I was reading through some posts from Bob Sutton, one of my favorite management gurus, and I ran across a post that contains one of my favorite Dilbert comic strips.
Bob Sutton’s post, as well as the comments that I made on his blog, reminded me of one of my favorite topics: misused Quality Assurance metrics.
“Treat monetary rewards like explosives, because they will have a powerful impact whether you intend it or not.” –Mary and Tom Poppendieck, authors of Implementing Lean Software Development: From Concept to Cash
Over the years, many people have asked me what Quality Assurance metrics they should use to evaluate employee performance. My advice is that Quality Assurance metrics should not be used directly to evaluate employee performance. The Dilbert comic strip may seem a bit extreme, but it’s exactly what happens when employee performance is based strictly on metrics. This is true regardless of whether monetary rewards are explicitly tied to the metrics or not.
In my comments on Bob Sutton’s blog, I mentioned three specific metrics that had unintended effects when used for evaluating employee performance:
Many people think that they have the ability to write a set of metrics that can be used to unequivocally gauge the performance of a Quality Assurance professional, but I have not yet encountered a metric that couldn’t be manipulated to favor the employees.
(If the metric can’t be gamed, it probably isn’t under the control of the employees, so it wouldn’t be effective at driving behavior anyhow.)
Actually, metrics are a great tool for identifying coaching opportunities and potential problems. However, in order to get honest metrics, they shouldn’t be used directly for employee evaluations or employee rewards.
When I’ve looked at the metrics that I mentioned earlier with an eye towards coaching, I had excellent results.
Good Quality Assurance metrics provide powerful tools for managing a Quality Assurance team when used properly. However, they shouldn’t be used in a vacuum. They should just be considered one data point among many.
I was only able to scratch the surface of this topic in this blog post. I plan to discuss specific metrics in future blog posts. In the meantime, if you want to read a much more in-depth review of the pros and cons of employee incentives, you can find one paper here.
I know that a lot of people feel passionately about Quality Assurance metrics, both pro and con. I’m very interested to hear about your experiences with Quality Assurance metrics. Have you found any that were particularly useful? Have you found any that had unintended consequences?
As I’ve mentioned before, a key aspect of building quality software is ensuring that it does what the users need it to do. In my experience, the backlog of feature request (whether written or held in the stakeholders’ heads) is always much larger than what the development team can build in a short period of time.
When the backlog gets too big, people could spend more time managing the backlog than actually building anything. What is more likely, though, is that most of the backlog is ignored, and the clutter causes great ideas to get lost. I have seen cases where key customer issues ended up unaddressed for months until the customer complained a third or fourth time.
I sat down with one of my clients to look at their backlog, and we found that they had over 400 backlog items that had not even been viewed for more than a year. They had more new, high-priority work coming in than they could deliver, so their backlog was growing. Clearly, nobody was ever going to review, let alone work on, the items that were over a year old. I suggested simply closing the backlog items that hadn’t been touched for over a year, but the client didn’t want to remove any items from the backlog without first reviewing them in a meeting with a team of key people, which was not going to happen.
The discussion reminded me of an episode of Hoarders where they were trying to convince someone to sell most of his 27 tool boxes. He agreed that 27 might be overkill, but he still didn’t want to sell them and insisted that the average person, who wasn’t a handyman like him, would need at least 7 toolboxes.
When a backlog gets hopelessly large, you may want to consider declaring backlog bankruptcy (based on the concept of email bankruptcy) and simply close all items that haven’t been looked at in over a year. If that sounds scary, I can understand. I tend towards hoarding myself, and I hate the thought of getting rid of something that might come in handy later. If declaring backlog bankruptcy, it may help to keep these ideas in mind:
Once you cleaned up the backlog, you want to try to keep it manageable. It helps to have a weekly triage process where the items are reviewed and prioritized. Some decisions that should be made during the triage process are:
I’ve found that it’s easier to identify which issues to review if you create a report that shows the priority of each item, the date it was entered, and the date it was last reviewed. This type of report helps ensure that older items are addressed.
What challenges have you had with backlog clutter? What actions have you tried to address the challenges? I’d love to hear your stories from the trek.
In a past role, I joined a test team that was in a User Acceptance Testing (UAT) “Death Spiral” that had caused the user base to lose confidence in the integration testing team. Based on conversations that I’ve had with others, I believe that the UAT Death Spiral is a common scenario that people encounter, and it can destroy a test team. It took some work, but we were able to pull out of the downward trajectory and regain a functional, productive partnership between the business folks and the integration testing team.
Our company had an understaffed integration testing team, aggressive deadlines, and a culture that valued meeting deadlines above all other goals. This meant that software often had known major bugs when UAT started. Even worse, there were typically areas of code that hadn’t been tested at all by the integration testing team before UAT started. It was common for the UAT team to find major, obvious bugs.
Because major bugs made it past the integration test team, the business felt a need to create a more formal, robust UAT team that would catch the numerous errors missed by the integration test team. The business folks assumed that the bugs were missed because the integration team wasn’t very good at their job.
As UAT became more robust, they realized that they needed more time to complete their testing. The company culture of tight deadlines meant that release dates could not be extended to accommodate in-depth UAT. Instead, the business insisted that the integration test cycle be shortened and that UAT start earlier.
I think you can see where this is going…
When UAT started earlier, they found even more bugs and the business lost even more confidence in the integration test team. The business then insisted on testing more of the functionality in UAT and starting UAT even earlier. This death spiral continued until the business had lost complete confidence in the integration test team. However, because the UAT team was not made up of experienced testers, the business was not finding all of the bugs either.
Basically, the integration team didn’t have time to do their job, the business was spending a huge amount of resources to test everything, even features already tested by the integration test team, and the quality of software in production was as poor as ever.
Breaking free of this downward spiral was more than just a logistical problem. It was a political problem as well. The integration test team feared that the UAT team was trying to take over their jobs and the UAT team felt that the integration team wasn’t competent. Rebuilding trust was critical for any new process to be successful.
I worked with both the integration and UAT test teams to plan a new strategy. The strategy was that the integration test team would first test anything that did not have a user interface. In addition, the integration test team would write, maintain, and run automated regression tests. Basically, they would test the areas that required their expertise. Only after these areas were tested and any major bugs were fixed would the UAT team start their work. We would divide up the test cases to reduce any overlap of testing between the two teams as much as possible.
Even though the UAT team agreed that this plan made sense theoretically, they feared that removing the redundant testing would mean that bugs were missed and worried that starting UAT later would mean that they wouldn’t have time to complete their work. I convinced them to give the plan a try on a smaller project. If the advantages to the new plan didn’t materialize, it would be easier to adjust for the lost time on a small project.
Fortunately, on the small project, everything fell into place. The integration test team was able to adequately test their portion of the plan before UAT started, and the UAT team knew exactly what parts were and weren’t tested by the integration team. The UAT team had a shorter test cycle both because they didn’t run as many redundant tests and because the initial quality of code was much better. Each bug takes time to find, fix, and retest.
Best of all, the software went live and had no problems in production.
For the rest of my time with this team, we followed the new process. This resulted in higher quality code with lower cost, and it had the added benefit of greatly improving the working relationship between the two teams.
I’d like to hear your experiences with User Acceptance Testing issues. Have you been in a situation where the business users lost confidence in the development team or integration test team? If so, what do you think were the root causes of the issues?
A key aspect of building quality software is ensuring that it does what the users need it to do. In my experience, the backlog of feature request (whether written or held in the stakeholders’ heads) is always much larger than what the development team can build in a short period of time. However, prioritizing these features seems to be difficult for people. Everything seems important, so everything gets priority 1.
Of course, if everything has the same priority, the stakeholders are de facto allowing the development team to prioritize the features. This can be a problem because the development team often doesn’t have the visibility to all of the factors that may determine the importance of the feature to the company’s success.
Here’s one example of a client who had an issue with prioritization, and how we arrived at a working solution.
A client had a new system replacing an existing business critical system. Unfortunately, their existing system had reached its technical limits before the new system was fully tested, and management made the decision to go live without much testing. Of course, the results were predictable. There were many errors in a production system that had to be fixed right away.
The “prioritization” method initially was that end users would come into the room of developers and tell them that they needed to drop everything and work on whatever issue the end user mentioned. The problem was that many different users were coming in each hour, and the developers didn’t get a chance to finish any task before being told to drop it and work on something else.
The first thing we did was set up a SharePoint list where users could report their issues. We created a process where the users would report their issues in SharePoint. Then, I would triage the list and assign the work to developers. This simple improvement resulted in a huge increase in productivity for the development team because they could complete tasks without interruption.
However, we weren’t always working on the most important issues. Users were choosing the priority, and every issue was the highest priority to that user. Even when we met with representatives from all departments together and set definitions for priorities, every issue was priority 1 on a 3-priority scale.
Our original 3 priority levels were called “High”, “Medium”, and “Low”. Because all issues were production issues, people didn’t want to minimize any by calling them “Medium” or “Low”. Everyone agreed that the issues were not all the same priority, but they weren’t willing to prioritize using those names.
First, the client came up with a category called “Priority 1 – Urgent”. This was higher than “Priority 1” and the client felt comfortable putting some items in Priority 1, and some in Priority 1 – Urgent. Still, way too many items were in Priority 1 – Urgent, so the development team was still choosing the priority.
Then, the client decided that the most critical items would be in a new priority called “Priority 0”. This was reserved for the top 5-7 issues to be worked on immediately by our 5-person development team.
This worked! The client was completely willing to prioritize into “Priority 0”, “Priority 1 – Urgent”, and “Priority 1” even though they were not willing to prioritize into “Priority 1”, “Priority 2”, and “Priority 3”. Just by changing the names of the priority levels, we were able to accomplish the goal of dividing the issues into 3 different levels.
We could then focus on the issues that brought the most value to the system.
What challenges have you had with prioritizing features? What actions have you tried to address the challenges? I’d love to hear your stories from the trek.