Intro to Unit Testing 10: The Business Value of Unit Tests
Backstory
I worked for a consulting firm for a while. We didn’t make anything particularly exciting — line of business applications and the like was about the extent of it. The billing model for clients was dead simple and resembled the way that lawyers charge; consultants had an hourly rate and we kept diligent track of our time, to the nearest quarter hour. There was a certain feel-good element to this oversimplification of knowledge work in the same way that it’s pleasant to lean back, watch Superman defeat Lex Luthor and delight in a PG world where Good v Evil grudge matches always end with Good coming out the victor.
It’s pleasant to think that writing software has the predictable, low-thought cadence of an activity like chopping wood where each 15 minutes spent produces a fairly constant amount of value to the recipient of the labor. (Cue background song, Lou Reed, “A Perfect Day“) Chop for 15 minutes, collect $3, hand over X chopped logs. Chop for 1 hour, collect $12, hand over 4X chopped logs. Write software for 15 minutes, produce a working, 15 LOC application for $25. Write software for 1 hour, produce a working, 60 LOC application for $100. Oh, such a perfect day.
When I started at the company, I asked some people if they wrote unit tests. The answer was generally ‘no’ and the justification for this was that you’d have to run it by the client and the client most likely wouldn’t want to pay for you to write unit tests. What they meant by this was that since we billed in quarter hour increments and supplied invoices with detailed logs of all activity, it’d be sort of hard to sneak in 15 minutes of writing automated test code. Presumably, the fear was that the client would say, “what’s this ‘unit testing’ stuff and why did you do it when you didn’t say anything about it.” I say “presumably” because this wasn’t the reason people didn’t unit test at this company just like whatever excuse they have at your company isn’t the real reason for not unit testing there. The real reason is usually not knowing how to do it.
Why did I start out with this anecdote and its centerpiece of the quarter hour billing and development cadence? Well, simply because software development is a creative exercise and far too spastic to flow along smoothly in a low viscosity stream of lines of code per minute. You may sit and stare blankly at a computer screen while contemplating design for half an hour, code for 4 minutes, stare blankly again for an hour, code for 20 minutes, and then finish the product. So, 24 minutes — is that a billable half hour or 15 minutes? Closer to 30, I suppose. Do you count the blank staring? On the one hand, this was the real work — the knowledge work — in a way that the typing certainly wasn’t. So should you bill 1.5 hours instead and just count the typing as a brainless exercise? Or should you bill 2 hours because the work is a gestalt? I personally think that the answer is obvious and the gestalt billing model cuts right to the notion that software development is a holistic exercise that involves delivering a working product, and the breakdown may include typing, thinking, white-boarding, searching Stack Overflow, debugging, squinting at a GUI, talking to another developer, going for a 5 minute walk for perspective, running a static analysis tool, tracking down a compiler warning, copying 422 files to a target directory, and yes, my friend, unit testing.
Those are all things that you do as part of writing good software. And, in a consulting paradigm, you wouldn’t cut one of them out and say, “the client wouldn’t want to pay for that,” because the client doesn’t know what its talking about when you’re under the software-writing hood — that’s why they’re paying you. They wouldn’t want to pay for “searching Stack Overflow” or “Squinting at the GUI” either, but you don’t refuse to do those things when you’re writing software. And so refusing to unit test for this same reason is a cop-out. When a younger developer at that firm asked me why I wrote unit tests and how I accounted for them in my billing, this was essentially the argument I gave — I asked him how he accounted for the time he spent compiling, debugging, and running the application and, bright guy that he is, he understood what I was saying immediately.
Core Business Value
This may seem like a roundabout and long introduction to this chapter, but it really cuts to the core of the business value proposition for unit tests. During development, why do developers compile, run, and debug? Well, they do it to see if their code is doing what they think it should do. Write some code, then make sure it’s doing what you expect. So why write unit tests? To make sure your code is doing what you expect, and to make sure it keeps doing what you expect via automation. The core business value of unit tests is that they serve as progress markers, sign-posts, and guard rails on the road to an application that does what you expect.
Unit tested applications are more predictable and better documented than their non-unit tested counter parts (assuming the same amount of API documentation and commenting are done), and there is an enormous amount of business value in predictability and clarity of intent. With a good unit test suite, you’ll know in minutes if you’ve introduced a regression bug. Without that unit test suite, when will you know? When you run the application GUI? When QA runs it a week later? When the customer runs it a month later? When something randomly goes haywire a year later? Each one of these delays becomes exponentially more expensive.
That’s not the only value-add from a business perspective (and I’ll list some other ones next), but it’s the main one, as far as I’m concerned. It also explains why the notion that you need to carve out some extra time for unit tests and figure out whether the customer wants them or not is preposterous. Do you think the customer is going to get angry if you explain that part of your development process is to execute the code you just wrote to make sure it doesn’t crash? If the answer to that is “no, of course not, that’s ridiculous” then you also have the answer to whether or not a customer would care if you happened to automate that process.
Of course, one thing to bear in mind is that a customer may not want to pay for you to learn on the job to unit test, and that’s a fair point. But if the customer (or your company, internally, if you aren’t a consultant) doesn’t want to foot the bill for this, then you should strongly consider picking it up on your own and then switching customer/company if they don’t buy in to something as fundamental as automating predictability. Unit tests are the software equivalent of accountants practicing double entry bookkeeping, doctors washing their hands, electricians turning power back on before leaving and plumbers doing the same with the water. Imagine if your plumber sweat welded a joint for your new shower, sized it up and then said, “meh, I’m sure it’s fine” and left without ever running the water. That’s what tens of thousands of us do every day when we just assume some piece of code works because it worked a month ago and you don’t remember touching it since then. Ship it? Meh, sure, whatever — it’s probably fine. The business value of unit tests is a stronger assurance that we know what’s going on than “meh, sure, whatever.”
Ancillary Business Value
Here are some other ways in which unit tests add value to the business beyond confirming that the application behaves as expected.
First, unit tests tend to serve nicely as documentation. This may sound strange at first; you’re probably thinking, “how is a bunch of code documentation when we have a whole activity associated with documenting our code?” Well, fact of the matter is that documentation in the form of writeups, code comments, instruction manuals, etc tends to get out of date as the product ages. Unit tests, however, are never out of date because if they were, the build would fail (or at least you’d see red when you ran them) and you’d be forced to go back and “fix the documentation.” If you keep your unit test methods clean and give them good names as described in earlier chapters of this series, they’ll also read more like a book than like code, and they’ll document purpose and intended behavior of the system.
Unit tests also guard against regression. When you write the tests, you’re confirming that the software does what you expect it to at that moment. But what about later? Maybe later you forget what you intended in that moment or decide that you intended something different and you change the code. Will it still work? In a lot of legacy code bases, the answer to that question is, “yikes — who knows?” With a thoughtfully unit tested code base, you can rig it so that a test goes red if a design assumption that you made is no longer true. For instance, say you write some method with the intention that it never return null, and say that eventually you and your teammates build on this method and its assumed post-condition, grabbing values returned by the method and never checking for null before dereferencing them. If someone later modifies that method and adds a condition in which it returns null, the only thing standing between them and introducing a regression bug is a unit test that fails if that method returns null.
The practice of automated unit testing has after the fact benefits such as documentation and guards against regression bugs, but it also helps during the course of development by having a positive impact on your design. I’ve long been a fan of and have previously linked to this excellent talk by Michael Feathers called “The Deep Synergy Between Testability and Good Design.” The general idea is that writing code with the knowledge that you’re going to be writing tests for it (or practicing TDD) leads you to write small, factored classes and methods that are loosely coupled and that this practice, in turn, creates flexible and maintainable code. Or, consider the converse and think of how hard it is to write unit tests for giant, procedural methods and classes. Unit tests make it harder to do things that make your code awful.
Lastly, I’ll throw in a benefit that summarizes my take on this entire subject and really drives things home. It’s a huge bit of editorializing, but I feel somewhat entitled to do so in my own conclusion. I believe that a serious piece of value added by unit testing is that it lends you or your group legitimacy and credibility. In this day and age, the question, “should you unit test your code” is basically considered to be settled case law in the industry. So the question, to a large extent, boils down to whether you write tests or whether you have excuses, legitimate or otherwise (and there are legitimate ones, such as “I don’t know how, yet.”) Don’t be in the camp that has excuses.
Forget justifying what you or your organization has done up to this point, and imagine yourself as a customer of software development. You’ve got a budget, and you’re looking to have some software written that you don’t have time, yourself, to write. All other things being equal, which group do you hire? Do you hire a group that responds to “do you unit test” with “no, we don’t think our customers would want that?” How about a group that responds with “well, there’s this database and this GUI and sometimes there’s hardware, so we really can’t?” Or do you hire a group that responds with, “we sure do, would you like to see some samples?” I bet it’s the last one, if you’re honest with yourself.
So be that last group. Add value to your users and your business. Write good software and consider your design carefully, and, just as importantly, automate the process of ensuring your software does what you think it does. Your credibility and the credibility of your software is at stake.
An efficient summary that I can point people at, along with several happy phrasings and analogies I plan to incorporate into my lexicon of persuasion. Very useful, thanks!
Thanks for the kind words, and I’m glad if you’ve enjoyed. Always happy to lend whatever help I can to make the world of software more predictable and repeatable π
A very usefull series, and a great summary. Thank You once more for sharing π
“Do you think the customer is going to get angry if you explain that part of your development process is to execute the code you just wrote to make sure it doesn’t crash?”
=> If I tell them it will cost 5 days to execute the code just to make sure it doesn’t crash, I can guarantee you a lot of them will get angry.
Is your point that writing automated tests takes you five days?
no no, 5 days was just as an example !
My point was that a lot of customer (in my experience) will always try to cut whatever isn’t pure “conception”.
But now I can show them your post π
Oh, gotcha. (That’s why I ask — I often misunderstand π )
It’s natural for customers to question the value of any activity that you list as a line item on the invoice. I think if there’s a take-away that I’d offer, it’s that “unit testing” isn’t actually a line item any more than “compiling” is. Those things are both covered by “implementation.” My take, anyway.
[…] […]