The novelty factor of bots and artificial intelligence (AI) is undeniable, and their functionalities are equally as impressive. From turning on the living room lights and playing your current favorite song to ordering household essentials with a simple voice command, bots can be helpful around the house. But what happens when you have a request beyond the bot’s reach?

Using AI and bots for customer service has become something of an art form, evolving into more of a strategy than an automated solution. Not only are bots limited in terms of capabilities, but automated responses don’t create bonds with customers–people do. For an industry so concerned with care, why would a robot replace the human touch? Instead, AI and bots can be used to empower human-to-human exchanges and make them more satisfying.

Let’s define the four types of interactions.

1. Informational
These are basic questions like, “what’s the weather in New York?”, “what’s the traffic report?”, “is my train on time”, “where is my order”, etc.  Bots can handle 100% of these queries. And the cost of incorrectly answering is generally pretty low: you get to the train station a bit too early.

2. Instructional
These refer to a bot’s ability to obey a command, such as “play a song”, “call Tom”, “send a text to Mom”, etc. Bots can also handle 100% of these interactions.  Again, the costs of failure are pretty low: the wrong song plays, the phone calls the wrong person, etc.

3. Transactional
These interactions are actually causing something material to change in the world. “Reorder pizza from Luigi’s”, “send flowers to Susan”, “request a Lyft”. We’d estimate that bots can handle approximately 80% of these types of requests. Now, though costs of failure are not low when it’s about mistaken orders, incorrection shipments or missed opportunities to upgrade airplane seats.

4. Customer service
These are interactions with a customer where the customer is trying to understand or to change a situation (order, service, delivery or similar).  Critically, many customer service queries stem from a issue, meaning there is already some level of frustration on the customer’s end.  Putting a fairly, but not completely, accurate bot in front of a frustrated customer is not a recipe for success.  We’d estimate that a bot is helpful in about 40% of interactions but the bot is only really useful for the early, information-gathering portion of the interaction such as, “what is your address?”.

Alexa is well-suited for handling informational and instructional interactions, but has proven to be confusable in transactional interactions. For example, Alexa followed a little girl’s command to order a $170 dollhouse, and Echos and Alexas have picked up commands heard from televisions and placed orders. Reversing those orders actually created an uptick in customer service issues, handled by real humans. As AI gets more sophisticated, its capabilities will certainly expand, but if brands want to maintain and nurture relationships with consumers, an army of bots won’t be replacing customer service representatives anytime soon.