By Adrian Bowles
(Aragon Research) – AI chatbots—software that accepts natural language input, interprets or classifies it, and then responds with natural language output or by triggering a system action—are experiencing a period of rapid adoption to augment or replace people in customer-facing roles. AI chatbots can now go well beyond improving operational efficiency to enable more effective employees, processes, and enterprises.
For most applications, AI chatbots will be part of a solution, not the solution itself. AI chatbots can enable conversations with customers, partners, and employees, but for all but the simplest requests, they should be integrated with more robust applications to provide the functionality indicated by the user’s intent.
How AI Chatbots Work Best
AI chatbots work best within a well-defined domain—such as customer service requests for a telecommunications provider—and a well-defined set of possible actions, such as “provide a refund,” “initiate a service/repair request,” “add a service,” “cancel a service,” or “elevate the call to a supervisor.” Applications that deal with more general domains or a broad range of tasks require logic that should be kept apart from the AI chatbot. It is a better practice to have the application itself incorporate the logic and algorithms to deal with the domain or task complexity and relegate the AI chatbot to managing communication between the user and the various application components.
As AI chatbot technologies become more powerful, yet simpler and less expensive to implement, they will become pervasive as the de facto interface that defines the user experience for a wide range of enterprise applications. With that in mind, it is important to understand some risks as well as the obvious opportunities of these technologies.
AI Chatbots Are an Extension of Your Enterprise
When an AI chatbot acts as an agent or representative of an enterprise—e.g., responding to online or email inquiries—it can establish or transform the relationship between the enterprise and its customers. As the interaction occurs using natural language, users naturally ascribe human attributes to the system and their impression of the software’s performance influences their impressions about the intent or competence of the enterprise it represents.
A great user experience can establish or build a positive relationship and lasting goodwill, while a poor experience can close doors and poison the reputation of even the most well-intentioned firm.
Key Factors to Evaluate When Building or Buying Your AI Chatbot
Building or buying an AI chatbot is a lot like hiring a person to represent your enterprise and its values and competencies to the world. Choose—or design—wisely.
Do:
- make sure the AI chatbot has sufficient domain knowledge to handle the most frequent requests without external intervention.
- maintain memory or state information within and between customer interaction sessions to personalize the service.
- have a consistent and obvious escape path for the user to exit a conversation, with an option to quickly escalate the session to a human
- track the depth of the conversation and use data from your own customer interactions to have your application guide the AI chatbot. Your application should recognize scenarios that indicate growing or impending frustration.
- allow the AI chatbot to respond under all circumstances. “I don’t know, let me find out” is a better alternative than endless questioning or irrelevant responses when the AI chatbot and application are unable or unlikely to find an appropriate solution.
- Whenever possible, have the AI chatbot learn from experience to improve its performance in future interactions with the same customer, or others that require similar actions or responses.
Don’t:
- require the user to know local/specialized non-standard domain-specific vocabulary.
- have any conversational path that could lead to a dead end. There should always be an obvious path to the previous or initial state.
- require too much user effort for any one input utterance. A conversational interface should be simple.
- ask for the same information twice. Failure to maintain memory/state within and between customer interaction sessions creates the (valid) impression that the system is serving the application rather than the user.
Make Your Chatbot an Asset, Not a Detraction
A well-designed and executed AI chatbot is an asset that will perform tirelessly and reflect well on your enterprise, while a poorly conceived or executed AI chatbot may be an efficient but significant liability.
You can read more about AI chatbot technology, trends, and emerging providers in our Hot Vendors in AI Chatbots report for 2017.
To continue the conversation on your individual AI journey, contact us.