Chatbots. Conversational agents. Software that talks back! AI-enhanced computer programs able to hold audio or text-based conversations that simulate convincingly how a human would interact… Whatever you want to call these code-based dialoguers, they are taking the enterprise by storm. From marketing, customer service, and e-commerce bots helping people purchase everything from flowers to flights, to workforce productivity ‘manager’ bots delivering reminders, deadlines, and tips tailored to individual employees… companies are deploying bots to serve all manner of objectives.
Such initiatives are also a common entry point for enterprise adoption of machine learning, but they are also just the tip of the iceberg when it comes to the opportunities and the risks of and artificial intelligence (AI). What follows are three areas companies often overlook when deploying chatbots.
1. Chatbots are extensions of brand persona
Chatbots are now being used by brands for service interactions, including simple outreach, education, feedback and survey collection, questions and answers (Q&A), tips and advice, etc. Many brands are using chatbots to extend the brand as a “friend,” easing pressures to buy by developing such bots with personality and the ability to engage far beyond the scope of sales or customer support. While this introduces new ways to scale insights and personalization, it also means companies suddenly find themselves depending on AI to communicate and convey the character of their brand.
This is, for the most part, unchartered territory for brands, but never more relevant as companies of all sorts are deploying these virtual avatars and assistants, conversational agents, robots, and other anthropomorphic brand extensions. Marketers and Communications should have a clear sense of who the brand is and what qualities it stands for, and these elements must now manifest in any kind of AI-driven manifestation of the brand. What follows are a list of new considerations:
• Brand persona (e.g. consistent with brand character/avatar; brand values; brand tone, etc.)
• Language used
• Triage, escalation plan (e.g. when to triage to human agent; communications training for human agent)
• New legal liabilities (e.g. what happens when users communicate potentially sensitive information to a branded bot, such as threats, crimes, suicidal information)
• New social liabilities (e.g. male vs. female bot persona; dress and class personification; responses to users’ provocation; all manner of unforeseen social
In 2016, Microsoft launched a chatbot called Tay. But in the course of 24 hours, it began spouting outrageous hate speech, as bots are trained, often to some unknown degree, on user interactions and public data. Microsoft quickly shut the bot down, but it was enough time to see unintended consequences of releasing such an ‘impressionable’ bot loose to the Twittersphere. (Image source: NBC)
Even in less dramatic cases, our inflated expectations around AI, combined with the limitations of the technology, mean risks of failure are high. Consumers expect to have a natural conversation with a chatbot like they would a human, without real topical constraints. But if the data isn’t there, the bot will fail.
Key takeaway: AI doesn’t just represent new opportunities for scaling brand engagement and personalization, it represents new requirements for brand governance and liabilities, and new risks of botched experiences, public relations (PR) crises, misrepresentation, or backlash.
Check out: Top Chatbot Companies
2. Chatbots need back up
Chatbots are merely the ‘last mile’ of information delivery. Although these efforts are often born of single departmental efforts, their enterprise value cannot be overlooked. While the front office may enjoy significant efficiencies as bots help triage, react, respond, and provide faster, and more personalized services, chatbots must be incorporated into broader ‘learning’ across the organization. For example, data collected through or as a result of
• Sales interactions can help organizations inform marketing or ad spend; channel strategies
• Product interactions can help optimize product development and design, or product sourcing
• Service interactions can help inform inventory replenishment, partnerships
• Relationship-building interactions can help automate in-store or online environments and layouts, labor allocations
Maturity in the data space is not AI—although advanced analytics are increasingly likely to use AI-based techniques. Instead, maturity must be viewed as a function of how touchpoints, products, services, systems, and people are interconnected to drive ongoing learning and innovation for the enterprise.
Key takeaway: Deploy chatbots with an eye for how data can inform other departmental learnings and efficiencies and actually create better user experiences.
3. Chatbots will eventually go away
For all the buzz and development in bots today, there are a variety of developments underway which are likely to dissolve the bot craze into something more operationalized and integrated, and less point-centric. Today most chatbots serve very narrow applications (e.g. top 10 frequently asked questions triage), but leave much to be desired when conversations veer into other problems, topics, products, never mind other business functions.
One reason for the de-emphasis of point solution bots is that a bot is only as good as the data that feeds it, and the more data and context available to feed bots, the better their capabilities will become. This takes time, and more data, more training, and ongoing tweaking of models. For example, consider the difference in utility between a travel bot that can only serve up recommended flights based on dates versus one that ‘understands’ trends in trip types based on user profiles, has data from millions of past interactions and conversions, mines additional data feeds for local hot spots.
Another reason chatbots will fade is that they are becoming one of numerous interaction modalities between consumers and AI. Consider, for instance, the rapid growth of
• Voice interactions, both hardware-based like via smart speaker, and those integrated into chatbot interactions on mobile or web)
• Virtual assistants, a much-debated name for a software agent capable of performing complex tasks for an individual, such as IT repair services, cross-jurisdiction or multi-system inventory availability, or (in the case of Google), creating an AI agent tailored to deliver personalized recommendations, prompts, and troubleshoot issues for each of its billion+ daily Google Assistant users.
• Conversational interfaces, in which NLP-based interactions are built into websites and actually inform web experiences such as language preference, design paths, or visual search (see image below)
A company called Sentient powers Visual Search, a AI-based way to both engage shoppers in product discovery and train the system on user preferences. (Image source: Shoes.com)
In addition to a widening range of modalities, both complimentary to and distinct from chatbots, maturity in the space will be characterized by who can deliver the most seamless and appropriate interaction experience at the right time. Sometimes, chatbots may be most opportune; other times a voice interaction; other times a real live human! All the time, the success of these interaction will be a function of the data, context, and yes, human touch, such interfaces foster. The imperative for the human element addressed is, thus, addressed first in this article; and the of data integration, second, b. Both will, over time, render chatbots less a novelty, and more just another tributary contributing to, and extracting learnings from across an enterprises’ circulatory system of information.
Key takeaway: Consider chatbots one of an ever-growing array of interaction modalities; not superior or inferior, but a tool to be deployed, managed, trained, and improved using both data and humans.