Lessons learned organizing the largest Machine Learning Meetup group in Latin America

This article has been migrated to our new blog. Find it at https://building.nubank.com.br/organizing-the-largest-machine-learning-meetup-group-in-latin-america/

As a data-driven company, Nubank has worked hard across the years to build a strong analytical team. Data scientists are one of the cornerstones of the company and usually responsible for everything related to modeling and sometimes policies.

As the field of Machine Learning/Data Science is too big to know-it-all, we have always encouraged our team members to attend events where they could learn from other people outside the company. However, from the very early days, we felt the need of doing more.

That’s why we were interested in building a community and inviting people from both the industry and academia to exchange ideas with a broader audience. Our goal was to learn and also give back some of what we have experienced as well — and the best way to do it seemed to be with a Meetup, a simple, ready-to-use platform that allows users to easily create and manage group events.

We have been organizing a Machine Learning Meetup in São Paulo since 2015. It has grown to become the largest Machine Learning Meetup group in Latin America, with over 7,000 people signing-up on our page. Over 2,500 people and 50 speakers have attended more than 20 meetings organized so far. These are some very exciting numbers, but it took us a while to get here.

Oftentimes, people will organize Meetups thinking that they will be a huge hit, that speakers will knock at your door every day and that a large audience will show up every time. Unfortunately, as we would find out, things don't work quite that way.

Since 2015, we have faced many challenges — and we wanted to share some of them to help future organizers.

So, what have we learned organizing the Machine Learning Meetup?

1. Finding a speaker is hard

• When you reach out to people and ask them to give a presentation at an event, they usually ask questions. Especially if it is for a free event. It took us some time before we could naturally attract speakers. Nubank’s brand really helped, as people in São Paulo know the company;
• Making a presentation takes time, making a good presentation takes even more! Even experienced speakers may feel discouraged to put enough effort into making one. So inviting them in advance is a good idea.
• Your location affects your acceptance rate. Machine Learning is still a growing field in Brazil and we are usually far from where the action is happening. So finding experienced people with good content to share is hard.

2. Don’t underestimate the importance of a good venue

• One feedback that we frequently got was how awesome it is to have a lot of space for people to sit, while also having room for them to network and move freely around. Finding a place to comfortably fit everyone is very important (we were lucky to use the amazing Day One room at Nubank);

The Day One room at Nubank, where we hold our Machine Learning Meetups

• But, we also found that holding a meetup at the same place every month limits the pool of people that can attend. Especially in big cities, like São Paulo. We are fixing that by trying to reach other companies to have them host our meetings once in a while.
• We try to organize events in different cities whenever one of us is traveling to a location that has a potential audience. Our 11th Meetup was in Brasilia, for example;

3. Hungry people don’t network

• People usually come to Meetups for multiple reasons:
  — the obvious one is to see the talk;
  — the other main reason is to be able to talk directly to the speaker or to other participants after the presentation. Networking is also one of the main reasons why we are holding these meetups — so that our data scientists have the opportunity to exchange ideas on what is happening in the field.
• We found that it is important to sponsor the pizza at the end of the event. This allows people to stay a bit and network — and it is also a thoughtful gesture, since meetups usually happen at the end of the day and most participants come straight from work. Hungry people don’t stay or want to network!
• Tip: remember to order for people with special dietary requirements, such as vegetarians.

4. Be mindful of the type of talks you want to accept

• Say “no” to company pitches. Some people do presentations mainly focused on a product, so you need to check the content in advance to guarantee it meets the audience’s expectations. You always want to focus on your community and not on a company’s interests;
• Offer your audience some variety. We are really proud that researchers, students, data scientists from industry, open-source contributors, nubankers and people from other countries have spoken at our meetups. The goal was to have technical and more introductory presentations in a wide range of topics — such as Ethics in Data Science, Open Data, Kaggle and Deep Neural Networks. (All presentations can be access here)
• We usually prefer long talks where people have around 1 hour to do their presentation. It gives them time to go deep into a topic instead of just presenting an introduction.
• From time to time, we try to hold Meetups with specific themes in order to reach new audiences — like the ones we did on education or open government data. People not involved in data science come to see how to apply Machine Learning to their subjects of interest;

5. You need effort — not a huge budget to make it work

• Our Meetups started small and were scheduled once in a while, but the issue with lack of periodicity is that it’s hard to build a real community — even though we already had over 1,300 people at the Meetup page at the time. On January 2016, we started to focus more and forced ourselves to have at least one gathering per month and to put more effort into reaching out to people
• We have sustained a steady growth without spending money with advertisement — nor investing time in social media (though this is something we would like to do). So far, our experience shows that word-of-mouth really works, which is why investing in content is crucial.
• However, the low commitment is a downside of a free event. We found that, even if your community is engaged, you can expect that nearly 40% of the people who sign-up won’t come.
• We usually display the event one or 2 weeks in advance but only open it for sign-ups 3 or 4 days before. If you open RSVP (sign-up for the event) too soon, people will sign up but many won't be able to come. And even if you send them a confirmation email a couple of days before the event, most won't cancel and still won't show up.

6. It is not easy, but it's worth it!

• It’s always interesting to build a community around a purpose. Organizing this Meetup allowed us to involve more people in Machine Learning and have them be interested in the subject.
• The Meetups are helping us engage more people at Nubank with our work. We are advertising them internally to have nubankers better understand what is Machine Learning and how it can help them in their areas.
• It’s a great way to get visibility in the market and reach candidates. Three people in our current Data Science team were found thanks to the Meetups;
• The field of Machine Learning is HUGE: it’s impossible to know and learn everything. A Meetup is a good way to invite experts and to be able to ask them questions directly, to exchange ideas, and to learn from other's challenges and mistakes so you don’t have to take the same path.

If you are interested in presenting or partnering with us, please reach out to the organizer on our Meetup page.

And if you want to be part of this great adventure as a Data Scientist, a Machine Learning Engineer (or any other position) at Nubank, apply here.