Monday, March 15, 2010

The Value Proposition

Suppose for a moment that you had a great idea for a new but non-trivial piece of software. Somehow, through a bit of luck, you managed to secure just enough money to get it built and released, but only enough for you to do all of the work by yourself.

That is, you've come into possession of a empty company with just enough money to pay your salary for a couple of years, but nothing else. The money is guaranteed, but until you find some other source of income that's all you are going to get.

It's a simple scenario, but a strong one in being able to evaluate software development methodologies. It works well, because your time is limited, and for every choice you make there is an opportunity cost in terms of what you could have been doing instead. It forces you to make better choices.

If you spend your days writing long-winded comments for instance, you've severely cut down on your development time. Given that you can't afford another programmer, comments aren't all that useful are they? Whose going to read them?

Of course, in this scenario you can't win until you get the software finished, packaged, and shipped to people. A whole bunch of people. But initially you don't even have a market, or a way to generate sales, just a great idea.

For the sake of simplicity, we'll say the idea is so great that in the hands of a reasonable company it will definitely generate revenue with some type of reasonable profit. That is, it is a winner, even if it's not a big one.

To get it going, you'll have to get several versions built and released to a number of different clients before the revenue becomes consistent enough to hire new people to replace yourself.

Still, just knowing that the idea can win, doesn't lead one to being able to make it so. Initially you'll need demos, marketing materials, and enough time set aside to practice a song and dance about why people should commit to this product. You are not only the software developer, you're also the salesperson, or at least the pre-sales engineer.


You realize that this is a chance of a lifetime, to be able to create your own unique product from scratch, so you'd be crazy to turn it down. Somethings are just worth doing.

The first priority would be to set up an environment in which to work. A development machine is obvious, but some type of source code control would definitely help in tracing problems, keeping track of changes, and tracking down bugs. A separate source server would be easier to backup.

You'll have to set up the whole environment, including the development and test machines, so you'll probably want something really simple and easy to maintain. You don't want to lose a lot of effort to system admin tasks, and you don't have enough to out-source this issue.

Of course for tools, you don't want to waste a lot of time learning some fancy environment; the simplest, but most straight-forward tools are always the best. You need to edit, and search the code, but you'll also need some powerful debugging from time-to-time. You really don't want to lose a week or two in trying to track down a stupid bug with only print statements.

Assembling a few lines of working code is not a hard problem, but as that code grows and grows, you'll quickly find that "structure" at multiple levels is important. It is there to separate the code, encapsulating it so it doesn't come back to haunt you. It also makes it way easier to triage bugs, and relate them to specific sections of code, something that will come in really handy later, when the first support calls start to eat into development time.

Structure is fine, but given that you'll be working on this system for years, and given that you'll need to do the sales yourself there will be plenty of interruptions, so you'll need to write down you plans.

Of course, spending six months to produce the perfect documentation, with arrows and charts and what not, is a killer waste of time. You only need enough documentation to remind you of exactly what you planned to do. Just enough to keep you honest. But enough to remind you what is important on extremely busy and disruptive days.

Early on demos will be necessary, so version control plays an increasingly important role. You must be able to code full stream, but fall back to a consistent version for a demo. Buggy demos hurt, and getting sales is your only chance at getting help. Of course a separate demo machine is important, and so is being able to update it quickly to the latest and greatest version. Another side issue is being able to quickly fill it with demo-related configuration and persistent data. And empty system doesn't demo well.

Now, in thinking about your design, you realize that if you just belt out the code you'll need way over half a million lines. Given that you've only got a couple of years, and you can't pound out that much code that quickly, you have to resort to re-using as much code as possible to reduce the code size into something manageable. Abstraction is the only way to do this. You find an intelligent way to implement some internal generic structure that allows you to hang all of the required functionality onto the same infrastructure, but without continuously duplicating each piece separately. Done well, the code comes down into the thousands of lines range.

Still, there are way more features than you initially need to get a sale, and you really want to get sales early to build up both experience and capital. As such, the 1.0 version of the system has got to be ready early, and you'll keep extending it, by adding in new features until it gets closer and closer to the product you imagined. While structure was really great in encapsulation and bug fixing, it really becomes crucial in handling how you will extend the system. The structure delineates what you think are the permanent lines, not the ones in which you'll end up replacing or refactoring code. In this way, the modules, libraries, components and other pieces make it far easier to do the extensions, particularly in small improvements. The pieces also make it easier to just test a small subset of the overall system. Testing too, is another time-consumptive problem that you have to watch out for.

Given that the the environment will be chaos once the product is out there, but you'll still need to be making a large number of improvements, any effort in structure, or in abstraction will pay huge dividends. In fact, you won't be able to win unless you invest effort enough in both.


The common denominator, again and again is in just doing on the most minimal amount of work possible to get to the results. If you get caught up on a side-track or a make-work project, it's a painful waste of your already limited resources.

The easy way to really value work, then is by tracing it backwards from some necessary requirement. Something you just can't live without. You need software to sell, so you have to write the code. People want it to work, so you'll have to test it. Other people will need to understand it, so they'll be documentation.

You need documentation for the users, so you'll have to both make the interface simple, and provide some trivial documentation. Of course, if you make interface changes frequently you certainly don't want to create a huge amount of extra work by having to keep the user documentation up-to-date. It should be simple, and as resistant to change as possible.

Since you have to sell, you'll need marketing documentation. People always want to know about the features, or how to use the product or a tutorial, or any number of other necessary 'sales aides'. These documents are usually summaries, but they too need to be kept in sync. In fact, they often need to be updated head of the code, given that you are trying to generate interest in upcoming features.

Nothing looks worse than poorly packaged software. While it's common with some shops, vendors do not look professional until they've wrapped their works in nice installers. It may not seem like much, but generally even a simple installer can take weeks if not months, and as the systems progress they need continuous updating as well. Many modern technologies are utterly pathetic to install, so it makes the problems even worse. Suddenly you have to be concerned about the dependencies, like databases, containers and libraries, not just for their technical abilities, but also for their capacity to interfere with the installer.

Of course, while you're on dependencies, you can't forget the massive number of sticky, icky licenses that are floating about. You need revenue, and you don't want to make you're competitor's lives easier by giving them the system, so you'd like to retain control of the source code. That means you have to be careful in choosing which libraries you use underneath. Some of them are legal landmines, just waiting to bite.

One easy thing to forget is support. In the initial days, for your first sales you still have to add in new features, but you'll also be both customer service relations, and support. You'll be on-site post-sales engineering, and will definitely be there for the first few installs, it is both necessary and prudent. System administrators will call with technical problems, but also with other weird and wonderful issues, like which standards you support. Users will call, and it won't matter how simple the interface is, or how well the documentation is, they will still call. They do this mostly because they know you are small, and they can be lazy. And of course there will be bugs.

The worse part about many of the bugs is that they come out of no-where, when you are busiest and just eat up huge segments of time. Just trying to understand and replicate takes effort, but then if the problem requires documentation and/or patching, a few days or a week can just disappear. No matter how good the coding and testing is, there will always be bugs, and some will be very difficult to find. Support is always underestimated, and can be a huge drain on resources.

The initial panic to setup the environment, then create a design and get coding as fast as possible, gives way to a much more disruptive environment once the first sales get going. Of course, you have to sell to get more people, and you really don't want to just hire someone immediately, because the initial cashflow is un-steady. Hiring costs time, and laying off your first employees is both time consuming because you procrastinate, and demoralizing even if they take it well.


Wining in a scenario like this takes an ability to really value work. In an environment with a lot of resources, it is very easy to place a value on useless work because it is deemed to fit some idea or process. But when there is just enough between you and success, you quickly find that a lot of what people think helps them, is really just excess baggage.

Of course some things that you might not think have value, come to play really strongly. Two examples are simple code, and simple interfaces.

Simple code doesn't mean brute force, but it does mean that the code does no more than exactly what it should. Abstractions massively cut down on the code size, so they are absolutely necessary but intricate, fragile and complex code is not. Code needs to be neat, consistent and easily readable, particularly late at night or when you are really busy.

Simple interfaces are another important feature, in that they both cut down on user documentation and support calls. If your system contains some uber-complex algorithm for handling stuff, it will provoke significant effort to educate the users, and that could cost you the win. A bad interface can eat a lot of time in creating it, but also a lot more in trying to support it. These are not resources you can afford.

What's interesting about a lot of our best practices in the software development industry is that they fail very quickly in this type of scenario. The same bloat and excess we put into the code, also goes into the making of the code. At the same type, you'll also find developers that have swung too far to the other way. They panic, becoming cowboy coders, avoiding all process and organization. Either too much or too little causes problems. In both cases, as the projects grow, the complexity increases and the resources get squandered on poor value propositions. If you have plenty of resources to waste, this might not be a problem, but even in large companies few developers have any real spare cycles.