In response to my previous blog post “Another Silly Quantitative Model”, Greg writes: In my current project, the product owner has assumed the risk of any financial losses stemming from bugs in our software. He wants to release the product to customers, but he is of course nervous. How do you propose he should best go about deciding when to release? How should he reason about the risks, short of using a quantitative model?
The simple answer is “when he’s not so nervous that he doesn’t want to ship”. What might cause him to decide to stop shipment? He should stop shipment when there are known problems in the product that aren’t balanced by countervailing benefits. Such problems are called showstoppers. A colleague once described “showstopper” as “any problem in the product that would make more sense to fix than to ship.”
When I was a product owner and I reasoned with the project team about showstoppers, we deemed as a showstopper
- Any single problem in the product that would definitely cause loss or harm (or sufficient annoyance or frustration) to users, such that the product’s value in its essential operation would be nil. Beware of quantifying “users” here. In the age of the Internet, you don’t need very many people with terrible problems to make noise disproportionate to the population, nor do you need those problems to be terrible problems when they affect enough people. The recent kerfuffle over the iPhone 4 is a case in point; the Pentium Bug is another. Customer reactions are often emotional more than rational, but to the customer, emotional reactions are every bit as real as rational ones.
- Any set of problems in the product that, when taken individually, would not threaten its value, but when viewed collectively would. That could include a bunch of minor irritants that confuse, annoy, disturb, or slow down people using the product; embarrassing cosmetic defects; non-devastating functional problems; parafunctional issues like poor performance or compatibility, and the like.
Now, in truth, your product owner might need to resort to a quantitative model here: he has to be able to count to one. One showstopper, by definition, is enough to stop shipment.
How might you evaluate potential showstoppers qualitatively? My colleague Fiona Charles has two nice suggestions: “Could a problem that we know about in this product trigger a front-page story in the Globe and Mail‘s Report on Business, or in the Wall Street Journal?” “Could a problem that we know about in this product lead to a question being raised in Parliament?” Now: the fact is that we don’t, and can’t, know the answer to whether the problem will have that result, but that’s not really the point of the questions. The points are to explore and test the ways that we might feel about the product, the problems, and their consequences.
What else might cause nervousness for your client? Perhaps he’s worried that, other than the known problems, there are unanswered questions about the product. Those include
Unanswered questions about the product are one indicator that you might not be finished testing. There are other indicators; you can read about them here: http://www.developsense.com/blog/2009/09/when-do-we-stop-test/
Questions about how much we have (or haven’t) tested are questions about test coverage. I wrote three columns about that a while back. Here are some links and synopses:
Got You Covered: Excellent testing starts by questioning the mission. So, the first step when we are seeking to evaluate or enhance the quality of our test coverage is to determine for whom we’re determining coverage, and why.
Cover or Discover: Excellent testing isn’t just about covering the “map”—it’s also about exploring the territory, which is the process by which we discover things that the map doesn’t cover.
A Map By Any Other Name: A mapping illustrates a relationship between two things. In testing, a map might look like a road map, but it might also look like a list, a chart, a table, or a pile of stories. We can use any of these to help us think about test coverage.
Whether you’ve established a clear feeling or are mired in uncertainty, you might want to test your first-order qualitative evaluation with a first-order quantitative model. For example, many years ago at Quarterdeck, we had a problem that threatened shipment: on bootup, our product would lock system that had a particular kind of hard disk controller. There was a workaround, which would take a trained technical support person about 15 minutes to walk through. No one felt good about releasing the product in that state, but we were under quarterly schedule pressure.
We didn’t have good data to work with, but we did have a very short list of beta testers and data about their systems. Out of 60 beta testers, three had machines with this particular controller. There was no indication that our beta testers were representative of our overall user population, but 5% of our beta testers had this controller. We then performed the thought experiment of destroying the productivity of 5% of our user base, or tech support having to spend 15 minutes with 5% of our user base (in those days, in the millions).
How big was our margin of error? What if we were off by a factor of two, and ten per cent of our user base had that controller? What if we were off by a factor of five, and only one per cent of our user base had that controller? Suppose that only one per cent of the user base had their machines crash on startup; suppose that only a fraction of those users called in. Eeew. The story and the feeling, rather than the numbers, told us this: still too many.
Is it irrational to base a decision based on such vague, unqualified, first-order numbers? If so, who cares? We made an “irrational” but conscious decision: fix the product, rather than ship it. That is, we didn’t decide based on the numbers, but rather on how we felt about the numbers. Was that the right decision? We’ll never know, unless we figure out how to get access to parallel universes. In this one, though, we know that the problem got fixed startlingly quickly when another programmer viewed it with fresh eyes; that the product shipped without the problem; that users with that controller never faced that problem; and that the tech support department continued in its regular overloaded state, instead of a super-overloaded one.
The decision to release is always a business decision, and not merely a technical one. The decision to release is not based on numbers or quantities; even for those who claim to make decisions “based on the numbers”, the decisions are really based on feelings about the numbers. The decision to release is always driven by balancing cost, value, knowledge, trust, and risk. Some product owners will have a bigger appetite for risk and reward; others will be more cautious. Being a product owner is challenging because, in the end, product owners own the shipping decisions. By definition, a product owner assumes responsibility for the risk of financial losses stemming from bugs in the software. That’s why they get the big bucks.