces and e-commerce sites have recognised the need to enrich the online shopping experience with the human touch. Achieving this means  focusing on a  consumer’s individual needs and preferences  to create a more personalised, natural style of interaction in which the customer is supported and guided throughout their shopping experience

Recently, the industry has witnessed a proliferation of chatbots and Virtual Agents (VAs). For some applications such as customer support, these reactive communication approaches can work well.They are typically placed in their own window, largely isolated from the user’s site browsing experience, but can such approaches really humanise the search experience in e-commerce? We believe that the fact that these are generalised solutions, not focused on individual user’s needs, makes them unsuitable for it, creating the need for an altogether new approach.

The Traverz agent goes beyond chatbots and VAs by delivering a more immersive and more personalised product search experience. It does this by taking advantage of the natural human desire for interaction and conversation, by combining  conversational AI with personalisation, rendered as part of a frictionless User Experience (UX), thereby creating a more natural and intuitive dialog with consumers. The value generated from this greatly improves user experience, fostering loyalty. So, how can conversational AI evolve to truly personalise the product search experience? Let’s have a look how chatbots and VAs work and how the Traverz agent goes beyond these technologies to help consumers make better choices.

What are these chatbots and VAs anyway?

Virtual agents and chatbots are software robots that are able to engage in a Natural Language conversation with users, and can even solve certain user problems autonomously. For example, they can solve customer service issues and provide answers to common questions about a contract or a service within a company. Users might encounter chatbots and VAs when filing an insurance claim or when engaging with a company’s customer service department over the web or via their messaging channel. Increasingly such agents represent a first line of response for many inbound customer queries by offering a more ‘human face’ to more conventional FAQs or email-based communication.

However, the terms ‘chatbots’ and ‘Virtual Agents’ are often unknowingly and erroneously used interchangeably to mean the same thing.

Chatbots are typically deployed in situations that are relatively constrained in their scope and technology. Most of them are question-answer or process-based solutions designed to solve either simple customer service requests or simple business processes. Typically, the user clicks on the chatbot icon and types a customer service-related input. The chatbot then finds the best matching answer from a set of alternatives in the knowledge base.

Source:Botmock 

In comparison, VAs are more advanced solutions that go beyond simple rule-based chatbots, and beyond the single question-answer dialogue set-up. They do not follow pre-programmed rules. Instead, they rely on advanced technologies such as Machine Learning (ML), that allow them to learn from examples and evolve beyond rule-based recognition. The ML also deals with new, unseen user inputs, and tries to solve these based on past experience.

Source: Microsoft Azure

By redirecting inbound customer service inquiries to chatbots and VAs, enterprises can expect to save anywhere between 15-70% in customer service costs. This, coupled with improvement in standard platforms for building and integrating VA solutions, and overall technological advances in VA capabilities, have spurred the growth of these digital assistants. Together, they are estimated to grow at a CAGR of 30% between 2019 and 2024, demonstrating their growing popularity.

It is  no wonder that this has led to a promise and belief that VAs can easily deliver cost-efficient online customer service and smooth user experiences by engaging in meaningful, natural conversations with end-users. However, many VA solutions are unable to live up to this promise. Frequently their lack of flexibility and disjointed aesthetic fail to provide the sort of experience that users require, especially when they have pressing customer service requests.

 


How do these chatbots and VAs actually work under the hood?

Chatbots and VAs are conceptually built around 4 key modules:

  • The NLU Module: This interprets the user’s Natural Language input and performs entity and intent recognition. In the simplest of cases, these technologies are implemented through a lot of ‘if-then’ rules, which identify certain keywords in the input. The resulting chatbot solution is little else than a large set of rules and a matching algorithm. In advanced VAs, however, the NLU module is based on an ML approach that learns and adapts with increased usage.
  • Dialogue Management: This module controls the flow between the user and the VA. For example, in a simple chatbot that interacts during an insurance claim, the dialogue management simply runs through each of the process steps, while asking questions and collecting the required information along the way.
  • Knowledge Base: This module maps the user’s questions and situations to answers that lie in the knowledge base.
  • Response Generation: Here, the answer that is matched in the knowledge base is processed into a human-like response using Natural Language Generation (NLG) technology.

 

Does the Chatbot and  VA approach work?

Natural Language Interaction is a powerful and modern way to engage with customers, with a strong human touch. So conceptually, VAs and chatbots appear to be headed in the right direction. Both can work diligently, though only within simple stipulated frameworks that eliminate any need to stretch their limits. Viable scenarios typically include supporting simple, customer service requirements by providing answers that are drawn directly out of a knowledge base. They work fairly well in cases where users provide personal details or other specific information during the interaction. These inputs prime the VAs and chatbots with information that is used by them as a starting point to begin delivering user-specific support – quite similar to a situation where users call customer support and provide details in order to initiate a query pre or post-purchase.

The point we must stress here is that VAs and chatbots are designed to be reactive, which means they need the user to initiate a discussion. They are set up as fully independent processes, with their contextual scope limited by the information provided by the user. Which is to say, they lack the native ability to either probe or widen the scope of the interaction for understanding the context of the user’s current situation. It comes as no surprise that VAs and chatbots aren’t well-positioned to provide the user with a human-assistant-like support during an online shopping journey, especially when it comes to selecting and evaluating a product.

A human-like virtual assistant changes that paradigm; it possesses advanced capabilities that translate into value for users. For instance, it can share relevant information with users, such as return policies while also engaging with them in their search for products. It does all this by making suggestions, asking questions, observing how users browse, etc. The assistant does this proactively and iteratively. It can do both – appear or vanish from the screen in a non-intrusive manner whenever it needs to respond to a user’s situation, and also engage in a continuous dialogue if need be.

E-commerce requires a more context-aware and interactive conversational agent

Successful e-commerce experiences are all about enriching a customer’s product search experience, not only by helping them to find a product or service that satisfies their needs, but also helping them to refine their needs and to better understand the relevant trade-offs that may exist in a particular product search-space. In other words, successful e-commerce is about guiding a customer through a journey of discovery and of supporting them on this journey,  highlighting interesting landmarks along the way and by guiding them to the right product destination given their personal circumstances. But like any good guide, it is important to know when to speak up and when to stay quiet, when to nudge one way and when to warn against likely dead-ends.

Platform developers are invested in providing users with web or app-based visual environments that provide access to a set of products to select from. But while this ‘web-UI’ has been refined for over 20 years now, it remains a static and re-active environment. This calls for the power of the natural-language-driven agent-based approach to be fully integrated with the platform itself, rather than being added (much like an afterthought) in a separate window. The team at Traverz has accomplished this feat, and at the same time also introduced several improvements in the web-UI.

To create a fulfilling product search experience, a conversational AI agent must support the user in the product search process – based on a full understanding of the user’s situation. The agent also needs to fully comprehend the terminology of the product vertical the user is searching in, and go well beyond the conventional question-answer mechanism before offering contextual advice and suggestions.

This is where the Traverz conversational AI agent shines. By adopting an inside-out approach, in which the agent is uniquely focused on the individual’s current situation as well as their interaction history, it is possible to interpret the user’s actions using the many insights that can be gleaned from their past history and current context. This is not to say that the Traverz agent does not leverage other inputs too. It can avail of all available product information and related user conversations to learn and fine-tune the agent’s conversational ML model, while retaining a central focus on the individual user.

It’s easy to notice that the Traverz solution is a far cry from the typical ML and Big Data approach, where developers wade through gargantuan volumes of user input data before eventually creating a general Natural Language Processing (NLP) model for the VA. The VA can then make somewhat intelligent interpretations of the user’s inputs, and generate reasonable responses to questions. Such a general model is always used in the same way, based on standard assumptions about a standard user’s situation. If these assumptions are off, the interpretation of the user’s intent and entities may be incorrect, and the responses will be flawed.

The Traverz conversational AI agent benefits from  a much more precise perspective on the user’s current situation, their recent search history, and their observed preferences – simply because the user has explicitly stated them. Consequently, it understands the search purpose and engages in a much smoother dialogue, making itself immensely useful. Notably, the Traverz agent does not rest its dependence on direct user requests alone, as in a standard NLP request-response loop. Rather, it offers support and guides users in several different ways throughout the product search process. To illustrate, the agent makes individual and personalised suggestions for important features, highlighting product attributes and sharing insights concerning the current search situation. In this way, businesses can provide consumers with human-assistant-like support during their product search process.

Summing up

The key element of the human search experience, or any human conversation in general, is the understanding of the user’s situational needs. Conventional chatbots and VAs offer general solutions, which typically do not factor in the user’s individual situation. Although their underlying technology is sound, their specific implementation approach renders them unsuitable for supporting consumers in their e-commerce product search journeys.

The Traverz agent redefines the product search experience. It centres around a user’s current situation, goes beyond the simple question-answer loop, and provides individual suggestions and situation guidance through truly personalised and natural interaction. Importantly, it connects directly into the site’s Web-UI and therefore forms a seamless part of the overall product search process. In this way, the Traverz agent is able to help consumers make better choices – which after all, leads to the win-win outcomes that e-commerce sites and consumers are both looking for.

To find out how you can build an e-commerce platform that delivers better insights and greater brand loyalty from customers, get in touch with us.

Share