By the end of 2015, more than half a million Americans had professed their love for Alexa, Amazon’s digital assistant/chatbot. It shouldn’t come as a surprise then to learn that in 2016, more than a quarter of a million Americans had proposed to Alexa. Fortunately, Amazon had the foresight to program Alexa with a number of witty and noncommittal answers for just such an occasion. Though silly, this anecdote illustrates an important point: US consumers are becoming more accustomed to and comfortable with interacting with artificial intelligence and brands are paying attention to this trend. Of businesses surveyed, 80% said they planned to integrate AI into their customer service efforts by the year 2020.

Of course, not everyone is in love with chatbots. While humans may not be able to tell the difference between communicating with a chatbot and communicating with a human early on in the conversation, it usually becomes apparent within the first several exchanges. Chatbots seem weird, or even creepy to some and many feel that humans still provide superior customer service. Since getting humans to accept chatbots is a huge step in the chatbot revolution, chatbot developers need to provide them with five key traits if they are to succeed.

Just a touch of personality

Creating a personable chatbot is a tricky balancing act. On the one hand, people don’t like chatbots to be too cold and impersonal. On the other hand, chatbots that are too life-like tend to creep people out. The balance, then, is to create a chatbot that is merely pleasant but not overly conversational. An effective chatbot will efficiently and accurately assist customers all while remembering it’s manners–greetings, please’s and thank you’s, etc…)

The ability to delegate

Artificial intelligence and related fields such as natural language processing simply haven’t come far enough yet to create chatbots that can handle any and all customer service issues. There will be questions that chatbots can’t answer and requests that chatbots can’t fulfill. An effective chatbot will be able to recognize early on in a customer service interaction when it isn’t suited for the situation and transfer the person to a live agent in the appropriate department who is best suited to assist that customer.

A willingness to admit defeat

Along with the ability to delegate must be the willingness to admit defeat. Too many chatbots, when unsure what a customer is asking, get stuck in a cycle of “I don’t understand” or “please say that again” responses.” When a chatbot gets stuck, customers will grow frustrated quickly if there isn’t an option to get routed to a live agent.

The ability to explain what it can and can’t do

As mentioned above, chatbots can sometimes be indistinguishable from humans at least initially. This can cause people to overestimate a chatbot’s ability and will lead to frustration when it can’t do what they expect it to be able to do. An easy solution for this is to have your chatbot provide customers with a list of functions it can fulfill early on in the interaction so customers have some direction about what kinds of things they can ask of the chatbot and which things they might need to talk to a human about.

Access to CRM data

Every good customer service department needs an up-to-date customer relationship management (CRM) database. This is where information about customers such as their preferences or notes about ongoing matters can be recorded so that they are available to other agents who might assist that same customer. Surprisingly, many chatbots aren’t programmed with access to this CRM data. This is a huge missed opportunity because one of the biggest advantages chatbots have over humans is the ability to process vasts amounts of data quickly. There’s no reason your chatbot shouldn’t be able to use machine learning and data from prior interactions to provide customers with the best possible service that is tailored for them individually.