SpeakOn: the app that hit #1 on Mexico's App Store and why it died

What SpeakOn was

SpeakOn was a messaging app like WhatsApp, but without the language barrier. You wrote in your language and the other person received the message in theirs, translated instantly.

It worked in more than 60 languages. Someone in Mexico could write in Spanish to someone in Japan, and that person read it in Japanese. The idea was simple: language should stop being a reason not to talk to someone.

How it reached #1 on the App Store

SpeakOn grew faster than anything I've ever built. It got 200,000 users in the first 48 hours and reached 350,000 sign-ups in total.

In June 2015 it was the number 1 free app on Mexico's App Store. In the overall ranking it was above WhatsApp, Messenger, Facebook, YouTube and Spotify. I still have the screenshot I posted that day, June 13, 2015.

The launch was done the way people did it back then: a launch party, money in media, and lots of celebrities talking about the app. Since it was free, there was zero barrier to downloading it. It went viral in the press and on social media at the same time, and that's how we hit those numbers.

Why SpeakOn died

SpeakOn died because of its tech, not for lack of users. I didn't know anything about code or software architecture, so I hired a team to build it: two backend developers, one on Android and one on iOS. The result was an app so badly built that it couldn't handle the load when it blew up.

We had a download counter in the office. I watched the numbers climb and, at the same time, watched people try to sign up while the backend went down because it couldn't take the load. No matter what we did, we couldn't stabilize it.

The app was built in Go (Golang). I brought in the best Go engineers from the United States, and the first thing they told me stuck with me forever: building a system is like building a house. If the architecture isn't ready for it, anything you add later is very hard. If you build a house and then want a pool on the third floor, and that wasn't planned from the start, there's no easy way to put it there.

Fixing it wasn't a matter of patches. It basically meant rebuilding it from scratch: the engineers estimated eight months of reconstruction, and we decided not to do it. Years later we released a version 2, but it never worked. The wave had passed.

The crash also cost me Y Combinator. I applied with SpeakOn and we made it to the first interview, but the app went down and we couldn't fix it, so they didn't accept us. That's when I started connecting with people from YC, and years later I ended up investing in 24 of their companies.

Hitting #1 with a product that can't handle its own users was the most expensive lesson I've ever paid for.

What I learned from SpeakOn

1. If you don't understand your product, you don't control your company. You don't need to code, but you do need to understand how the thing you're selling is built and where it can break.
2. Growth doesn't forgive bad architecture. An app that holds up for a thousand people can collapse with a hundred thousand. You have to build for the day it works.
3. Don't outsource the core of the business without technical oversight. Hiring an outside team isn't the mistake. The mistake is not having anyone on your side reviewing what they build.
4. Going viral isn't a business. Hitting #1 means nothing if you don't know what to do the next day.
5. Launching in stages protects the product. A waitlist or a gradual rollout lets you grow at the pace your tech can handle.

Today, every company I build starts with the question I didn't ask myself in 2015: what happens if this works.