In my first blog post, I reflected on my startup. Here, I’ll go a bit deeper into rethinking why it failed to evoke the social change I had planned. To use the schematic provided by Lessig, the full scope of the problem involves the constraints created by the market, by the law, by social norms, and by the architecture of the solution.

In the case of my company, I think I had done a good job of understanding the market dynamics in general. The product I created would realistically create more opportunity for all players in the space… but it could potentially pose an issue to incumbent powers. As I stated in my first blog post, in order to make our model as best as it could be, we needed access to a massive amount of high-quality ticketing data… and yet, I had failed to see that the only people who had that kind of data were the very monopolies we were trying to undermine. This ultimately meant that if we wanted to make something that worked, the practical use of what we made would be determined by the company that owned it. This created an aspect of the regulation of the technology that we didn’t have control over.

While the threat to the incumbent powers proved a problem, theoretically, this could have been overcome with the other constraints. For example, we could have found a way to shift the social norms of artists. The reason that the big monopolies are the only place to find quality ticketing sales data is that they are private data sets. There are public data sets available—in fact, we tried to train our model on them. These free public datasets were of poor quality because promoters who ran events would self-report the data. Often wanting to influence musicians to work with their brand or venue, we found that this data was often inflated or inaccurate. In a similar way, musicians had no benefit of posting public data… it took effort, tech-savvy, and could contain information that hurt their brand. Indeed, one of the benefits of having a privatized data set from a big conglomerate is that they, just like our models, needed to understand why shows performed badly so they could figure out how to avoid making the same errors in the future. Because the data was internal to the company, there was no downside of honesty. If we had created a platform in which we worked directly with middle-class musicians to collect the data about their shows and show them how working with us would benefit them, we could have become the locus of the data set we needed. As a team, we were afraid to take on such a big challenge (we’d need lots and lots of artists to be valuable enough) and for that reason didn’t pursue this option. As for law, right now the major conglomerate is under investigation for potentially violating anti-trust laws, but that process is slow moving.

As for architecture, I believe the solution I outlined of working with the musicians themselves to collect the data is the best bet… and that, of course, depends on the structure of the platform used to deliver our AI model. Strangely, the company we exited to is actually best positioned to deliver the solution I outlined, provided that there are enough venues able to get out of their exclusive relationships with the large conglomerate. I guess I had been thinking of this as a failure when actually, we most likely lucked into having the product land at a place where it now has the highest likelihood of impact.