In this guide
Make a small change a learning task
  1. 01Understand the problem
  2. 02Predict and test
  3. 03Study observations
  4. 04Adapt or make routine
A practical PDSA-informed view. A small before-and-after difference does not automatically establish causation.

Continuous improvement is a repeatable way to examine a problem, try a bounded change, study what happens and decide what to keep or investigate next. Small changes still need a clear question and a review. Activity alone does not show that work improved.

Use this approach when the process is sufficiently understood for a local test and the people doing the work can participate. If the intended service or organizational purpose itself is wrong, a series of tiny adjustments may be inadequate. Double-loop learning helps distinguish improving a routine from questioning the rule behind it.

Start with a problem people can recognize

Describe the difficulty, who encounters it and what happens during the work. “The team is inefficient” is a judgement with no clear observation behind it. “Returned books wait on a trolley while staff repeatedly sort them into different routes” identifies something the team can examine.

In this hypothetical library, staff suspect that separating some returns earlier would reduce repeated handling. Before changing the trolley, they need to understand which items require checking, which can go directly to shelves and which must follow another route. A current workaround may be protecting a necessary condition.

The Institute for Healthcare Improvement’s PDSA guidance describes planning, trying, studying and adapting a local change. This guide applies that learning discipline to ordinary workplace improvement. It does not claim that a library test is a clinical intervention or that any change labelled PDSA is well designed.

Describe the problem and the current work

Observe the route from the return desk to the next usable state. Record where an item waits, which checks happen and who handles exceptions. Ask staff to explain what the formal procedure leaves out. Include busy and quieter conditions where they materially affect the task.

A library colleague examines a crowded return trolley while other staff shelve books.
Figure 1. Describe the problem and the current work.Illustration credits

Do not collect more detail than the decision needs, but establish a usable starting point. If the team wants to reduce repeated handling, define what a handling event means and how it will be recorded. If the concern is delay, define the beginning and end of the interval.

Keep the process boundary visible. A faster return desk is not an improvement if unresolved work accumulates elsewhere. Interface management can help when the proposed change moves a check or responsibility between roles.

Write a problem statement that leaves room for alternatives: which items are handled repeatedly, under what circumstances and with what consequence? Avoid embedding a preferred trolley design in the statement before investigating the work.

Plan a test with a prediction

Choose one change and state why it might help. The fictional team might try a clearly separated area for items needing a further check, so staff do not repeatedly examine them while shelving ordinary returns. That is a hypothesis about the work, not an established solution.

Use a compact test record:

FieldWhat to agree
QuestionWhat should this test help us understand?
ChangeWhat exactly will be done differently?
PredictionWhat do we expect to observe, and why?
BoundaryWhich people, period and item types are included?
ObservationWhat will be recorded and by whom?
SafeguardHow will ordinary service and exceptions be supported?
ReviewWhen will someone decide what follows?

IHI’s PDSA worksheet explanation emphasizes documenting an individual test and the learning across cycles. The table above is an original adaptation, not a copy of the worksheet. It can be used without specialized software.

Make the first test small enough to understand and support. Small does not mean careless or invisible. Tell the relevant colleagues what is changing, keep an available recovery route and avoid making several unrelated changes whose effects cannot be distinguished.

Choose measures that show more than speed

IHI’s measurement guidance distinguishes outcome, process and balancing measures. In the fictional library, time until an item is ready for its next destination could describe an outcome; repeated handling could describe part of the process; misplaced items or extra checking effort could reveal side effects.

Select measures according to the purpose rather than insisting that every small test needs a large dashboard. A brief observation record can be useful if its meaning and limitations are clear. Qualitative accounts may explain a difficulty that a total conceals.

Write definitions before comparing results. Does the count include all returned items or only ordinary books? Are unresolved exceptions excluded? Does the person recording change their interpretation during the test? These are measurement-quality questions, not details to postpone until the chart is complete.

A simple before-and-after difference does not establish that the change caused the outcome. Staffing, demand and case mix may also change. Use experiment design when the decision requires a stronger causal comparison or a larger commitment.

Study the test before making it routine

Compare what happened with the prediction. Ask which observations fit the proposed mechanism, which do not and whether another explanation is plausible. Include interruptions, exceptions and practical burdens, rather than discarding them as noise because they make the story less tidy.

Two library colleagues compare the arrangement of books on two return trolleys.
Figure 2. Study the test before making it routine.Illustration credits

For the fictional trolley, the important discussion might concern whether staff can reliably distinguish the routes at the return desk. If they cannot, the problem may be the information available at that moment rather than the physical layout. A mixed result should lead to a more specific question.

Decide whether to adapt, repeat under different conditions, expand cautiously or abandon the proposal. Pivot-or-persevere guidance helps separate a disappointing result from a reasoned decision about what remains worth testing. No result is reported for this imagined library.

Taylor and colleagues’ systematic review of PDSA application in healthcare found uneven use and reporting of methodological features. The lesson for this guide is modest: naming a cycle is different from carrying out and documenting its learning. The review does not estimate the benefit of the fictional library arrangement.

Make a useful change maintainable

When a change appears worth adopting within its tested boundary, agree the working method, owner and conditions for review. Show colleagues how to perform the ordinary task and handle exceptions. Preserve the reason for the arrangement so later staff can distinguish an essential step from a historical habit.

Use learning transfer to connect explanation with practice during real work. Training attendance does not establish that people can use a new arrangement under pressure. Give them access to the necessary tools and support.

Do not use standardization to close further inquiry. Record which conditions were examined and which were not. A change that works on a quiet shift may need another test before wider use. Revisit the method when demand, technology or service requirements change.

Boundaries and the next cycle

Continuous improvement can be misused as a demand for endless cost reduction or as a way to move more work onto people with little authority. Keep service quality, workload and the purpose of the work in view. Some problems require additional capacity, a different policy or an organization design decision rather than another local experiment.

The example and test record are original practical adaptations. The sources establish improvement concepts and their limitations; they do not validate this exact routine or promise a quantified gain. Begin with one observable problem, one proposed mechanism and a review someone owns. The next cycle should follow the learning, not the obligation to report that a change was successful.

Sources: