"How Do You Know?"

In spring of 2017 I was about to move from San Francisco down the peninsula. I had just decided to pause an ambitious side project that attempted to further unsupervised machine learning by automating clustering and instead turn my attention to another project dealing with art¹. Y Combinator had announced their first startup school; I applied and was able to join a group of other entrepreneurial types.

My art project² was going to sell physical art. I had developed an app that enabled people to browse and buy artworks with Stripe. Sounds prosaic but there was an angle of course. Even so, this endeavor was unfortunately not a sit-behind-the-screen and dominate³ type of project like my ambitious unsupervised machine learning project. It required that I go talk with customers and artists. 😱

It took time, but at the very least I was able to start emailing artists via a yearly Bay Area art event that produces a magazine with all of the participants’s contact info. I had a pretty good success rate and set up a meeting with one artist, with another meeting on the way. Not only was I making progress, but also discovered that when you tell artists that you want to sell their work, some want to hear more.

At this point it was about 4 or 5 weeks in, and things were moving too slowly. Yes, it is important to Do Things That Don’t Scale but I was treating this like it was Y Combinator’s actual program and if I didn’t have anything to show by presentation day, why bother?

So I pivoted to another iOS app. This one leveraged the Lob printing API to produce post cards with artwork on it. The hypothesis was that people would want to send a postcard with artwork from the location they traveled to.

At the weekly meeting, the group lead Aaron Glazer, CEO of Taplytics, had a guest from his former YC Group. We were discussing what was holding us back, and my contribution usually went like this:

I am thinking about trying to sell this concept to real estate agents to thank clients they’ve worked with. But I don’t think they’d want it.

This may seem dumb, dear reader, but I believe a lot of people suffer from this fallacy. You find a reason why it won’t work. Disappearing photos would be great! But that won’t work because you can always take a photo of the photo or perhaps the more familiar Dropbox is something that could be trivially built with rsync and big tech solutions will be pervasive soon enough.

So during this meeting halfway through startup school these guys asked me: How do you know? And they didn’t just say it once. Every time I made an assertion they would ask me, “How do you know?” There is only one way to know whether something will work, and that is to test the market. You needn’t build anything! You could. You could build a prototype. You could produce a video like Drew Houston did with Dropbox. But if you have any assertion without evidence to back it up, you may as well trash it. God doesn’t exist. Is that so? How do you know that?⁴ Likewise, I didn’t know whether real estate agents would be interested in artwork postcards.

After that call I did not follow up with real estate agents because I decided to pivot to augmented reality. But the lesson stuck with me. It is important to ask questions like whether it is the right time and what is the distribution strategy⁵, but you don’t want to make any assumptions without data.

I’ve heard people talk about eliminating items from idea lists, but I’ve never heard advice on how to do that. Consider how awful the idea of AirBnB was. Paul Graham tried to convince AirBnB to change their idea!⁶ Probably what made it work was Michael Seibel’s advice to leverage Craigslist on both the supply and demand side as a distribution strategy.

Whenever there is an idea I’m unsure of, it is now easy to think How do I know? If it is something reason would easily talk me out of like an AirBnB concept, I try to get some data. A handful of data points or a few posts on forums is not enough, just as it wouldn’t be enough in a scientific setting. And as with science, I should at least be able to walk away from the experiment learning something new.