Good news: the chatbots scam can be avoided!

by bold-lichterman

A year ago, in these columns, I had awakened (and struck) some consciences with a platform in the form of a rant on the chatbots scam. But yes, you know, those little chatbots who’ve been in the business of marketing and customer support for a few years. It must be said that the promises are beautiful and that specialists are flourishing, self-proclaimed disruptors of chatbots, or rather “conversational experiences based on Artificial Intelligence”. It’s more glamorous and I denounced it in the previous forum, I will not come back to it here.

No, what interests me today is rather to identify good practices observed in the field in contact with large groups supported for several years on the subject. Proven methods and tips that are not necessarily in the small manual of the perfect marketer, but which have been proven. Because yes, the chatbot is not just a gadget designed to generate a bit of hype by being “the first”, it can be of real service as long as we try to tackle the subject from the right end.

What is a working chatbot?

The question is quite complex, it must be admitted. We all have biases when we test experiments like this. However, there is a pragmatic version of this definition of efficiency, the user simply needs to spend time in the conversation to fulfill the objectives that the designer has set, without falling into an infinite loop or being lost and blocked. . You must therefore find the right mix between the actions you want the user to do and the real fluidity of the conversation proposed.

The sources of frustration are numerous and, fortunately, we can avoid a lot of them with a few basic ingredients that are ultimately very simple.

For example, it seems essential to allow the user to get out of the script defined beforehand. By including the equivalent of clickable buttons if text is typed. It sounds obvious, but if you’ve tested a few chatbot experiences you will understand what I’m talking about. Also remember to plan for simple cases such as understanding “yes / no”, “thank you”, “hello” and their usual variations. Nothing rocket science, no strong artificial intelligence, just common sense to bring a little fluidity to the conversation. In short, make sure that the expected service is real and rendered, without interference from simple technical problems.

Unfortunately, these are not isolated cases that I take here, it concerns far too many projects, led by startups as by major players in social media marketing.

Between artificial intelligence and very human stupidity

For this there is no secret or revolution compared to a classic web project, you have to spend time designing and iterating. The conversational interface involves new aspects, that’s obvious and we’ll come back to it, but we must not forget the basics! Artificial intelligence should only be there as a backup, it is especially not the solution to all problems.

Take the example of chatbot solutions that serve as dynamic Frequently Asked Questions (FAQs). Typically these approaches are not able to provide real conversational experiences. They can, more or less well, answer a specific question. It’s a good start, but it’s nowhere near enough to deliver something fluid and complete. All the more so when the answers proposed by the robot are of the form “here is what I found, does that suit you? yes, no, maybe, click to be sure and to calibrate me please ”.

No, today it is impossible (and totally illusory) to want to let an algorithm, however intelligent it be, manage a complete conversation. Facebook and its billions of dollars can’t do it either. I invite you to browse a research paper from the Facebook lab on the subject of conversation simulation. We are making progress, but today’s industrial solutions are still very far from it.

By dint of absolutely wanting to use artificial intelligence, and a fortiori mathematics applied to language, we are on the wrong track and we are moving away from the real needs of users.

The main resource of a successful chatbot? The real people who are behind the machines!

I hope not to surprise anyone. A successful chatbot is not a machine that passes the Turing test, but a combination of human design to guide and support with a dose of artificial intelligence to optimize, go further, and personalize.

But the basis is human intelligence during the design phase. It is not a question of sticking to a basic decision tree like a good old telephone answering machine “for service X type 1, for service Z type 2” as we see it all too often, especially on Facebook Messenger , with the ease of buttons to click. We must go further by analyzing all the paths or situations that may arise so as to never be caught off guard.

Then we can add language recognition bricks to manage even more cases, based on a history of conversations with humans for example. Or by iterating by having it tested throughout the design process by “normal” users outside the project. In our approach, models are conventionally trained to detect loops, detect aggressiveness to pass the baton to a human operator … these kinds of little things so important for the fluidity of the conversation experience.

More or less advanced technical bricks, sometimes good old multi-criteria search scripts are enough, will also be able to allow us to access more directly to a given branch according to a profile.

Do we really need artificial intelligence for that? In many cases I am not sure. Once again, it is a question of thinking carefully about the conversation scenarios upstream, even if it means testing the refinement over the test phases or in reduced deployment.

Forget the “fun and quirky” chatbots

After a conference which gave a few examples of failures and successes in the conception-writing of conversations with bots of major French brands, I was surprised that I was asked a question that seemed very important to its author: “so, do you or you have to say when you’re a bot? ”.

The question is relevant, of course, but far too simplistic. Because editorial work can allow us to go much further. A real effort on the drafting makes it possible for example to guide the users towards the options that the robot understands, while trying to avoid the “exotic” entries which could make it ramble. This is all the more important when designing experiences for voice, which does not have clickable buttons like on Facebook Messenger, and for which the user has very limited attention. Don’t overlook this part! In my various tests, I have already seen chatbots that were rather well designed technically and crashed because they were poorly written (and I’m not talking about spelling mistakes, which are also legion, sometimes).

So yes, the chatbot is an experience that is meant to be conversational and therefore more fun and closer to the end user, but this is no reason to talk to him like a friend crossed in the street. This is a very big mistake that many still make. The work of experienced and conversational writers is a plus, but as with design, real-world testing is the only valid fire test.

There are many very interesting players on the subject in France and abroad, very specialized in each of these areas, they are unfortunately often not the most visible or the most communicative. They are the ones who do, and who put their work back on their keyboard without complaining and without pouring into the abusive press release. Take the time to identify them.

Last year I warned against the chatbots scam, the message is still relevant and today I would not change anything in the forum in question. But it’s high time to roll up our sleeves and take pleasure in creating truly beautiful conversational experiences for users, with the right mix of techno and common sense.

The contributor:

Artificial intelligence from dystopian fantasy to reality on the groundThomas gouritin supports SMEs and large accounts in their transformations, with digital support. Producer of the Regards Connected series (Youtube channel and podcasts), he explores our technological future to popularize complex subjects such as artificial intelligence and convey messages of pragmatism to be applied in business. The subject of chatbots is essential today, Thomas approaches it in a pragmatic way with, in addition to project support, conferences aimed at demystifying the subject without “bullshit” and with workshops allowing everyone to get their hands dirty. design to understand, learn, and do.