Bots should ‘talk’ exactly like humans, right? Wrong. They should sound human but always nudge the conversation towards a desired outcome. Just like movie characters do, said one leading conversation designer at the Chatbot Summit in Berlin.

What’s the most 21st century job you can think of? I challenge you to come up with something better than ‘conversation designer’.

It’s not exactly driving buses or baking cakes is it?

But in the digital age, in which people routinely talk to intelligent machines, someone has to put words into the bot’s ‘mouth’.

After all, as bots proliferate across our daily lives, we will become more and more aware of good and bad encounters. Those experiences won’t happen by accident.

Hence conversation designers.

Unsurprisingly, many AI companies are now working hard to make machines more helpful, natural and persuasive. They’re aware that software engineers have driven most of the advances to date. Now, they recognise that coding alone won’t cut it. Different skills are needed.

This is the view of Hans van Dam. He is the founder of Netherlands-based Robocopy, a specialist in bot conversation design, and has spent many years working on this challenge. At the Chatbot Summit & Voice Summit in Berlin last month, he outlined what he has learnt.

The big take-away is that bots should not speak exactly like humans. Rather, they should use a contrived form of dialogue that sounds authentic but which subtly moves people to a desired goal.

He says a good parallel is screenwriting.

“On TV and in the movies, nobody speaks the way humans usually do,” he says. “It seems natural, but it is polished dialogue, created with a purpose, brought to life by a strong character… Somebody wants to achieve something, and this somebody has to overcome certain obstacles to come to a resolution.”

Writing for the screen is an art. And so is writing for bots. One specific challenge that bots have to overcome is when humans/customers prevaricate or digress. Here, the first task is to ensure the bot is now bewildered. The second is to make sure it (politely and subtly) directs the customer back to the goal of the conversation.

Writing for the screen is an art. And so is writing for bots. One specific challenge that bots have to overcome is when humans/ customers prevaricate or digress. Here, the first task is to ensure the bot is now bewildered. The second is to make sure it directs the customer back to the goal of the conversation.”

On stage, van Dam gave a couple of examples.

Bot: ‘Do you want to pay now?’
Customer: ‘I’ll use MasterCard’.

Clearly this is not the yes/no answer the bot was looking for.

Another example.

Bot: ‘How many rooms are there in your house?’
Customer: ‘Let’s see. There’s the kitchen, three bedrooms, two bathrooms…’.

Again, not a direct answer to the question.

In the above cases, the bot has to re-direct conversations that are veering off course. However, conversation designers can be proactive too. On stage, van Dam talked showed how proven psychological techniques can nudge people to desired outcomes.

For example, they can double up the positive options in a multiple choice answer.

  1. Of course
  2. Yes
  3. Maybe
  4. No

They can also use the proven psychological technique of eliciting multiple ‘yes’ answers (which improves the chances of getting another). This is done by writing a question such as:

Are you ready for the next question?

…for which the answer is always yes.

While these techniques are important, van Dam says conversation designers should remember that bots can’t do everything. Instead, they should focus on the easy wins.

He notes that 80 per cent of customers enquire about the same small family of queries. Bot designers should focus on these cases, and let humans agents handle the small number of more complex scenarios.

Tim Green

Features Editor, MEF Minute

  

Member Login

Subscribe to our mailing list

* indicates required