Quality Is Value to Some Person

A team I worked with spent an afternoon arguing about whether their software was ready to ship. The engineers said no: forty known defects sat in the tracker. The head of sales said yes: three customers were waiting, and they would sign on delivery. Both sides thought they were arguing about quality. They were not. They were arguing about whose opinion of quality counted, and nobody in the room would say so out loud.

That argument is the most common one in software. It is also one of the oldest. Four people spent their careers answering the exact question the team was stuck on, and they never fully agreed with one another. So put them in a room: Philip Crosby, Joseph Juran, W. Edwards Deming, Gerald Weinberg. One question on the table. Is this good enough to ship?

Crosby answers first because for him, there is nothing to deliberate. "Quality is conformance to requirements. Nothing more, nothing less. You wrote down what the software has to do. It either does that, or it doesn't. Forty defects against the spec are forty failures. And don't tell me it's good enough. Good enough isn't a standard. Zero defects is a standard."

Someone asks whether zero defects is realistic, whether quality that strict is affordable.

"You have it backward," Crosby says. "Quality is free. It's not a gift, but it's free. What costs money are the unquality things: every job you didn't do right the first time, every defect you're now paying to find and fix. Those forty defects already cost you. You just haven't added up the bill." He waves off the idea that quality means a better product. "It has nothing to do with goodness or luxury. If a Cadillac conforms to all the requirements of a Cadillac, it is a quality car. The word means conformance. It does not mean class."

Juran is not satisfied. "Conformance to what, though. You talk about the requirements as if they came down from a mountain. Quality means fitness for use, and use is judged by the person using it, not by the person who wrote the spec. Conformance is one part of fitness. Only one. A program can conform to every line of its specification and still be useless for the work someone actually needs it to do." He pauses. "And there are two different things you're both calling quality. One is the features that make people want it. More of those usually cost more. The other is freedom from the defects that make people regret it. Fewer of those usually cost less. Decide which one you're arguing about before you argue."

Deming has been waiting. "You're both still talking as though quality were sitting inside the product, waiting to be measured. It is not. Quality can be defined only in terms of the agent. So before anyone votes on whether it's good enough, answer the prior question: who is the judge of quality?" He lets it sit. "Quality should be aimed at the needs of the consumer, present and future. Not the consumer you have a contract with this morning. The one who decides next year whether you still have a market. The hard part was never counting the defects. The hard part is translating what a person will need, before they can say it, into something you can build."

Weinberg has the shortest answer in the room. "Quality is value to some person."

It sounds too simple to end the argument the other three have been having for 30 years.

"The work is in the last three words," Weinberg says. "Not value. Value to some person. And by value, I mean something you can pin down: what is a person willing to pay, or to do, to get their requirements met? Quality is not a property of the software. It is a relation between the software and somebody who does, or does not, get what they want out of it. Change the person, and you change the quality, without touching a line of code."

He turns to Crosby. "Your definition isn't wrong. It's incomplete, and the missing piece is a person. Conformance to requirements only moves the question down one level. Whose requirements? Written by whom, for whom? Requirements are written by people, for people, and in software, they are seldom even close to correct. A program that perfectly conforms to the wrong requirements is worth nothing to the person those requirements forgot. Your Cadillac can conform to every requirement of a Cadillac and still be the wrong car for the man who needed a truck. You did not define quality. You named one person and hid him: whoever wrote the spec."

Crosby does not concede. "Then fix the requirements. That's a discipline, not an excuse to call quality unknowable."

"I'm not calling it unknowable," Weinberg says. "I'm calling it relative. Watch." He goes around the room. "Fix the person to the one who wrote the spec, and you get Crosby: conformance to requirements. Fix the person to the one using it, and you get Juran: fitness for use. Fix the person to the consumer, present and future, and you get Deming. Not one of you is wrong. Each of you quietly chose a person, then called the choice the definition of quality. I'm only refusing to hide the choice."

Two of his own products make the relation cut both ways. A word processor that drops whole paragraphs from book-length documents is low-quality to the novelist who loses a chapter, and acceptable-quality to the manager weighing the thousands of customers who never write that long. A cribbage program that scores hands wrong, that counts eight where the right answer is four, was high quality to Weinberg, who paid the shareware fee and kept playing, because even broken, it gave him something he wanted. One product is full of defects, yet it's exactly what someone wanted. The other could be provably correct and worthless to the person it forgot. Conforming to requirements is neither necessary nor sufficient for quality.

Deming nods slowly, because he has said a version of this himself. "Who is the judge of quality. We agree on the question."

"We agree on the question," Weinberg says. "We part on what to do next. You want to translate the judge's needs into measurable characteristics, reduce the variation, and make the relation tractable. I want to say the part you will not. More quality for one person usually means less quality for another." Every quality decision spends a finite budget of attention, time, and money, and it spends it on somebody. The user who wants more features against the user who wants nothing to change. The customer who wants it now, against the customer who wants it bulletproof. There is no single dial setting that is best for everyone, because people want different things. "So every statement about quality is a statement about some person. And every is-it-good-enough is secretly a different question: whose opinion of quality counts when we decide, and how much, against everyone else's? That question is both political and emotional. It always was. You can measure your way around it, Ed. You cannot measure your way out of it."

Now, end the thought experiment and go back to the room with the forty defects.

The team was never going to settle it by gathering more data about the software. The software was not the disputed variable. The engineers had fixed the person to themselves: people who value an empty tracker. The head of sales had fixed the person to the three customers who value having the thing at all. Both were right. They were measuring value to different people and mistaking it for a disagreement about the code.

So the move is not to recount the defects. It is to ask the question the four of them spent their lives circling: whose value are we counting, and whose are we willing to sacrifice? List the people with a stake in it. Write down what each of them actually wants. The wants will contradict each other, and the contradiction is not a flaw in the list. It is the thing the quality argument was hiding the whole time. Once it is on the table, "is it good enough?" turns into "good enough for whom, and who decided that?" The first question nobody can answer. The second one a leader can.

Weinberg gave the room one more law for the question it refuses to ask. He called it the Zeroth Law of Quality: if you don't care about quality, you can meet any other requirement. Stop defining what good means, and every deadline becomes reachable, every budget achievable, because there is nothing left to fail against. That is what the defect count was really for. The number felt safe because it pretended the decision was made by the data, not by a person, about a person. It never was. Quality is value to some person. The only real choice the team had was whether anyone would say whose.