Oi Chatbots + IBM Watson.
Rodrigo Esch
UX Lead at Oi
Made with

Oi Chatbots + IBM Watson.

Helping users achieve what they need using conversational interfaces.


Since Watson won Jeopardy TV show back in 2011, the hype about IBM's artificial intelligence only increases year after year. Despite all the exaggeration around it, conversational interfaces are here to stay.

At Oi, we face challenges regarding customer support savings, helping reducing calls and customer pains creating new experiences for people manage their plans and seek for support. Providing points of contact near as possible to the users everydays lifes in a way that they can do things by themselves, is not only a important business strategy but also a user demand. Messaging apps are used by people every minute, every day. It makes sense to automate process building easy and fluid conversational interfaces.

My Role

As the UX Lead, I was responsible to provide a clear vision of the chatbot strategy, defining the MVP along with the product owner, and make sure the chatbot delivers a fluid conversation helping users complete their tasks.

We develop a good practices guideline, mapping componentes to different, messaging platforms, defining voice tone and personality and testing the bot constantly to gather insights.

Our Vision

In partnership with business areas, our strategy was to put the most used feature on Minha Oi in a chatbot as the MVP, in 3 different channels: Oi website, Facebook/Messenger, and later Whatsapp (the last one because is the most used message app in Brazil, and Oi was chosen by them as a successful case of agile methodology, letting us build our chatbot on the platform first than other telecom players).

We needed to be very clear to users that this was in fact a robot, not a human. And when they need to talk to a human, the chatbot must leave the way.

The MVP brought the most used functionality in Minha Oi nowadays: payment with bar code. Users would be able to use the chatbot easily to get their code to make payments monthly, and even receive a notice by mail when their bill is available again. Quick and easy, without the need to download Minha Oi app (a well known user pain point).

Later on the roadmap, technical issues will be treated in chatbot as well, when we will need to reinterpretate the Técnico Virtual user flows to adapt them to a conversational interface.


Oi Chatbots + IBM Watson.

User Flow: MVP and Exit Points

We adapt the user flow to pay a bill on Minha Oi to a conversational interface. We also design how we would treat other subjects not covered by the chatbot on the MVP, the called Points of Exit.

It is important to notice that in a interface like these, made with natural language processing (NLP) there is no linear flow. The user can talk about anything anytime. We as a UX team needed to prepare for all of these possible scenarios.

Oi Chatbots + IBM Watson.

Voice Tone and Personality

Oi has well defined brand guidelines and some of its attributes match the personality we thought the chatbot should have to really be helpful and approachable by users. We definitely didn't want to make the chatbot sound like a human, but that shouldn't make it a cold and distant cyborg.

We tried to make the bot's speech honest, friendly and direct, setting the right expectations to the users. In our MVP, the bot definitely would not be able to talk to any subject so we communicate that right away to the users. We did on our landing page a fun illustrated roadmap as we represent the bot as a baby who is just born and soon in its next step becoming a youngster who is studying and already knows how to talk about bills. Then, he will be bigger, and much more knowledgeable.

Oi Chatbots + IBM Watson.

Defining Interface Components

This is not a conventional interface design, and the cases to be based on today are still very limited when it comes to chatbots UI documentation, so we started to build a document with all the componentes, divided by message platform telling exactly when use each element. It helps the developers a lot.

Oi Chatbots + IBM Watson.

Copy and Curatorship

One of the biggest challenges of any project like this is the curatorship. 

This is heavy work, since the artificial intelligence needs UX Writers inputs to build its knowledge base. Watson works with standards like Intents and Entities. They're groups of phrases, terms and words that needs to be classified and associated with conditional responses. The image on the left show a spreadsheet used to create the interactions.

Oi Chatbots + IBM Watson.

Surveys and Tests: How We Refine

The very first launch of our chatbot was in a controlled environment where only some selected users were invited to test. This was something really important because telecom companies in Brazil get backlash comments a lot in every possible social media, and we needed to see how the chatbot would perform interacting with real users.

We send to them a small survey, to get first impressions. This help us a lot to identify gaps and make improvements.

Here are the points of contact where the Bot can be found in Oi's ecosystem. First we release the bot on the Minha Oi pages, and then Facebook.

Minha Oi / Landing Page

Oi Chatbots + IBM Watson.

Facebook / Messenger

Oi Chatbots + IBM Watson.


Minimizing Whatsapp constraints:

Whatsapp now is officially allowing companies build chatbots in the platform. One of the main barriers to a great bot experience in it is the absence of rich components like buttons and cards. They are very helpful to put users in the right path, like shortcuts, but in Whatsapp you are restricted to only links, images and maybe a bold and italic text style.

We tried to build a logical interface with these little pieces showing the users call to actions throught bold text and some emojis. We also start to use numbers are indicators to possible options in a menu.

Oi Chatbots + IBM Watson.

Next Steps

Soon, the Técnico Virtual flow will be released in all channels. We are planning to release more functionalities in all platforms.

Our hypothesis is that Whatsapp will be a hit, much more than Facebook and Minha Oi. We will be following the metrics to validate that impression.

What Key Learnings and Takeways

Some things me and the team learned during the process:

  • As a said before, the user can talk about anything anytime. We as a UX team needed to prepare our bot for all scenarios;
  • The way of documenting and thinking about user flows are very different than doing it when building a screen to screen interface;
  • Good copy is key on the project. The UX Writer must be at the top of the game, refining all the time the conversation based on users feedbacks;
  • IBM Watson is an AI powerhouse, but it takes time and effort to make a chatbot really good;
  • Analyse the data on a daily basis is really important, but testing with users "in loco"made us realise that in every conversational interface the user behaviour is incredible different. For example, we had to make changes in components and copy in Messenger in a way and Whatsapp in another way.

Impact in the week of launching.

Efficiency rate in get payment code. Beats the Minha Oi average in 10%.
People subscribing to online billing (more than Minha Oi average, the official channel).

Dear Rodrigo Esch,