My biggest mistake with Glassboard was having blinders to the fact that the design decisions Sepia Labs made when building Glassboard would likely clash with mine. That doesn’t have anything to do with visual design. Instead, I’m referring to the platform architecture decisions they made as a funded startup, or spin-off, or whatever they were.
The folks at Sepia had a parent company in Newsgator that was funding everything on Glassboard from the server costs, their salaries, and all the perks that go with that. More importantly, they were funding these things with the typical startup mentality in mind: get as many users as humanly possible and then worry about monetization later.
Glassboard under Second Gear had a completely different mindset: try to convert our most passionate existing users to paid users, and then grow from that. I sadly failed at doing that. There are a variety of different reasons for that failure, but one of them is how much time I spent undoing some architectural decisions from their previous owner.
That’s not a knock against Sepia or NewsGator. They built a great product. They just built it with a far different set of constraints than I had as the new owner funding this out of pocket.
E-Mail
One of the biggest was e-mail. Glassboard was sending over 150,000 emails a day when I took over. The SendGrid bill was $2400 in November of 2013, which is kind of crazy. I was able to cut that bill in half by negotiating a better rate to account for how much email we send, but it was still four figures a month.
Sepia didn’t need to worry about justifying that cost because it was another feature in the checkbox to try and acquire users. Don’t want to use an app and get push notifications? We’ll send you an email every single time someone comments on your board posts!
E-mail to Second Gear’s Glassboard is, however, a sunken cost because none of those costs were ever passed on to my customers. My thinking at the time was that I could convert enough users to premium accounts to fund the transactional email for everyone. In reality, that proved to not be the case.
I spent far too long in that mindset. When I finally wised up and made transactional email a premium user feature, my costs dropped down to under $100 a month. It also pissed off more users than it converted, but that was to be expected. The lesson I learned is that people wanted those emails, but they didn’t want to pay for them. It was a nice-to-have thing, but not a must have. I wish I had learned that lesson $15,000 ago.
Computing Costs
The biggest project I undertook during my time with Glassboard was converting the entire backend to use computing resources more efficiently. There were three different cloud services that powered Glassboard:
Notifications Role: this handled dispatching push notifications, emails, and archive processing. It ran on 2 medium-sized cloud service instances.
Attachment Processor Role: this would take the videos users uploaded, compress them, and make them more mobile friendly. This ran on a single Large cloud service instance. Why I didn’t kill video on day 1 is beyond me. What a poor job on my part.
Glassboard Platform: This was the kitchen sink project. It ran the API, the web app, web hook processors, and pretty much everything else. It was 4 medium sized instances when I took it over (I may have the number wrong, but it was quite a few).
I was able to get a few quick wins by enabling auto-scaling so that the instances would rise and fall based on usage patterns. It’s silly to pay for a ton of servers at 3AM on a Sunday when there’s not nearly enough users to justify the costs.
The biggest issue is that the only way to scale Glassboard in that incarnation was to throw more hardware at it. I spent an ungodly amount of time converting the entire backend platform to use C#’s async-await pattern. I am fairly confident I touched every single file in the project. One does not merely convert one portion of their code to be asynchronous. It’s a rabbit hole that you keep going down until you reach the bottom.
It was a lot of work, but once it was deployed I cut my server costs across the board in half.
Of course, by the time I got to that point, it was too late. I cut costs on a product that wasn’t commercially viable. Oops!
Lessons Learned
Building products with a bootstrapped mentality is completely different than a startup mentality. When bootstrapped, every decision you make affects the bottom line, and that is a bottom line you care about from day one. Trying to convert a platform that wasn’t designed with that in mind proved to be too great of a challenge for me as the sole proprietor of Glassboard. Rather than focusing on improving the core Glassboard product, I spent most of my time trying to cut costs where possible to curb our losses.
Would it have been more feasible if I had some help? Most likely, but I still don’t think that would have saved the platform.
If I had a do-over, I think my first decision would have been to shut down the Glassboard service as it was in November of 2013 and relaunch it without any of the legacy user accounts or data. If you wanted to continue using the service under Second Gear, subscribe now. That would have lowered costs up front and allowed me a lot more runway and likely time to delay a lot of cost-cutting projects that took up most of my 2014. Of course, there’s obvious tradeoffs in this decision too.
Hindsight is a bitch.
Achieving Zen With Auto Layout: I'm writing the new book on Auto Layout and doing it with your help. Purchase today and get beta access to help shape the content of the book. Learn more and get a free sample chapter at https://gum.co/autolayout.