Leave the persona for last

In any kind of organisation, public or private, you'll face resource constraints. We can dream of having a team that's big enough so we'd never have to choose what to prioritise, but reality is often different. This means you need to have good reasons to spend resources on any particular task!

In this article, we'll take a closer look at how to prioritise the task of adding human-like elements to your VA - things like creating a name and personality -  based on the work of Makany and colleagues.

The imitation game

Ever since computers were created, people have tried to make them behave like us. "What if this computer could trick us into believing it's human?" became such a big question that we're even making official tests for it!

In the 50s, Alan Turing created a test where a human has to decide whether they're talking to a computer or another human being. He called it the imitation game, but today you're more likely to know it as the Turing Test.

Of course, it is an intriguing question - all the more so now that Generative AI is really stepping up its game. But making a VA behave in a human like way is not like flipping a switch! Even with LLMs it requires time to write a prompt or conversation that conveys human likeness. With all other important tasks in a VA project, is it really necessary for us to spend resources making the VA imitate a human?

Match your users' expectations

To answer that question, you have to know what your users expect. With that in mind, Makany and colleagues interviewed 19 people - 10 of whom had used a VA from a small and medium enterprise (SME), 4 SME owners, and 5 VA developers.

From these interviews, they distilled six themes, each of which had different kind of expectations from each participant group (user, owner, developer).

I want to highlight their main finding: beyond politeness, the users do not really care about human-like features (name, gender and personality). To them, it is much more important that the VA understands and can help them, and does so efficiently - a finding in line with other work that we've done together with SINTEF, which you can check out below.

Prioritise functionality over personality

Let's be very clear: it is not bad to add human likeness to your VA! But if you have limited resources, don't start by spending them on deciding which personality traits your VA is supposed to have. Instead, they're much better spent ensuring that you can answer your users.

In other words: prioritise tasks like improving model accuracy, scoping knowledge coverage and reviewing conversations over creating a personality for your VA. Once you reach a point where, for example, your unknowns are at an acceptable rate, you can be more comfortable spending resources on adding human like features.

Perhaps the easiest way to remember this is by thinking of personality as the cherry on top. Your VA will work without, but can become even better with.

Of course, when you get to the stage where you're thinking about personality design, we've got you covered. Check out the article below to start thinking of personality in a structured, and goal-oriented way to help you design your VA consistently.