03 August 2010

Measuring Progress in Software Development


I am about to take on the leadership of a new, still-in-formation developer team, on a project - the first of several - of critical importance to the client. This means that everything is up for negotiation: team structure, development methodology, coding styles, frameworks to be used,... everything!

Initially my role was confined to that of Consulting Architect, but, by force of circumstance, has evolved to Architect and Team Leader pro tem for a few months while the client gets their dev team properly resourced and settled-in. Naturally I'm trying to help that along as best I can.


The client initially planned to use a BDUF (Big Design Up Front), waterfall approach to the project. The requirement is extremely well-known and quantified, in a very well understood business domain.

I have never believed in my tummy that BDUF is in any sense realistically or practically achievable, though, even long before the Agile Movement tore the idea to shreds. It is impossible to foresee every detailed design element, no matter how hard you work at it. On the other hand, some Agile proponents seem to say that no up-front design is necessary... Perhaps my hearing is playing tricks with me. I cannot agree with them, either.

So call me a proponent of SDUF: Some Design Up Front.

And on the Process front, I don't think there's a lot to argue about when we contrast a waterfall/sequential process with an agile/incremental process. For me the critical difference lies in how we report and feed-back progress and how frequently we do this. And what we do about the feedback we receive - how flexibly we accommodate direction changes from customers, business sponsors, unit-tests,... to change the still-in-the-pipeline development and requirements without completely trashing the budget and time-to-market constraints. An also-essential aspect of agile development is "to reflect on what has gone before, and to adjust what we do to make
things better." [Ron Jeffries]

Waterfall possibly still does have a place in some circumstances. I can't honestly say that I've ever actually been party to such circumstances, though I've certainly been on projects where some of our business partners thought they needed hard, contractual milestones with no going back. (In reality we always "went back" anyway, when necessary, after some amount of renegotiation.)


A very greenfield situation, this, which some people would immediately call a "wonderful opportunity", but which I very much see as a "two-edged weapon"...

The question that has been most on my mind is, "What should we measure?"

I am a very firm believer in the old saw, "Tell me how you Measure me, and I'll tell you how I Behave."

Measure a sales-person by the number of sales, and you'll get a high order volume of the easiest-to-sell products, regardless of whether they represent the best margins or quality-of-business for the company. Measure the same sales-person by margin-value of product, and you'd best hope that your high-margin products are ones that lots of people want to buy. Measure them by the number of sales calls they make and you'll have lots of calls that don't result in sales.

Here is where I believe that some Scrum proponents are going wrong... We take Features and break them up into Tasks - the developers' unit-of-work. And they measure Task completions using a burn-down chart of Tasks completed versus time. This can easily result in a situation where many Tasks are being completed, but not so many Features. A situation where Features reach an 80%-complete state, and then get stuck, for any of a variety of reasons, all of which amount to "Nobody wants to complete those Tasks" because they're boring,... or they're "just" test Tasks,... or they're difficult (because not well understood), or...

The solution is really simple. Just measure Feature completion instead of Task completion. Then the team only gets rewarded when Features or User Stories get completed. We only get beer and Pizza when the Business gets value.

But is this enough? Can we go further? Is there a way to tie developer reward directly to delivered Business Value?

In the situation I'm headed into, Business Value should be pretty easy to quantify: The product to be built is one that will directly generate revenue for the company, so we can very easily quantify how much Business Value the software is generating. (Successful completion of the product will also deliver a huge  strategic Business Value by enabling new revenue streams, but that's also quite easy to quantify, and, indeed, is the prime reason the client is taking on this quite substantial investment in the first place...)

Are there ways to close the loop? To feed-back to the dev team on how much business-value their efforts are generating without making money too much of an up-front issue? Then, too, I have a reservation: Developers can have notoriously short memories, and the sort of value we're talking about here is only delivered on longer time-scales... Maybe it's good to have both long-time-loop feedbacks as long as we also have the short-timespan feedback in place as well... Waterfall's failures are largely a result of too little feedback taking too much time for us to correct project course when we need to.

My instinct is that moving towards a continuous deployment process (the step beyond continuous integration) might help to shorten this feedback loop, which is completely the point of "agile" thinking, but I'm still not really clear on how we might implement it.
Related Posts Plugin for WordPress, Blogger...