Improvement science is a disciplined, evidence-based approach to making systems work better. Rather than relying on intuition or one-time fixes, it uses small, repeated cycles of testing to figure out what actually changes outcomes in real-world settings like hospitals, schools, and public services. It borrows from manufacturing, statistics, and behavioral science, and its central premise is simple: every system is perfectly designed to produce the results it currently gets. If you want different results, you have to change the system.
How It Differs From Traditional Research
Traditional research, like a randomized controlled trial, typically asks “Does this treatment work under ideal conditions?” Improvement science asks a different question: “How do we make this work reliably, here, with the people and resources we actually have?” That distinction matters. Academic research often prioritizes questions of scientific interest, while improvement science prioritizes practical importance.
The two fields also think differently about expertise. Traditional research tends to rely on external experts who design an intervention and then push it outward to other settings. Improvement science flips this by emphasizing local knowledge. The people already working inside the system are treated as essential sources of insight, because they understand the daily friction points that no outside researcher would notice. This doesn’t mean improvement science is less rigorous. It simply applies rigor to a different goal: learning what works in practice and adapting until it does.
The Intellectual Foundation
Much of improvement science traces back to W. Edwards Deming, the statistician and management consultant whose ideas transformed Japanese manufacturing after World War II. Deming laid out what he called the System of Profound Knowledge, built on four interconnected pillars:
- Systems thinking: Understanding how different parts of an organization interact, so that fixing one area doesn’t accidentally break another.
- Theory of knowledge: Gaining understanding through structured experimentation rather than guesswork.
- Understanding variation: Recognizing the difference between normal fluctuations in performance and meaningful changes that signal a real shift.
- Psychology of change: Appreciating what motivates people, why they resist new ways of working, and what support they need to succeed.
All four pillars work together. You can redesign a process perfectly on paper, but if you ignore variation in how it plays out day to day, or if you neglect the human side of adoption, the improvement will fail. Deming’s framework insists you hold all four dimensions in view at the same time.
The Model for Improvement
The most widely used improvement science framework, developed by the Associates in Process Improvement and popularized by the Institute for Healthcare Improvement, centers on three questions:
- What are we trying to accomplish? This forces teams to define a clear, measurable aim rather than a vague aspiration like “improve patient safety.”
- How will we know that a change is an improvement? This requires selecting specific measures and tracking them over time, so the team doesn’t confuse activity with progress.
- What change can we make that will result in improvement? This pushes the team toward testable ideas rather than wholesale overhauls.
These questions can be addressed in any order, but all three must be answered before testing begins. They serve as a compass that keeps improvement work focused and accountable.
Plan-Do-Study-Act Cycles
Once a team has answered those three questions, the actual testing happens through Plan-Do-Study-Act (PDSA) cycles. Think of each cycle as a small, fast experiment.
In the Plan stage, you identify a specific goal and predict what you think will happen. In the Do stage, you run the test on a small scale, maybe with one patient, one classroom, or one shift. Keeping it small is intentional: if something goes wrong, you can adjust quickly without causing widespread disruption. In the Study stage, you compare what actually happened to what you predicted. Did the wait time drop? Did the error rate change? In the Act stage, you decide what to do next based on what you learned. You might adopt the change, modify it, or abandon it entirely and try something else.
The key difference between PDSA and conventional project management is that it’s cyclical, not linear. You don’t plan everything up front and then execute. You run through the cycle repeatedly, refining your approach with each pass. This iterative format is often compared to applying the scientific method in real time within a working system.
How Improvement Gets Measured
Improvement science relies heavily on tracking data over time rather than comparing a “before” snapshot to an “after” snapshot. The primary tool for this is a run chart: a simple graph that plots your measure (say, infection rate or wait time) along a timeline, with a line showing the median.
Run charts use four rules to distinguish normal fluctuation from a real signal of change. A “shift” is six or more consecutive data points all above or all below the median, suggesting the system has genuinely moved to a new level of performance. A “trend” is five or more consecutive points all heading in the same direction. An “astronomical point” is a single data point so dramatically different from the rest that anyone looking at the chart would agree something unusual happened. The fourth rule looks at how often the data crosses the median line: too few or too many crossings suggests a non-random pattern.
These rules matter because human beings are notoriously bad at distinguishing real change from noise. A single good week can feel like a breakthrough, and a single bad week can feel like a crisis. Run charts impose discipline on that interpretation, helping teams invest energy only when the data shows a genuine shift.
Diagnosing Root Causes
Before testing a change, improvement teams need to understand why the current system produces the results it does. Several practical tools help with this diagnosis.
The simplest is the “5 Whys” technique. You start with the problem and ask why it happened, then ask why again for each answer, continuing through five layers. For example: a lab worker was cut by a knife. Why? The knife was left by the sink. Why? The area wasn’t cleared the previous day. Why? Clearing isn’t a daily habit. Why? No standard operating procedure exists. By the fifth answer, you’ve moved from the surface event to a systemic gap that can actually be fixed.
For more complex problems, teams use a causal tree. The worst outcome (or near-miss) sits at the top, and the team maps out every contributing cause beneath it, branching downward until they reach root causes. The team then selects the two or three most important root causes for corrective action. This structured approach prevents knee-jerk reactions like firing off a memo after every incident, when often the right response is simply to monitor whether the problem recurs.
The Human Side of Change
Improvement science recognizes that changing a system means changing what people do every day, and that’s inherently difficult. Deming included psychology as one of his four foundational pillars for good reason: even a well-designed process will fail if the people using it don’t understand it, trust it, or feel supported in adopting it.
Research on change management across 16 different models identified five strategies that consistently appear: communicate about the change, involve stakeholders at every level of the organization, pay attention to organizational culture, align the change with the organization’s mission, and provide encouragement and incentives. Among these, the actions practitioners used most frequently were gaining leadership support, listening to feedback from frontline employees, and adjusting strategies based on that input.
This feedback loop mirrors the PDSA cycle itself. Just as you test a process change on a small scale and refine it, you also test how people respond to that change and adapt your approach to implementation. Skipping this step is one of the most common reasons improvement efforts stall. A team can have perfect data, a clear aim, and a proven change idea, but if they roll it out without listening to the people who have to execute it daily, adoption will be shallow and short-lived.
Where Improvement Science Is Used
Healthcare has been the most prominent adopter. Hospitals and health systems use improvement science to reduce infections, shorten emergency department wait times, improve medication safety, and decrease readmission rates. The Lean methodology, originally developed for manufacturing, has been adapted for healthcare settings to eliminate seven categories of waste: unnecessary transportation, excess inventory, wasted motion, waiting, over-processing, overproduction, and defects. In a hospital context, “waste” translates to anything that frustrates patients or adds burden without adding value to their care.
But the discipline extends well beyond medicine. School systems use improvement science to raise graduation rates and close achievement gaps. Government agencies apply it to service delivery. Nonprofits use it to improve program outcomes. The core logic is transferable: define what you’re trying to achieve, measure whether your changes are working, test ideas in small rapid cycles, and pay attention to both the data and the people.