Sadly, a lot of these seem obvious but I guess he needed to learn them the hard way?

Unit economics matter a lot more than in pure software businesses.With Justin.tv, when we wanted to generate more revenue, we would look hard and figure out yet undiscovered places to shove another ad unit. Actual COGs (bandwidth and servers) were only ~30% of our gross revenue – most of our expenses were software engineers working on improving the product. With Exec Errands, we paid out 80% of the $25 / hour that we billed for errand runner time on job. That means that all marginal customer acquisition, customer service, and recruiting sufficient supply of errand runners had to be paid for out of that remaining 20%. Paying for mistakes (workers messing up a job) or customer refunds quickly ate up any profits.

Turnover of errand runners was very high. Most competent people are not looking for part-time work. The exception to this are people looking for supplemental income, but those people have full time jobs which conflicted with our high demand periods: mid-morning of the work week, Monday - Friday. Hiring new errand runners was expensive, and it was difficult to get them to stick around when we couldn’t guarantee work – the average utilization (time on the job / time available) was only 50%. Additionally, most of our errand workers wanted to work during normal work days, meaning there was an undersupply on weekends.

Demand was very spiky. It turns out that $20 / hour does not provide enough economic incentive in San Francisco to dictate when our errand runners had to be available, leading to large supply gaps at times of spiky demand. Without some form of Uber-like surge pricing, it was impossible to ensure that we had consistent availability.

Customer acquisition was hard. Lyft and Uber are viral online-to-offline businesses, because many of the times you call a ride you take it with others (who subsequently think it’s an incredible experience and download the app). Exec Errands was a not nearly as viral (and cleaning not at all), because you could use those services without ever telling anyone.

Customer activation was hard. Only about 10% of people who signed up for Exec Errands activated (put in at least one job request). When they heard about Exec, people thought it sounded like a cool idea, but rarely did they have an errand waiting in their back pocket that they hadn’t yet run – making it hard for them to try us immediately. This led to reduced activation rates and high customer acquisition costs (since fewer signups turned into actual customers).

We shouldn’t have run jobs ourselves. In the early days, if a job request came in and there were no Execs available to run them (as was often the case due to spiky demand), we sent someone from the office to run the jobs ourselves. We called that eating our own dog food and giving ourselves a sense of what could be improved about the product for both errand runners and customers. Unfortunately, this gave us a false sense that the quality of service for our customers was better than it was, as the average competence of our employees was higher than our average recruitable errand runners.

Verticalize. Ironically, Exec was very similar to Justin.tv in this way: the product got a lot better when we started focusing on a specific vertical, and surprisingly, grew much faster. With Justin.tv, we discovered that a small number of users were very passionate about gaming, and thus we started working on Twitch, a community based around watching gaming video, which seemed like a very small niche to us at first. Now, Twitch has 45 million monthly viewers, and is many times the size of Justin.tv. I failed to learn the lesson the first time around, and started Exec Errands as a general errand running site that you could put any job into, from shopping to cleaning to getting your car washed. When you do many things, it is hard to do any one thing well, and Exec Errands turned out to be a merely adequate tool for many jobs, but not an exceptional one. When we started focusing only on cleaning, we were able to vet workers against a smaller skill set, provide specific tools, and streamline the interface to make purchasing more intuitive and simpler (and thus increase conversions), resulting in much better growth.

I don't know why he wasted all that time with Exec when Twitch is the business he was made to do.

Maybe he just needed to get it out of his system?