Tag Archives: system development

The legacy system modernization trap

Government agencies have often failed at efforts to modernize their legacy systems. Even when the effort is not labeled a failure, it is expensive, time-consuming, and risky. There is a good reason for this: we should not be doing legacy system modernization projects at all.

Why not? To begin with, “modernization” is not a business goal. It does not in itself represent a business outcome that adds value to the enterprise. As a result, its scope of activities can never be clear, the business cannot feel a sense of urgency (and as a result wants to hold off releasing until the system is perfect), and good prioritization and trade-off decisions cannot be made. It is a project that is all risk and no return: there is already a functioning system, so the agency is only risking failure. A modernization effort can never end, since being up-to-date is a goal that slips away as we approach it.

Legacy modernization also does not lend itself to an agile approach – that immediately should be a give-away that something is wrong. We cannot release the modernized product incrementally, beginning with a minimally viable product, because the business will not want to use something that does not at least match the functionality of the legacy system. This means that the scope of the first release is fixed and cannot be adjusted to fit within a time box. The legacy system must be replaced in a single release.

Yet there is a way to modernize a system. The first goal should be to bring contemporary technical practices to bear on the legacy system as is. And the first step in doing that is to bring the legacy system under automated test. A wonderful reference on doing so is Working Effectively with Legacy Code by Michael Feathers. Once the system is under automated test, regressions can be avoided and the code can be restructured and refactored at will. The second step is to re-architect (as little as possible) to make re-engineering pieces of the system efficient. In many cases this will involve creating elements of a service-oriented architecture – or at least finding ways to loosely couple different pieces of the system.

Now to the real point of the exercise. We need to work from real business needs – capabilities that the system does not have. For each capability we change the system to deliver it. We do so in an agile way, and we use our automated tests to feel comfortable that we are not breaking anything. Over time, the system becomes easier to change as components are replaced by more “modern” components. But we never do anything without a good business reason to do so.

In other words, we’ve modernized our system without doing any such thing.

Advertisements

Optimize the whole

Let’s talk about how to reduce government IT cycle times (mission need to deployed capability). As the Lean principle urges, we need to “optimize the whole.” The “whole” in this case goes well beyond system development. We can implement Continuous Delivery practices but if we don’t address the other parts of the value chain, our impact will be tiny. And we’re shooting for a big impact, right?

To go from mission need to deployed capability, here are some of the steps we may need to pass through: gathering requirements into a monolithic “program,” documenting the program and its plan, satisfying oversight bodies, securing funding, executing one or more procurements (typically for development services, hardware, and software), onboarding contractors, developing and testing the product, having it retested by independent verification and validation (IV&V) partners, configuring production hardware and networks, receiving an Authority to Operate (ATO – verification that it is secure enough to release), passing it through a change control board, and deploying the system.

I’ve probably missed a few things here. But a few things should be clear. First, the “value add” development part is tiny compared to the whole (both in time and cost). Second, there are many opportunities for leaning out the process, even if we still need to execute each of these steps. Third – and most subtly – there are real “business” needs, in the government context, driving each of these steps. We really do need to ensure that the system meets FISMA security requirements. We really do need to secure funding through an appropriations process. And so on.

So the question for me is: how can we meet these “business needs” of the government through the leanest possible process?