Adoption of chatbots—coded programs that can engage in some degree of conversation with human inputs, often through the help of artificial intelligence (AI) or machine learning—is undoubtedly a growing trend. There are thousands of chatbots in use today, on websites, messaging apps, and social platforms. It follows that bots would find a prominent place in the retail world.
Over the past several years, many companies have implemented AI technology in effective and scalable ways, and the chatbots may lead others to do the same. That said, business leaders should be aware of what chatbtos cannot do as well.
Though results vary, chatbot vendors claim that natural language processing (NLP) software can assist retailers with improving processes, converting leads, building brand loyalty, and growing customer engagement. We researched the space to better understand how chatbots come into play in the retail industry and to answer the following questions:
- What types of chatbot applications and software are currently in use in retail?
- What tangible results have chatbots driven in retail?
- Are there any common trends among these innovation efforts? How could these trends affect the future of retail?
This report covers four vendors offering chatbot software or applications for retail. We have selected companies that offer different methods of chatbot creation because we believe that these varied methods may appeal differently to different retail endeavors. All of the vendors highlighted below have demonstrated significant commercial success or are backed by PhD-level talent in their AI development efforts.
This report aims to inform business leaders in the retail space of what chatbot technology is currently available, enabling them to make educated decisions about chatbot adoption. Further, this report will hopefully reduce the time and effort retail business leaders spend searching for information on at least a few chatbot companies in the market today.
Established in 1995, LivePerson was one of the first companies to offer an online chat software that connected customers with company representatives strictly through a web-based interface. Since its inception, the company has purchased multiple startups centered on customer service, data analytics, and online chat. Early in 2017, LivePerson announced LiveEngage for Bots and claims it has since been integrating AI into its chat services. As of October 1, 2018, the company claims more than 40% of its conversations involved chatbots.
The LiveEngage software claims to “sell products and answer questions” by interacting with customers in high-traffic venues such as Facebook Messenger, WhatsApp, Line, and Apple Business Chat. According to the company, “Bots can help retrieve pricing information, product specifications & availability, highlight promotional offers, and provide shipping & installation information.”
The main use cases seem to be automating the 69.2% of retail interactions that LivePerson deems “highly suitable for bots.” Bots are intended to cover the more mechanical elements of retail selling, leaving human sales reps free to give more attention to complicated or escalated situations.
LivePerson further claims to simplify bot creation and management by providing its clients with a pre-built graphical user interface known as BotStudio. According to the company, BotStudio allows non-technical users, including “content creators and customer care professionals,” to create bots for any niche. These non-technical experts may act as bot managers and use the AI management console to monitor the performance of the bots.
This AI management console, per LivePerson, can be used to gauge customer satisfaction with chatbots, using real-time sentiment analysis. The console dashboard lists out all agents—both humans and bots—engaged in conversation, indicating customer details like country, browser, current page, and visit start time. It also tracks and shows a numerical custom metric called the Meaningful Connection Score (MCS)—an indicator of customer satisfaction with the conversation that uses NLP to synthesize whether the user seems positive, neutral, or negative about the conversation. When an MCS drops too low, LivePerson claims the bot will seamlessly transfer the customer to a live agent for further assistance.
When bots do not perform as desired—failing to respond effectively to user queries and allowing the MCS to drop—managers could use the LivePerson software to join the conversation, and the chatbots’ machine learning capabilities should allow them to synthesize manager replies and adjust responses over time. Chatbots and live responders may work simultaneously or separately, and businesses may also integrate third-party chatbots into customer chats through the LiveEngage dashboard.
LivePerson additionally offers analytics reporting to support any interactions made. The software generates reports on bot conversation time, bot conversion rate, and more.
A video depiction of LivePerson bots in action can be found on the LivePerson website.
Recently, Aramark, the food and drink supplier for multiple ballparks, partnered with LivePerson to do a trial run of an Apple Business Chatbot. Game attendees at the Citizens Bank Park were able to scan a QR code to initiate a chat with a bot, which provided a menu of drinks for the user to choose from. Fans could select the drinks they wanted and pay for them via Apple Pay. Service reps could then deliver the purchased beverages right to the customers’ seats. The 10-game trial run ended in August, but we were unable to find the details about customer engagement and returns—it may not yet be publicly available. Data specifying the relative successes of the LivePerson software would increase the relevance of this case study.
LivePerson represents many other brands, including The Home Depot, Citibank, Orange, and HSBC. Company CTO and Duke graduate Alex Spinelli was the global head for Alexa OS at Amazon.com before joining LivePerson.
Unlike LivePerson, which offers GUI bot building and messaging alongside live chat services, Pandorabots is an online web service dedicated exclusively to chatbots. The company provides an online space where users build and deploy their own bots using artificial intelligence markup language (AIML), a well-used open source language developed between 1995 and 2002. Pandorabots calls its developer’s space the Sandbox, and businesses may use the Sandbox to develop, test, and enhance their bots using original code and previously coded AIML libraries. The Sandbox also functions as an AIML interpreter, so developers are notified of coding compilation errors as they build.
Pandorabots also offers end-to-end bot building and management for larger brands.
Businesses may contact Pandorabots to open an account that suits their unique needs, or they may begin with a free account. This account provides a Quickstart tutorial to explain the features of the Sandbox, then allows the developer to choose a name and primary language for the new bot. The developer then uses AIML to build the bot and compose rules and patterns to dictate how a bot engages with users.
Bot creators can also draw from open-source libraries, as well as premium libraries that come with monthly fees. Pandorabots clients can even license codebases for existing bots, including Mitsuku, a four-time Loebner Prize–winning conversational bot that claims to use natural language processing and is “widely considered the best AI conversationalist,” according to the company.
Once a bot is complete, the developer may deploy it via web browser, SMS, or a variety of third-party services such as Slack or Twitter. Developers may return to the Sandbox at any time to edit the chatbot or to monitor a 30-day record of the bot’s interactions, number of clients contacted, number of chat sessions, and number of interactions per session. This preliminary data allows Pandorabots to assign a monthly fee of $0.0025 per message to any bot with more than 1,000 monthly interactions.
How each bot works—and the retail problems it could solve—depend on what the creator needs. For retail cases, bots could be tasked with presenting product information, directing users to product pages, and even answering consumer questions. In most cases, the bot takes user input—a text-based question or statement—and runs it through its developer-programmed categories and patterns to create a text-based, contextual response.
In 2016, American Eagle Outfitters deployed two chatbots on Facebook Messenger using the Pandorabots software. The bots responded to a variety of content, including customer service, product fit details, and shopping. Between November 2016 and June 2017, the bots exchanged over 4 million messages and achieved a 25% click-through rate to the American Eagle Outfitters website, though it’s not stated what American Eagle’s CTR was prior to implementing the bot. However, a full 75% of people using the Aerie bot were new to the brand. Pandorabots claims to also have worked with Coca-Cola.
In terms of AI talent at the company, founder Dr. Richard Wallace holds a Ph.D. in computer science and was the original creator behind the above-mentioned AIML (as well as the A.L.I.C.E. NLP chatterbot). Current talent includes Pandorabots’ active Senior Artificial Intelligence Designer, Steve Worswick, who also created Mitsuku.
Established within the past few years, mode.ai creates and manages chatbots with a more visual focus than other pre-built bots, making it a particularly relevant option to businesses in the retail shopping space. Bots can be deployed either on a company’s website or through Facebook Messenger. A customer interacting with the bot can upload photos or describe what they’re looking for via text, and the bot scans the client’s inventory data—which CBO Karen Ouk indicates the company can pull from existing store inventories on sites like Shopify—to find products that match or resemble the user’s request.
Retail often relies on visuals—human retail sales agents engage customers in conversations about aesthetics and visual preferences. To approximate that experience, mode.ai claims its bots leverage image analysis and computer vision to synthesize both user inputs and inventory to find matches to customer requests. Once a customer has found a product that they like, the bot can also simplify purchasing by either routing users to a product purchase page or even offering one-click pay for Facebook Messenger.
The below video shows a demo of how a customer could interact with a mode.ai stylist bot via text alone, and then with an image to show the type of product they’re looking for. The bot processes the text and image inputs, then searches a product inventory to return a matching item.
In an interview with WWD, Levi Strauss & Co.’s senior vice president, Brady Stewart, indicated that customers who interact with Levi’s Ask Indigo bot (powered by mode.ai) are between 50% and 80% more likely to convert—though Stewart doesn’t indicate what that increase is in comparison to. More clear parameters on exactly how mode.ai’s bot has bolstered conversions and over what time frame would lend greater authority to this particular use case, but even the lower end of the cited number range seems impressive.
Though mode.ai is relatively young, it may have set itself up as a solid player in the AI world, securing a spot on CB Insight’s AI 100 list in 2017. In addition to working with Levi’s, mode.ai has also worked with Louis Vuitton, Mavi, and ProFlowers. The company is also backed in its visual conversation elements by team members like Senior Engineer and Scientist Dr. Michel Vidal-Naquet, who has a Ph.D. in Computer Vision from the Weizmann Institute of Science.
Retail business leaders that seek the flexibility to choose between both end-to-end chatbot production and in-house chatbot development may investigate vendors such as Creative Virtual. Creative Virtual builds, deploys, and manages chatbots, or allows others to do so by contracting its software, known as the V-portal. The company claims its V-Person chatbots and live chat support could improve sales and brand loyalty, boost task efficiency, and increase cost savings.
Businesses that wish to employ Creative Virtual submit an online form and then consult with a team of customer-success advisors. According to the company, together, they devise a scalable strategy to be implemented across multiple platforms.
Initially, V-Portal serves as a space for data integration from a company’s internal databases and call centers. Based on that data, developers (either from Creative Virtual or within the business) can create a chatbot to suit the business’s needs, then deploy it across multiple platforms, such as SMS, Facebook Messenger, the company’s website, and more. Each interaction with the V-Person chatbot or virtual assistant is sent to the Natural Language Engine to evaluate the intent of the query and craft the appropriate response based on the source and language of the customer query.
V-Portal also offers many GUI components and controls for chatbot evaluation, management, and editing. It offers transcripts of customer interactions and in-depth visual analytics for unanswered questions, call deflection, conversion rates, and satisfaction levels. The video below explains the intent and the basic functions of V-Portal.
The company does offer retail-specific V-Person solutions that it claims can increase sales, streamline customer movement between channels, and improve support and training for sales agents. However, while Creative Virtual appears to provide clear case studies to document its application for clients in the financial and non-profit sectors (including the Commercial Bank of Dubai, Enterprise Bank, and the Royal Society for the Prevention of Cruelty to Animals), we were unable to find a published case study for any of the company’s retail clients.
It’s possible that outcomes from non-retail studies—from increased efficiency to increased conversion rates—could potentially apply to retailers, but a documented study of chatbot use and efficacy in the retail space would better verify the company’s claims.
Creative Virtual’s current CTO, Peter Behrend, has 18 years’ experience with natural language processing and AI work, and has been an influential driver behind the company’s technology development for over 14 years.
Takeaways for Business Leaders
For retail businesses with knowledgeable in-house developers, software like the Pandorabots Sandbox may provide a viable option for chatbot construction, deployment, and management. Retail business leaders with little access to technical expertise may find software with GUI, such as LivePerson, a more realistic option. Companies with little technical expertise but a sizeable AI budget could opt for more condensed bot production and management, offered by LivePerson, Creative Virtual, or mode.ai.
As of now, well-designed chatbots have shown tangible results in the retail sector, providing positive click-through rates, conversions, and AOV for online purchases. However, we would not expect significant results from superficial or ill-managed chatbots that fail to fully utilize consumer data.
As consumers evolve to become even more digital, we would expect an increased number of retail chatbots and an expanded role for these chatbots during the course of the next five years. Retail businesses—especially those with young, digital-native consumer bases—would do well to seek intelligent chat-based solutions to meet customers where they are.
Header Image Credit: Medium