Don't Under-Think It: Making Critical Decisions When Building an iOS Application
When we were approached about building Whisper, my first thought was that this would be a great opportunity to apply a lot of my philosophy of app building. They had an incredibly smart founder, a really good idea, and funding. What they didn't have was a technical team. This is the type of project where our team shines. What follows here is an attempt to generalize some of the decisions we made. One was sufficiently painful enough that I wrote a summary and then decided to make it a second post. The rest are generally applicable to lots of applications and projects.
First Technology Decision - Ruby vs Erlang
Our team has a lot of experience with Ruby, Erlang, Scala, and C#. We only seriously considered Ruby and Erlang for this project. In the end, we went with the Hybrid model - Rails for Admin/Backend, Erlang for API. This turned out to be a great decision, as we're in the top 10% of traffic for all mobile apps with only a handful of servers. Was it a good decision to build a hybrid system with Erlang as the front line? YES. There are lots of good technologies out there, and you don't have to use either of these - but I'm a big fan of using the right tool for the job, and both Rails and Erlang/OTP have their place.
Second Technology Decision - Real-time sockets
We debated adding a special socket for deletes. The reason for this was we wanted to be able to remove inappropriate or hateful content without waiting for a device to refresh. This turned out to be an incredibly powerful channel that we now use for all sorts of things aside from deleting Whispers -- Metadata updates, connected user counts, deciding whether a user is connected and should get a push notification or a more powerful real-time socket update, determining activity, and more. After this experience, and with the ease of creating a socket channel in Erlang, I don't want to build another web application without including this "backchannel". Was it a good decision to implement a real-time channel in addition an HTTP/JSON API? YES.
Third Technology Decision - NoSQL vs SQL
For this application, we decided to use Riak, NoSQL database that has become more popular in the past year, because we expected a lot of users and a lot of Whispers. We decided that the additional up-front cost of Map/Reduces would be OK because we could really scale to dozens of Riak servers and Terabytes of Whisper data. This was not the case, and we'll cover what happened with Riak in a second post like I mentioned in the introduction. Was it a good decision to use Riak? Cue the ominous music and stay tuned...
Fourth Technology Decision - Security
We spent a lot of time thinking about device security and ways to prevent spam, jailbroken devices, fake users, and more. Should we build "banned users" into the app? Should we build filters for text content? Should we have multiple classes of users - trusted and untrusted? After months of operation, I can say:
- Extra time up-front to ensure algorithmic rigor always pays off. Defenses we thought would be useful, such as authenticating all requests, requiring SSL, and others we can't mention here -- though they took a bit more time at the start -- have been incredibly useful.
- Extra time on ontology is NOT worth it. By that, I mean classes of users, admin levels, etc. Until you SEE how people will use an app, behavior is incredibly hard to predict; and some of our up-front user segregation machinations were useless.
Fifth Technology Decision - Dials and Knobs
I have a forthcoming post about software designs, and what separates successful projects from unsuccessful ones. One of the major ways I can tell a team is professional and serious about building a successful app is when they don't come in with a lot of settings. A technology project's odds of success are inversely proportional to the number of customizable settings in the first version of that product.
In fact, I'm going to pause for effect here.
A technology project's odds of success are inversely proportional to the number of customizable settings in the first version of that product.
Whisper resisted for weeks the idea of adding even a settings page. We finally HAD to add one as Apple asked us to allow username customization, so we did so. Contrast this to many of the proposals I get each week with 20 to 25 ways to configure e-mail alerts and push notifications. Quite honestly, no iPhone carrying human has the emotional and mental bandwidth to think deeply about your alert preferences, and that is one of the reasons I'm not cynical about humanity's prospects. Do we need a lot of customizable settings for the first version of our app? The answer is always NO.
And Finally, The Most Important Decision - How much do I trust this team?
The client team in LA had wireframes and ideas for the app along with a great designer. But how much did they trust us? Do we focus on a perfect specification and an iron-clad contract down to the last penny and week? That puts a huge burden on up-front requirements gathering. And it focuses our relationship on a piece of paper. My favorite lawyer often says in the course of drafting bizdev agreements between large entities, "This piece of paper is worth nothing if the two parties in this room aren't committed to the relationship." A good contract is important, that's not up for debate. We work hard to make clear, concise contracts. But we work harder to be committed to the relationship. A scope of work is similar. We work HARD to build accurate estimates around projects, but we always remember that we're not locking ourselves into a fixed vision of an app before we've seen it in our hands.
In this case, we decided to build a scope of work from the wireframes as a target for our development effort. This is most useful because it gives us a sense of what the app is going to take to get it done. Then we basically throw it away after putting every task in that document into Pivotal Tracker. Pivotal becomes our Truth.
Fast forward to a few weeks later, after we had a prototype in-hand -- we threw out two MAJOR components we thought were critical to the first version. We didn't spend two seconds thinking about renegotiating a contract -- we just kept iterating.
Was an initial estimate enough to make a go/no-go decision? Yes. Did we know within 10% what our costs will be? No. In fact, we didn't know what the winning product was until we had built it. But was it a good decision to trust each other without knowing exactly where we were headed? As Whisper is currently the fastest growing social media app in the app store, I'd say YES.