Most Chatbots Are Mediocre, and Marketers Should Intervene
Chatbots have a bad reputation, and it’s well deserved—they disappoint most people who try to use them. As a result, brands that deploy a chatbot risk damaging their own reputations by association.
Does that mean marketing professionals, as brand architects and stewards, should steer clear of conversational AI? No, on the contrary, because the main reason most chatbots disappoint is that so many companies treat their chatbot effort as an IT initiative and fail to take a human-centric design approach.
Of course, no chatbot can be effective without solid technology, which requires care and feeding from developers and other technologists. But although the best technology professionals are highly human-centric in their work, most are not because of the priorities baked into their training and their employers’ metrics.
Marketers have an opportunity—and, in my view, a responsibility—to be part of the solution. Although the people outside of IT with the most depth in human-centered design are, unsurprisingly, designers, the design team’s closest natural partners and allies are the smart marketers. By “smart marketers,” I mean those who see the big picture—who understand the importance of retention, not just of acquisition, and therefore of the importance of creating experiences that deliver the ease, effectiveness, and emotional resonance that will not just get customers to give the brand a try but keep them coming back. They embody the purpose of marketing set forth by Peter Drucker: “to know and understand the customer so well the product or service fits him and sells itself.”
That may seem inspiring but a bit abstract, so here’s what it means concretely. When it comes to chatbots, that purpose entails four things: understanding customers’ real needs; deciding which of those needs to address with the chatbot; designing the chatbot to align with the brand; and nurturing the chatbot over time so that it understands your customers better and better. Let’s look at each of these in turn.
Understanding customers’ real needs, in chatbot lingo, boils down to identifying a set of intents that the chatbot should be able to respond to. The term intent comes from the user-experience discipline, and an intent can be informational (such as inquiring about a store’s hours of operation) or transactional (such as replenishing a recurring order). The most common way to identify an initial set of intents is to comb through logs of text-based chats that customers have had with human agents and through transcripts of call center conversations. A more sophisticated approach also includes conducting discovery research to identify adjacent intents: needs that your brand could address but that customers don’t typically think to ask about.
Next, deciding which needs to address with the chatbot requires triaging first: winnowing the long list of identified needs down to the subset of intents that are both appropriate for a bot to handle and that a bot is capable of handling. It then requires prioritizing: ranking the intents based on factors such as how frequent they are and how easy they will be to implement.
Designing the bot to align with the brand is crucial, because the fact that chatbots interact with people using natural language causes customers to anthropomorphize them much more than they would a traditional app or website user interface. This means that the bot’s word choice, tempo, syntax, and other linguistic attributes must convey a tone that aligns with the brand’s attributes. The same goes for the bot’s extralinguistic attributes, such as its name (if it has one), its gender (if one is implied), its visual appearance (if it has an avatar), and how it sounds (which could be as basic as a chime to indicate that it has responded, if it’s a text-based chatbot, or as complex and subtle as its tone and manner, if it’s a voice-based chatbot).
Finally, nurturing a chatbot over time so that it understands customers better and better is a consequence of the open-endedness and infinite variability of human languages, which is unlike the deterministic and finite nature of traditional graphical user interfaces. It requires frequently reviewing the list of requests the chatbot could not handle and improving the chatbot so it does better next time—either by equipping it with new rules if it’s a deterministic chatbot, or using machine-learning training methods if it’s AI-based.
Beyond these concrete steps for creating an effective chatbot, there’s also an underrecognized benefit of chatbots to marketers: Customers express, in their own words, exactly what they want and expect from the brand. What a gold mine for insights into customer needs! It’s something that even the most carefully instrumented graphical user interfaces and the most sophisticated analytics and heat maps have never been able to provide.
I’m bullish about chatbots’ long-term potential, despite the fact that most of them today are mediocre at best, but it’s going to require attention from people who understand what human-centered design means and who are willing not just to observe from the sidelines but to act and lead to design better chatbots.
David Truog is vice president, principal analyst, at Forrester Research. Truog researches innovation, tools, and best practices at the intersection of technology and design, with a focus on conversational AI (voice and text virtual assistants or chatbots), the metaverse, extended reality (augmented reality, mixed reality, and virtual reality), and other emerging technologies.