Bot vs spouse: How much do customers talk to chatbots?

By 2020 the average person will have more conversations with bots than with their spouse.¹

How would you go about making such a prediction? To start, defining these terms and measuring them is not a simple task.

First, we should start by defining what a “bot” is. The term “bot” can mean a lot of things, but here we should only talk about bots that you can have a conversation with.

Second, what is a conversation? Is it a time-boxed, linguistic interaction? Does it need to be spoken language, or can it be text-based? How many turns are needed to count as a conversation – i.e. does a simple question and answer count as a two-turn conversation? And should we measure based on the total amount of time spent in conversations, or should it be based on the total number of these time-boxed interactions?

Third, let’s define a time frame for these conversations – let’s limit it to one day.

If we define a conversation as a back-and-forth over SMS or IM, then maybe asking a bot a question or requesting it to do something for us would count. In this case, typing a search into Cortana or Google might count as a conversation. And in this case, the average person probably already does have more of these simple conversations with bots than their spouse in a day.

If conversations must involve more than two turns, a simple question-and-answer pair does not count, and simply counting the number of conversations might not be enough, so we’ll want to look at time spent. In the best-case scenario for bot usefulness, we would assume that these bots will become as useful as smartphones. Various studies have shown people use smartphones three to five hours per day.² I wasn’t able to find a good statistic for how long couples spend talking to each other on average per day, so let’s assume working couples probably spend at most two to three hours together per day, and let’s assume 20-30 minutes of that is spent in conversation. If bots become as useful as smartphones, they could easily surpass 20-30 minutes.

However, today, when we are on our phones, most of our activity is passive. Sure, we spend plenty of time actively texting and commenting, but most of our time is spent passively consuming media—reading social media, reading articles, watching videos, etc.

The real question here is how much time are we spending asking questions, or trying to get the kinds of things done that would require a conversation? In other words, what if bots understood natural language as well as humans? In that case, we might replace all of our business interactions with bots (e.g. calls to customer service, making reservations, etc.).

But, on average, we probably don’t spend a lot of time talking to people to get business done – how much could we really get done talking to a bot? For example, could we replace all of our time spent scheduling meetings by talking to bots (e.g. like Amy from Or could we talk to a personal shopper bot that would save us the time of reading reviews and comparison shopping online? These kinds of conversation with bots would arguable save time, but tasks like these would add up.

Alas, bots are still quite far from understanding natural language as well as humans, and they are quite far from being able to do complex tasks like comparison shopping and reading reviews.

In summary, it depends on your definitions of “bot” and “conversation.” Under some definitions, the average person is probably already talking to bots more than their spouse. Under other definitions, we have a long way to go before we get there.


Have a conversation with our bot!

See what talking to a bot is like. Have a conversation with Nina.

Engage us

About Paul Tepper

Paul Tepper is the Worldwide Head of Nuance’s Cognitive Innovation Group (CIG). The Cognitive Innovation Group is focused on applying the latest advancements in machine learning and artificial intelligence to automate and improve the customer experience across channels. Paul is responsible for setting Nuance’s AI Strategy and leading product development efforts in collaboration with Nuance’s Definitional Customers and Nuance Research. Currently Paul is focused on machine learning advancements in conversational AI, machine learning, natural language understanding, question answering and dialog modeling. Paul has over a decade of experience in software development and AI research. He holds a Ph.D. in Computer Science & Communication Studies from Northwestern University, an MSc in AI & NLP from the University of Edinburgh and a BA in Computer Science, Linguistics and Cognitive Science from Rutgers University).