This hit me like a freight train when I read it.
In Alan Shalloway's Lean Online Training, we're learning about the Myth of Optimization Through Decomposition, which states that trying to go faster by optimizing each individual piece does not speed up the system.
In the physical world of manufacturing, attempting to run every single machine at 100% utilization results in large piles of unfinished product just sitting around waiting to get through the next step of the pipeline or for a buyer. These unfinished products incur significant costs in terms of inventory and storage. And, whenever the product line is changed or stopped, whatever is sitting in that pipeline winds up being thrown away. This is why physical operations do best when they use a Just In Time strategy -- creating only what they need and no more. It turns out that operating each machine at 100% utilization is actually a really bad business decision.
In the world of software development, the parallel to running every machine at 100% utilization is making sure every employee is busy 100% of the time. And, just like in the physical world, this results in large amounts of unfinished works in progress that incur significant costs and risks. Knowledge degrades quickly, requirements get out of date, the feedback loop is delayed so we don’t learn what we’re doing wrong. The result is unfinished, untested, misunderstood, and often flat-out unnecessary code bogging down our product, degrading its quality, and, actually slowing us down.
- It's difficult implementing a feature that was specified so long ago that no one can remember what it's for.
- It's hard tracking down an error in code developed so long ago that no one remembers how it was implemented.
- It's slow adding new features when the software is muddled with unfinished, untested code (that isn't even needed!).
Thus, Lean teaches us that striving for 100% utilization is not the answer. It doesn't get the product completed any more quickly, and the only thing it creates is waste.
The only way to go faster is the optimize the whole. In other words, find your bottlenecks -- the things that are slowing down the process, incurring delays, and adding waste -- and remove those. And when you do, a funny thing happens, it lets your developers work faster! They're happier, you're happier, and ultimately the customer is happier.
From my own experience I offer some indicators that reveal the truth of this:
It's difficult implementing a feature that was specified so long ago that no one can remember what it's for.
Imagine managing development for 3-4 major products and shared infrastructure, each of which has a product backlog from dozens to over one hundred things! Imagine product owners that want to estimate everything they can imagine in a product over N releases up-front, “so we can inform the contents of each release partially based on how big things are.”
In my experience with, say, e-commerce web applications, writing user stories of any reasonable fidelity, it’s unusual to pack more than ten things into any release. More than that requires too much time for the release or too many people to get the job done in a reasonable time. When you’re done it’s likely that ten more things have appeared that are at least as important to the business as the next ten in the backlog. A backlog longer than ten is waste in this situation.
I need to say something about story fidelity. When I see a backlog of dozens or hundreds I usually see very finely-tuned stories. In my experience, a story that doesn’t stand on its own when implemented in a product is too fine-grained for a product backlog. A story needs to describe a complete picture so that when someone looks at a story two months later, someone unfamiliar with the backlog reads a story, they can quickly and easily understand the feature.
Now, I realize I have made a nasty generalization with my ten-item-backlog example. The point is that a backlog of 100 is pretty darn difficult to prioritize and manage. The backlog simply becomes a list of things someone one day thought were needed. Estimating it is waste. Prioritizing it is probably impossible.
Finally, as a development manager you need to resist aggressive product owners who try and pack as much as they can onto your agenda. The belief is that if every available hour of every resource is planned that we are working at maximum efficiency. Wrong. Dense-packing software development teams like that guarantees lots of overtime or missed deadlines or both. If you schedule everyone to the limit you have no “surge capacity”. Without surge capacity you’re dead; you are forced into nights and weekends and you have grumpy people.
It's hard tracking down an error in code developed so long ago that no one remembers how it was implemented.
I have a rule: Whenever you work on old code you always refactor it to leave it better than you found it. If you’re a good programmer do you ever remember working on old code you couldn’t improve? I don’t.
It's slow adding new features when the software is muddled with unfinished, untested code (that isn't even needed!).
I recently helped my company with some technical due diligence evaluating the acquisition of another company and its software. Company B uses a development process in which they release their product religiously every X weeks pretty much regardless of whether new features are completely finished. They have conditioned their user community to expect partially completed or incompletely tested stuff; indeed they say their users enjoy being treated to “sneak preview” features and Company B uses feedback to improve these features before they are completely done. As a business model this works for them and that is wonderful.
I argue that care needs to be taken with this style of development. It is easy to get distracted and start adding new things without finishing old things leaving code littered with partially completed work. Clever branching might help mitigate this problem, but doing so adds complexity to the development process nevertheless.
