Take the statement, “be sure that the report is accurate.” To this it might seem reasonable. But when two people begin evaluating a report, one checks the totals, one checks spelling, and one checks that all sections have been included. Each person thinks he or she is ensuring quality, but that quality requirement does not tell them what to evaluate or what results they should expect. Measurable acceptance criteria translate this kind of quality expectation into conditions that are objectively observable, verifiable, and documented.
Start by identifying the subject of the evaluation and its objective. The requirement reads, “Customer orders need to be processed fast.” The subject here is the order-processing process, but fast is not defined. A more complete requirement might state that a complete order must be input to the system within 30 minutes of being received during normal business hours. Now we have input, the boundary of the process, the output, and a quantified time expectation. The person responsible for the process also has a tangible target to focus on instead of simply trying to work faster.
Good acceptance criteria will address several related points. What is the subject under evaluation? Which of its properties is relevant? What is the acceptable range (range of values, specific condition, tolerance, etc.)? How will you determine this? The requirement that states that “the label must be correct” could become something that says that “the product code, batch number, and expiry date appearing on the label must correspond with the approved production record, all fields are legible and all fields fall within the labeled space.” The quality control checklist might be expanded to reflect these three criteria (correctness, legibility, location). The reviewer does not need to guess what the definition of correct is.
Quantities are nice but not everything that is measurable needs to be quantitative. Some conditions may be observed directly as a pass or fail condition. A piece of work may be judged as adequate if it lists the sequence of steps, a list of actors, a list of required records, and a final sign-off. An incident report may be judged to be complete if it lists the type of failure, the location of the failure, the date of occurrence, the severity, and an observation to support this. This is measurable because if two different people evaluate the same piece of work in question, they should reach a similar conclusion using these criteria. Use quantified language to describe terms like appropriate, acceptable, clean, or good quality, unless the document also explains what they mean in context.
As an exercise, pick a quality expectation you have used to describe something that you know very well. Write it in the top of a page, then explain what you would need to see to be certain that the expectation was met. The statement is “deliver a finished meeting summary.” What are the key components? When is it due? Where does it go? How is the person in charge of an assigned action known? Rewrite the expectation using what you wrote above. Now, find someone else to evaluate the expectation against some actual sample work they have created. If the two of you disagree, there is something in your definition that you have yet to fully resolve.
Avoid defining so many details that evaluating the item you are looking at becomes more difficult than completing the original task. Each of your criteria should have a clear quality benefit. Measuring 10 formatting criteria in a document may be irrelevant if your customers care mostly for accurate technical information. And do not lump multiple criteria into a single point. “The form must be complete and correct” obscures what are at least several different quality issues that may cause this check to fail. Distinguish between completeness, data accuracy, approval, and traceability when all of these conditions could each fail on their own.
The ultimate test for the quality you just defined is not whether it sounds more official. It is whether the expectation you just defined can help someone consistently evaluate the item in question and make a clear conclusion. Read the wording you just developed with the question in mind of “Can someone new in the organization read this expectation, find the relevant data, make their own determination as to whether the expectation has been met, all without asking someone for additional details?” If the answer is “yes,” you have taken an expectation about quality and converted it into a measurable quality acceptance criterion.