6 July 2011
A common question in IT departments is whether to provide a capability by building custom software or by buying a package. For longer than I've been programming the debate has raged about how to make that choice. My base position on this is founded on the UtilityVsStrategicDichotomy. Boiled down this means that if the business process you are supporting is part of your competitive advantage you should build custom software, if not you should buy a package and adjust your business process to fit the way the package works.
Despite the clear excellence of my opinion, not a lot of companies seem to do this. Often they neglect the dichotomy, which is one problem. But the problem I want to focus on here is the common trap when they buy a package.
You'll notice above I said "buy a package and adjust your business process to fit". I have two reasons to say this. Firstly if you are buying a package to support a utility business process, then there's no differentiation in the business process - therefore you might as well do that business process in a way that fits the package. Of course this is a very software-person-centric view of the world. Despite the fact that a team isn't doing differentiable work, they'd still rather do it their way, rather than the way some silly software package wants to do it. As someone who believes in people over process, I naturally have a lot of sympathy with that point of view.
But the result of this natural action is that companies start doing significant customization of the package… and this is where the trouble begins. The fact is that most packages aren't designed in such a way as to make customization really viable, at least not on a significant scale. Typically they lack what my colleague Scott Shaw calls "deliverability" - such things as support for version control, testing, and a deployment pipeline. This makes changes brittle and hard to control.
You can bear this if your customization is small, but in many situations it isn't. Recently a colleague came across a package customization that ran to 300KLOC of customization code. That's more code than the entire codebase of our Studios product suite, twice as much as the codebase for one of our larger client projects that's been running a strategic business for a decade. Once you're at those sizes you cannot expect to manage without the tools and processes that you'd use for custom software.
This problem tends to come most to a head when the vendor releases an upgrade of your package, and you find that there is a prohibitive amount of work involved in doing the upgrade because the customizations will break with the upgrade. Gartner recently estimated that it would take $500 billion to bring corporate systems up to latest versions (rising to $1 trillion by 2015). That's a big number, but the real cost is how much money has been wasted in customizations that weren't worthwhile or could have been cheaper with a purely custom route.
So what can you do about it? First off, I think it's important to be somewhat hard-nosed about package customization for utility business functions. Is the cost on the software side really worth it? Although agile approaches matter less for utility functions, the notion of taking small steps is valuable. Can you use the package without customizations initially and try to see how well it works in practice? People will naturally be uncomfortable with the change, because people naturally are. But given some time they may find that things they thought would matter now matter much less.
We can look more seriously about how customization is done. Some approaches are more difficult to deliver than others. Look for someone else that's done a similar level of customization on an older version of the package, and find out what it took for them to upgrade. That may help get a truer picture of the costs. In general for packages we should treat upgradeability as first-class cross-functional requirement.
When you get a vendor to customize a package, it's very hard to get them to follow the customer's delivery practices. The risk that they won't follow those practices is high and should be taken into account as part of your risk planning.
A lot of organizations try to limit the number of languages or frameworks they use. We must remember that many of these customizable packages are, in effect, another language or platform. As a result any arguments made against adopting another language should apply equally well to a package customization effort.
Indeed a common argument against introducing new languages is that it makes it hard to find developers in that language. This issue is usually particularly true for packages, since these often offer a narrow range of employment opportunities. Furthermore the nature of much package customization work deters able people, which makes it even harder to find good people who are more comfortable with polyglot programming. It may be that the costs involved with the difficulty of hiring people could be greater than any cost saving in using the package over a custom development in a mainstream programming platform.
Rather than customize the package, look to see if the package can expose data and functionality through an effective API, and write custom applications for custom capabilities. Often people don't like this because it means they have to use separate applications for different parts of the same work-flow. That may be a smaller burden to bear, and can be made smaller with web interfaces. Vendors may make this hard as it can reduce lock-in, but ease of collaboration should be an important part of choosing which vendor to go with.
But herein lies a trap. One of the big sources of customization is integrating between different vendor packages. This is a big reason to prefer a single vendor for multiple packages rather than picking best-of-breed. Picking a single vendor makes it easier to do the integration as it's in their interest to get it right. If it's a utility business function, then the value of best-of-breed packages is limited anyway.
What it all boils down to is that package environments usually provide a very poor platform for software development. Such packages cost a lot more to customize and keep current than many people tend to think.