Imagine a bucket. Here’s one to help.
At the bottom of the bucket is the mainstream. If you know something that other people don’t, you are at the edge of the bucket.
Good startup ideas start with unique information — something that the founders noticed that no one else did.
This is painfully obvious to anyone thinking about innovation.
Tito, Prashast and I shut down our startup, Product ML, a few months ago. We went through a process of coming up startup ideas and it was much harder than we thought. This is the story of why we struggled to find a “good startup idea”.
Let’s do a startup! This was our first mistake. Decide you want to do a startup and then come up with the idea. But no! We are better than everyone else. The standard advice doesn’t apply to us.
Googling “how to come up with startup ideas” brings up an excellent essay by Paul Graham.
Focus on a problem you have, understand that well, build a product to solve the problem. Validate with customers. Since you have the problem, you are the first customer. Wow, a blueprint for entrepreneurship.
Let’s think, people! What problems do we have?
Let’s do something BIG
I was in Burma/Myanmar a few years prior and the internet was very slow. There are lots of remote places where internet is slow. Maybe we should build a satellite internet network?
We quickly developed a “practicality filter”. We didn’t know the first thing about building satellites. Or the internet.
I hate doing laundry. I would pay for someone to come to my home, pick up my laundry, and bring it back the next day. Maybe we should build an Uber for laundry?
A “passion filter”. We were engineers! We want to build something cool and have other people use it. Did we want to hit the streets and sign up drivers for this Uber for laundry thing? That doesn’t sound like fun.
But, we didn’t have enough meaty problems of our own. We started reaching. Unconsciously, we stopped looking for problems and were now coming up with ideas.
At some point, someone said: “Healthcare is broken, right?”.
We didn’t know the first thing about healthcare. We weren’t in the industry. We didn’t know how it worked. We had no unique insights. And we couldn’t see anything that we could build a startup around. Theoretically — sure! But nothing that we were uniquely positioned to solve.
No matter! Let’s list lots of ideas, we’ll do some research and find out what problems are there.
That Trello board looks like a list of startup ideas, not customer problems. Our ideas were shamefully generic. We didn’t know where to start.
We should build this!
We were a year into this process. We had just spent the last 4 months trying to validate a prototype for a “Get things done” style app — only to realize that no one wanted it.
We had already quit our jobs. And we had spent the last year trying to come up with ideas. We were wiped, minds were blank. There are no other ideas!
We accidentally stumbled on to the idea for Product ML. And it happened while we were walking around San Francisco contemplating our fate.
Prashast was talking about a problem at work. They were processing lots of data for their ML models. When data volume dropped in a country, you didn’t know if it was a bug — or just some random thing like a holiday where people did less on their phone.
I had a similar problem in my day job as a gaming product manager. In a game that has micro-transactions, metrics like average revenue per user fluctuate, and you never know if these fluctuations are expected, or if you did something wrong in your product.
Every game product manager can change things in a game to make more money. Features like progression maps in a game are great for long-term retention, but the best holistic outcomes happen when you change the core game design itself. Candy Crush plays differently for you each time, and you don’t always win. These random outcomes can change how much you play, and whether you spend money.
Game designers and PMs play God to change how the game plays by controlling the game configuration. To do this, personalized to a user, without screwing up the game and maximizing long-term revenue was (still is!) very hard. Game PMs would love this. I know, I was one! Adaptive game difficulty can be a powerful tool, but it is difficult to get it right at scale.
Tito had built the analytics infrastructure at McAfee. We started talking about the data engineering challenges involved in gaming.
Product ML was born. Given our backgrounds, this was an idea staring at us in the face. It took us a year to notice it. Then we began to dream. We were a platform for adaptive product design, you see. Using machine learning! Game design is only our starting point!
And there we remained. After another year of trying, we didn’t get to product-market fit. We had customers and proved that they made more money using Product ML. We couldn’t convert trials into long-term engagements. We lost every “build vs. buy” conversation with customers. We pivoted to build ML tooling, eventually open sourced our code and went out in a blaze of glory, as Prashast said at the time.
The Information Distribution Bucket
Reflecting on this journey, I realize that this idea was somehow unique to us. Some combination of our experiences made us see this problem where other people didn’t. “A problem that few others realize is worth solving”.
Only a young person could have come up with Facebook. Undergrads with computers and fast internet + some hacking skills + Harvard dorms for online Hot-or-Not activity. These were the unique ingredients that resulted in Facebook being born. No 40-year-old living in the suburbs with two kids would have the right context to come up with Facebook. This is also why Snapchat was founded by young people.
A combination of unique personal experiences and knowledge caused Facebook and Snapchat to be founded.
Remember the bucket?
At the bottom of the bucket is the mainstream. You have the same information as everyone else. The edges of the bucket are when you have information that is not available to other people i.e. unique information.
Let’s consider building an OS for quantum computing. I get my quantum computing news from TechCrunch. I am very much in the mainstream, and I don’t know more than you do. Certainly not enough to build a quantum computing startup.
Contrast that to someone who is building the software layer to execute a prime factorization problem on quantum computing hardware. They are uniquely positioned to build an OS for quantum computing. I am not.
The founders of Riot Games met at university and were big fans of Asian MMOs. They saw what was missing in the Western gaming market and built League of Legends. They could only have done that if they were rabid fans of Asian MMOs. That unique information, along with their experiences and social context, created the unique conditions for building that startup.
Once an information edge is found, it is used to create something new.
New information, in the form of new ideas and new technology, is introduced into human culture. Some new information sticks around and eventually becomes mainstream. Not all new information sticks around, though. The market decides.
A good test for unique information is that other people don’t see it yet. Or think that it is too niche. Because the mainstream doesn’t have access to the same information that allows them to see your vision.
Good startup ideas don’t come from reading Techcrunch and Quora
So I ask myself — what aspects of my life are mainstream? I try to focus on unique information that I possess, and nurture it.
I hope you do too.
I’ve since joined SuperAwesome, the world’s #1 kidtech company. We are #1 because we are defining the category. We are at the edges of the bucket when it comes to kidtech. Life on the edge is hard, but it is very interesting.
Product ML didn’t scratch the itch to build an early stage startup for Prashast. He’s at Deepcell, and loving the world of biotech startups.
Tito is traveling the world. As predicted, he’s bored of it. But not that bored that he wants to stop. He’s open to remote working opportunities, you can find him on Linkedin.