Echo Dot (3rd Gen) - Smart speaker with Alexa - Charcoal

Use your voice to play a song, artist, or genre through Amazon Music, Apple Music, Spotify, Pandora, and others. With compatible Echo devices in different rooms, you can fill your whole home with music.

Buy Now

Wireless Rechargeable Battery Powered WiFi Camera.

Wireless Rechargeable Battery Powered WiFi Camera is home security camera system lets you listen in and talk back through the built in speaker and microphone that work directly through your iPhone or Android Mic.

Buy Now

Would you use your own chatbot?


Felipe Magalhães Bonel
“I’m sorry, I wasn’t able to understand your message. Mind saying that again with different words?”

Would you use your own chatbot?

If the answer to this question is a daunting “no”, then maybe the time has come for you to rethink some important matters.

Why am I saying this? Well, because there is nothing more harmful to the implementation of a new interface or communication channel than a poorly designed user experience. And since that, unfortunately, is becoming more of a rule than an exeption, it is really important that we discuss usability-centered approaches to not end up holding hands with certified agnostics who keep insisting in punchlines such as “well, chatbots don’t work at all” or “virtual assistants aren’t getting there because this or that algorithm isn’t sharp enough.”

Nonsense! Here, between us, now: I have never seen, during my carreer as a product/UX person, a single human being blaming mobile for the gargantuanic amount of bad apps on the market, nor have I seen anybody having the courage to say “oh, this graphic interface hype won’t last much longer”. No. We should be thinking of ways to overcome this hype on the expectations created upon new technologies and get to understand what are the fundamental questions to be asked before putting effort into creating a potential frustration generator. I am serious — between doing it lazily and not doing at all, sometimes the second choice is a wiser option.

First things first: this might be the most complex question to ask yourself and, therefore, it is certainly the most important one. From a practical and pragmatic point of view, your virtual assistant is able to help somebody acheive any objective? If yes, move on. If the point is “oh, but I want one!”, hold your horses, buddy.

“B-b-but Gartner said that, by 2020, 85% of companies will be using automated interfaces to inteeract with their customers! I need to have a chatbot.”

Aaaaaall right. 2020 is in less than four months, and I really would like to think that 85% of customer service around the world could be automated, conversational, massive-one-to-one by then. Since we’re quite distant from that acheiving that KPI, what we have on the real world is a massive amount of purposeless bots running around messenger apps. These guys end up being more a source of constant headaches for the users than a cluster of efficient solution providers, unfortunately.

I guarantee: the amount of people potentially interested in talking to an institutional chatbot in order to ask the year that your enterprise was estabilished, who are the key employees and what is the market niche in which you operate should be close to zero. Thad said, it is highly recommended that you drill down deep in your business database to understand three basic things:

  • What are my customers’ main pain points?
  • Can these pain points be addressed with a conversational flow?
  • Do I have systemic conditions (databases, microservices, APIs) to enrich these flows with relevant information?

Affirmative answers to the questions above, if neatly organized, can be great material for organizing a consistent implementation roadmap, centering on your user pains and on guaranteeing true ROI for your business.

This is a marvelous gain-gain case, as it ensures previsibility and efficiency on the implementation and, most importantly, good nights of sleep with clear conscience that no consumer is calling you names because the chatbot told him to install your mobile app on his cell phone to gain access to a transactional interaction that you prioritized on the conversational channel without being able to do so.

With the above-mentioned roadmap (or at least a first sketch) in hands, a praiseworthy next step should ensure that the approach to the conversational flow design will be suitable for your main objective. Much of the process of dialog writing involves driving the user toward solving their problems, which demands some sharp predictive skills. In this scenario, nothing more reasonable than to formulate use hypotheses based on the maximum information at hand about your users’ behaviours, how they are satisfied with your customer service and, most importantly, all of the efficiency and retention metrics that you can gather.

“Researching user behaviour is for pussies, here we have a foolproof feeling”.

Go ahead, then, but you are utterly bound to fail at it. Lots of good chances of enchanting a customer with a memorable experience are lost due to assistants being unprepared to handle common situations at best. Others might lose their hand on being insistent on a specific matter without provisioning a scape valve for the user.

Why bother sending 5438 notifications via Messenger or via WhatsApp if your end-user doesn’t have a single reason to talk to you? Why train your NLP to receive praisal when your bot’s main function is to handle angry customers with financial problems?

It’s all about context and purpose, and you have to be prepared for it.

Otherwise, you’ll just end up like me, with a bot answering “I also like you ❤️” to a user that had just typed in “I give up on talking to you”. These interactions, although ironically comic, never end nicely.

On the other hand, as long as you are fully aware where you are getting into, you will be able to design flows capable of rocket up all that planned ROI. It’s a simple merchant principle: A happy and satisfied customer will always come back.

I was once an extremist on the necessity of overtraining sharp-edge NLP everywhere to ensure that comprehension wasn’t a problem. I’m still like that [author note: in fact, none of the bots that I design go live without a hardcore dataset of natural language and proper conversational design], but it is worth spending some minutes to think if some choices aren’t just overkill.

“But Felipe, CIO Magazine said that if I don’t buy my own ‘Watson’, I won’t be innovative!”

Ok, let’s try a different approach to this question: supposing that your office is three blocks away from your home and your children study at a school just down your street, would you make priority investments in a car, or else? Probably not, considering maintenance costs, fuel, traffic tickets and extra care to safeguard your vehicle’s health.

When it comes to Artificial Intelligence, we could apply the same logic. You might want to have a top notch natural language processing engine, but if you don’t have pacience (or anyone available) to monitor accuracy rates, harvest feedbacks and provide proper training, why complicate?

Sometimes, it is better to stick to your IVR or set up predefined answers in a closed flow than losing 90% of your users because your voicebots behaves like a véia surda* due to out of tune speech-to-text, or because your virtual assistant isn’t able to disambiguate “recharge balances” and “internet franchise” just because your user says “my internet balances are over”. Right? These things might look like soft details, but can completely change the meaning of a user’s journey in using a cognitive interface.

[*= author note: “véia surda”, translatable as “deaf old lady” is a fictional character from a popular comedy show in Brazil, “A Praça É Nossa”. In her particular sketch, humor consisted in making puns based on low-quality mishearings like confusing “I thought you were mad then” with “I thought you were Matt Damon”]

On the other hand, having at hand a continuous improvement framework is a buletproof way towards excellence. I take as a premium example the case of Telefonica-Vivo in Brazil, which created a “bots school” at their Cognitive Call Centre. If their language processing accuracy today hits 86% in a compendium of millions of interactions a day considering a complex NLP+STT+telephony setup, it is due to the fact that there are a lot of human beings, from quality overseers to data scientists, working a lot so that everything turns out fine.

Stay cool: you won’t need a specialist sector on your company with 18 people fully dedicated to train your bot. But keep in mind that it takes a lot of hard work.

There’s no magic trick in here.

Rather than being skeptical or evangelistic about conversational interfaces, the main point I’m trying to make is: be loyal to your target audience. Don’t be afraid to ask all the questions necessary to ensure that your product is able to fill its purpose, enchant your users and be able to stay on its own feet with a reliable and simple user interface.

Ultimately, empathys still figures out as the best tool for this moment of decision taking. If you, taking place of your user, isn’t able to get to the other side of your flows with satisfactory results, it is because something is going terribly wrong. When in doubt, reach out to a Product or UX person for a more fresh, user-oriented view. Remember: Although you are not your user, being able to empathise and put yourself in their place is very important.

After all, what’s worse than a first-time user being frustrated? Some first impressions can be extremely hard to revert.

Read More


Please enter your comment!
Please enter your name here