How Human Does Your Bot Need To Be? Science Says Not Much!

human, robot, chatbot

Recent research suggests that going for the all-singing, trying-to-be-friendly, almost-human bot isn’t the best way to go for most use cases. So, just how much banter or empathy should your bot show? That’s a key design decision to make when building chatbots.

Bots can express themselves in three basic ways, They can deliver the basic information that any user needs to know. They can emulate the nature of the brand (be that as a chatty character with a sense of fun or gravitas as required) or they can feign interest and emotion in the user’s wellbeing, their day or other aspects to broaden the interaction.

When deciding to build a bot, your team might think that sticking to the basics is the best way forward, at least for a first bot. And that instinct is backed up by the U.S. Media Effects Research Laboratory that works alongside Pennsylvania State University’s Institute for Cyber Science.

As part of “Computers in Human Behaviour” research, the upshot of their work was that test subjects preferred it when the bot stuck to a common range of expected replies, rather than taking up their time with preamble and attempts to engage outside the core topic. Medical Daily has a detailed summary of the test.

Building Bots for Speed or Smarts

As with any business and technology decision, the level of detail or engagement you want a business chatbot to go into is very much up to the design team, tempered by budget and time constraints. Chatbots can be both smart and fast, but the key element around these decisions is a chatbot is supposed to save your customer time, so how much of that do you want to take up being chatty?

With digital businesses measuring engagements and web browsing in seconds, that time is very precious both to the customer, and the business – especially when it comes with a risk of losing a customer the longer a chat goes on.

Yet, every customer is different and some markets, think consumer magazine or online bingo chatbots, might benefit from chats that boost stickiness. Similarly, every business and technology leader has their quirks, perhaps considered bots as tools rather than collaborators. From those who still write everything down on paper to those who live their whole lives in Slack, via Alexa or on Messenger, their leaning to tradition or simplicity vs modernity might sway any debate as to how smart the bot needs to be.

Therefore, when planning a bot, the designers need to factor in the playtime they expect with their audience, their demographic and technology-awareness. Then, the result or message they are trying to get across (service chatbots will differ from marketing chatbots and so on) and other factors.

Coming up with a resulting scale or ratio of empathy/humanising to fact-based content should allow any business to work out how far beyond “hello, how are you?” the chatbot should be in terms of friendliness. We’re not aware of a magic equation that can provide a golden number, although we suspect that AI experts are working hard on one to do that job.

Ultimately, the business or service should know its customers well enough to estimate how much people are prepared to put up with. A bot handling trade accounts or rapid customer service interactions like pizza or cinema ticket orders will likely be light on chat. But a college student wellbeing bot or a popular weather app chatbot might need more of a persona to drive and extend the engagement.

We expect further research and different answers as bots become an everyday interaction for a growing percentage of the population. The key takeaway for any business or developer, that perhaps has a small but growing audience, is to start doing the basics well, and then add chat and further interactions over time, monitoring the metrics to see if they drive or curtail engagement.

Whatever the results, don’t be afraid to share them, as many in business and the tech industry are keen to see how much leeway there is in users’ goodwill toward bots and are happy to highlight any breakout examples of beyond-the-norm interaction done well.