Why Compliance Training Fails Without Human-Centered Design

Why Most GxP Training Changes Records, Not Behavior

Ask a site quality head how training is going and you will usually hear a completion rate. Ninety-eight percent on time. Two overdue assignments, both for people on leave. That number is real and it matters during an inspection. But it answers a different question from the one the business cares about, which is whether people perform the task correctly when nobody is watching.

The U.S. GMP regulations frame training around capability, not acknowledgment. Under 21 CFR 211.25, each person must have the education, training, and experience “to enable that person to perform the assigned functions,” and training must cover the particular operations the employee performs.¹ Nothing in that language asks for a signature. The signature became the default because it is easy to produce and easy to count. Our earlier article on training effectiveness beyond the read-and-understand signature covers what regulators expect as evidence of effectiveness. This article covers the step before that: why the usual training designs fail, and how to design training that has a chance of working.

Three Habits That Define Most GxP Training

Three habits account for most of the training hours in a typical pharma or biotech quality system.

• Read-and-sign. A controlled document is assigned to everyone in a role. The person opens it, scrolls to the end, and signs that they have read and understood it. For a long procedure, the assignment might take five minutes or an hour depending on how carefully it is read. The record looks the same either way.
• Retraining after every revision. When an SOP changes, the new version is assigned to everyone who had the old one, whatever the size of the change. A typo fix and a new critical step generate the same assignment.
• Long slide decks. Annual GMP refreshers, data integrity modules, and onboarding courses are delivered as presentations, live or recorded, followed by a short multiple-choice quiz that tests whether the person can recognize a sentence from the slides.

Each habit has a sensible origin. Read-and-sign ties people to the current document version, which document control needs. Retraining on revision keeps everyone on the same version. Slide decks are quick to build and easy to audit. The problem is that each one was designed for the record, not for the learner. None of them asks what the person needs to do differently on Monday morning.

When Training Volume Teaches People to Skim

There is a second-order effect that leaders often miss. When a technician receives dozens of read-and-sign assignments a quarter, most of them for minor changes that do not affect their work, they learn a reasonable lesson: these assignments are administrative, and the fastest way through them is to click to the end and sign. That habit does not stay with the minor changes. It applies just as much to the one revision that adds a new critical step, which arrives in the same queue, looks the same, and gets the same treatment.

In other words, high training volume does not just waste time. It teaches people to stop reading. A system that sends everything with equal urgency has no way to show that one item matters more than the others. Human-centered design starts by fixing that.

What Learning Research Says About Reading, Recall, and Practice

The research on how adults learn and retain skills is large, and much of it was done outside pharma: in schools, universities, safety programs, and corporate training. That matters, and we will come back to the limits. But the main findings are consistent across settings, and they point in the same direction for GxP training.

Rereading Is a Weak Way to Learn

In 2013, a team of cognitive and educational psychologists led by John Dunlosky reviewed ten common learning techniques and rated how well each one works across learners, materials, and tasks.³ Two techniques received a high utility rating: practice testing and spreading study over time. Rereading and highlighting were rated low. The authors chose to examine rereading and highlighting partly because learners rely on them so heavily, which made it important to check whether they work.

Read-and-sign is rereading by design. A person reads a procedure, sometimes for the fifth or sixth time across revisions, and is then assumed to have learned it. The research suggests that this is one of the least effective ways to build durable knowledge, even before you account for the scroll-and-sign behavior described above.

Recalling Beats Reviewing, and Learners Cannot Tell the Difference

A well-known experiment by Henry Roediger and Jeffrey Karpicke compared students who restudied prose passages with students who took recall tests on the same passages.⁴ When the final test came five minutes later, the students who restudied did better. When it came two days or a week later, the students who had been tested remembered substantially more. The detail that matters most for compliance training is this: repeated studying increased students’ confidence in their ability to remember the material, even though it produced worse long-term retention.

That finding describes the read-and-sign experience closely. Reading a familiar procedure feels like knowing it. The person signs in good faith. The feeling of fluency is real, but it is a poor guide to what the person will recall at the moment of use, weeks later, under time pressure.

Passive Formats Lose to Active Ones

The strongest workplace evidence comes from safety training, which shares many features with GMP training: rule-based tasks, serious consequences, and a long history of classroom delivery. A 2006 meta-analysis by Michael Burke and colleagues pooled 95 studies covering 20,991 participants and grouped training methods by how much they required the learner to participate.² The least engaging methods were lectures, pamphlets, and videos. The most engaging were behavioral modeling and hands-on training.

The most engaging methods were, on average, about three times more effective at building safety knowledge than the least engaging ones. In five studies of low-engagement training that measured knowledge again later, the average effect fell by about half over periods from one week to one year. The authors concluded that training combining behavioral modeling, substantial practice, and dialogue is generally more effective than other methods, and that their findings challenge the emphasis on more passive computer-based and distance training.

Two points of fairness belong here. First, every method in the Burke review produced meaningful improvement in on-the-job behavior; passive training is not worthless. Second, the gap between methods was much larger for knowledge than for measured safety performance. The honest reading is that engagement matters most for what people know and keep, and that behavior on the job also depends on everything around the training.

The same direction shows up in education. A meta-analysis of 225 studies in undergraduate science, engineering, and math courses by Scott Freeman and colleagues found that examination scores improved by about 6 percent in active learning sections, and that students in traditional lecture courses were 1.5 times more likely to fail than students in courses with active learning.⁵ Average failure rates were 21.8 percent under active learning and 33.8 percent under lecturing.

What Happens Before and After Training Matters as Much as the Session

A 2012 review of organizational training research by Eduardo Salas and colleagues makes a point that is easy to forget when the training group owns only the session: what happens before and after training matters as much as what happens during it.⁷ Before training, the authors recommend creating a suitable learning climate and conducting a training needs analysis, which they describe as a diagnosis of what needs to be trained, to whom, and in what organizational system. After training, they recommend checking that transfer to the job took place and evaluating the methods so they can be improved.

Read that list against a typical SOP retraining cycle. There is usually no needs analysis: the assignment goes to whoever had the previous version. There is no check of transfer: the signature closes the record. The session in the middle is the only part that exists, and it is the weakest part.

The Limits of This Evidence

Much of the learning research was done with students, and the safety training studies come from a range of industries. None of it was done on aseptic gowning or batch record review. Leaders should not take any single effect size and promise it for their own site. What the evidence does support is a direction: active, practice-based training with feedback outperforms passive reading and listening, and the gap is largest for knowledge that people need to retain over time. That is enough to justify redesigning the highest-risk training first and measuring the result.

Human Error Is Usually a Design Problem Before It Is a Training Problem

Before redesigning any training, it helps to ask whether training is the right fix at all. A large share of training in GxP companies is triggered by deviations attributed to human error, with retraining assigned as the corrective action. That habit assumes the person lacked knowledge. Often they did not.

The Person Approach and the System Approach

James Reason’s 2000 article on human error, written for clinicians but widely used in other high-risk industries, describes two ways of looking at error.⁸ The person approach focuses on the individual’s lapses: forgetfulness, inattention, weak motivation. The system approach starts from the premise that people make mistakes and asks what conditions made the mistake likely. Reason summarized the system view in one line: “though we cannot change the human condition, we can change the conditions under which humans work.”

Retraining after a human error deviation is a person-approach fix. It sometimes works, when the cause was a gap in knowledge or skill. It does nothing when the cause was a confusing batch record layout, two similar-looking labels stored side by side, a procedure step that cannot be done in the order written, or a shift pattern that puts the hardest step at the end of a twelve-hour day.

What FDA Has Said About the Conditions Behind Human Error

FDA has started to say this plainly in warning letters to drug manufacturers. In a May 27, 2026 warning letter to a sterile drug manufacturer, Pharmathen International S.A., FDA wrote that the firm had not implemented effective CAPA to reduce recurrence of contamination hazards, including operator error that had contributed significantly to recurring sterility and media-fill failures over five years.⁹ The agency’s example of the gap: “your investigation did not include in-depth analysis of the conditions that create the circumstances for human error.” FDA went on to call for in-depth root-cause analysis “to identify effective CAPA including design remediations.”

The phrase “design remediations” is the important one. The agency is not asking for more training. It is asking the firm to change the conditions that produce the errors.

The UK Health and Safety Executive makes a related point in its human factors guidance. On procedures, it states that “Procedures do not replace competency,” and recommends using task analysis, including walking and talking through the task with the people who do it, to decide what a procedure should contain.¹⁰ On competence, it states that “Training is an important component of establishing competency but is not sufficient on its own,” and notes that consolidating knowledge and skills through practice is a key part of developing competency.¹¹

Four Gaps, Four Different Fixes

A simple sorting step before any training request saves a great deal of wasted effort. When a deviation or a performance problem points to people, ask which of four gaps is behind it.

Only the first two gaps are training problems. In our experience, a review of a site’s recent human error deviations often finds a substantial share that belong in the design or belief columns. Moving those out of the training queue reduces training volume and puts effort where it will change the outcome.

Human-Centered Compliance Training Starts With the Real Task

Human-centered design is a well-established approach in software and product design. ISO 9241-210:2019 sets out requirements and recommendations for human-centered design principles and activities across the life cycle of computer-based interactive systems.¹² The standard was written for interactive systems, not for training, and nobody should claim conformance to it for a GMP course. But the habits behind it apply directly to training. Design from an understanding of the people who will use the thing and the work they do. Involve them. Test early versions with them. Change the design based on what you see. Repeat.

Applied to compliance training, human-centered compliance training means designing for the person doing the task, in the place and conditions where they do it, rather than for the document or the training record. It has four parts, and the rest of this article takes them in turn:

1. Understand the real task before writing anything.
2. Write short content for one role and one task at a time.
3. Replace reading with practice, scenarios, and recall.
4. Measure behavior and error rates, not completion.

Running through all four is a fifth concern: how people respond to being told to change, which we cover in its own section.

The Procedure Describes the Task as Someone Imagined It

An SOP is written by an author, usually a subject matter expert or a quality specialist, who describes how the task should be done. The people who do the task every day often do it somewhat differently. Sometimes the difference is a harmless shortcut. Sometimes it is a workaround for a step that cannot be done as written. Sometimes it is a real risk that nobody in quality knows about. Training built only from the SOP teaches the imagined version. Training built from observation teaches the version people will meet on the floor, and it often shows where the SOP itself needs to change.

Salas and colleagues put the training needs analysis first for this reason.⁷ HSE’s recommendation to walk and talk through the task with users makes the same point from the procedure side.¹⁰ You cannot design training for a task you have not watched.

A Five-Step Task Discovery Routine

This routine does not need a large team. A trainer and a quality engineer can complete it for one task in a few days. The output is a short list of critical steps, decisions, and known failure points, which becomes the starting point for the training content.

Short, Role-Specific Content

Once the real task is understood, the content can be much shorter than the SOP. That is not a shortcut. The SOP remains the controlled reference. The training material has a different job: to prepare a specific person to perform specific steps correctly.

Write for One Role and One Task

Most GxP training modules try to serve everyone who touches a process. A data integrity module written for operators, analysts, reviewers, and managers ends up general enough that none of them sees their own work in it. A module for QC analysts reviewing chromatography data can show the exact screens they use, the audit trail entries they should look at, and the specific patterns that should prompt a question. The same principle applies to clinical operations: a module on informed consent documentation for site monitors should look different from one for study start-up staff.

HSE’s guidance on procedures recommends a format, style, and level of detail appropriate to the user, the task, and the consequences of failure.¹⁰ The same test applies to training. If a section of the training would read the same for every role, it probably does not belong in role-specific training.

Cut Everything Not Tied to an Action or Decision

A useful editing rule: every part of a training module should connect to something the person does or decides. Regulatory history, the full scope statement, the list of related documents, and definitions of terms the person already uses every day can stay in the SOP. If the training material cannot point to the step or decision that a paragraph supports, cut the paragraph.

This produces modules that are much shorter than the courses they replace. Short formats delivered close to the point of use, sometimes called microlearning and just-in-time learning, are a related topic with their own design questions, and we will cover them separately. The point here is narrower: content becomes shorter when it is built around the task rather than around the document.

Separate What Changed From How to Do It

For SOP revisions, human-centered design changes the retraining question. Instead of “who had the old version,” ask “whose task changed, and what do they need to do differently.” For a minor editorial change, the answer may be nobody, and the document system’s version control is enough. For a change to a critical step, the affected people need a short explanation of what changed and why, followed by practice of the new step if it involves a skill. Our article on LMS curricula covers how to classify revisions and route assignments; the design point is that a person who receives fewer, more meaningful assignments is more likely to take each one seriously.

Practice and Scenarios Instead of Reading

The largest single change in human-centered compliance training is replacing reading and listening with doing. The evidence from the Burke review points to behavioral modeling, substantial practice, and dialogue as the combination that works best.² Each of those can be built into GxP training without new technology.

Show, Then Practice, Then Discuss

Behavioral modeling means watching a competent person perform the task, including the critical steps and the reasoning behind decisions. For a gowning procedure, that is a demonstration by an experienced operator who talks through what they check and why. For a data review task, it is a reviewer working through a real (anonymized or training) dataset on screen and explaining what made them stop and look closer.

Practice means the learner does the task, in a training environment, with feedback. For physical tasks, that is a mock-up line, a training cleanroom session, or a supervised run. For data and documentation tasks, it is a training instance of the system with prepared records. Dialogue means a short conversation afterward about what went well, what was hard, and what the person would do in situations that were not covered.

Build Realistic Errors Into Simulated Practice

Many GxP trainers avoid letting learners make mistakes during training, out of a reasonable instinct that the goal is error-free work. The research suggests the opposite approach works better for transfer. A meta-analysis of error management training by Nina Keith and Michael Frese pooled 24 studies with 2,183 participants.⁶ Error management training combines active exploration with explicit encouragement to make errors during training and learn from them. Compared with error-avoidant training, it produced a positive overall effect (d = 0.44), a larger effect on performance after training (d = 0.56), and the largest effect on tasks that differed from what was trained (d = 0.80). The authors concluded it may be better suited than error-avoidant methods for helping people handle new situations.

The last finding matters a great deal in GxP work, where the risky moments are usually the unusual ones: an alarm that does not match the troubleshooting guide, a record that looks almost right, a sample that arrives outside its window. In practice this means planting realistic problems in practice material. A training batch record with a missing second-person verification. A chromatogram with an integration that should have been questioned. A label that is close to correct but carries the wrong lot. The learner finds the problem, or misses it and sees the consequence, and the trainer discusses what to look for. All of this happens in simulation, never on product or in live GxP records.

Use Recall Questions, Not Recognition Quizzes

The ten-question multiple-choice quiz at the end of a slide deck mainly tests whether the person can recognize a sentence they saw a few minutes earlier. Roediger and Karpicke’s results suggest that retrieval practice has value beyond assessment: testing is a way of improving learning, not just measuring it.⁴ Better checks ask the learner to produce an answer from memory, apply it to a situation, or make a decision:

• “List the three checks you perform before starting line clearance.”
• “This audit trail shows a result reprocessed twice after the run. What do you do next?”
• “A shipper arrives with the temperature logger showing a brief excursion. Who decides, and what information do they need from you?”

Dunlosky’s review also rated spreading practice over time as high utility.³ A few recall questions on critical steps, repeated at intervals after the initial training, are a simple way to apply both findings.

Scenarios for Judgment-Heavy Roles

Not every GxP role has a physical task to practice. Quality reviewers, clinical research associates, pharmacovigilance case processors, and regulatory staff mostly make judgments. For these roles, scenarios are the practice. A good scenario is drawn from real (anonymized) events at the company, includes the ambiguous details that make the real decision hard, asks the learner to choose and justify a course of action, and shows what happened or what an experienced reviewer would do. Short group discussions of scenarios also give supervisors a way to see how their staff reason, which a quiz score never shows.

Working With Change Resistance

Even well-designed training meets resistance. Compliance training is almost always mandatory, often arrives with a deadline, and frequently asks people to change a habit they believe works. Behavioral research has useful things to say about why that produces pushback, and about what reduces it.

Reactance: Why Mandates Invite Pushback

Psychological reactance theory, first proposed by Jack Brehm in 1966, describes a predictable response to perceived loss of freedom. A 2015 review by Christina Steindl and colleagues defines reactance as “the motivation to regain a freedom after it has been lost or threatened,” and notes that it leads people to resist the social influence of others.¹³ In a GxP setting, the freedom at stake is usually the freedom to do one’s own job in the way one has learned to do it. A new requirement presented as a directive, with no explanation and a completion deadline, is a classic trigger.

Reactance does not usually show up as open refusal in a regulated workplace. It shows up as minimum compliance: the signature without the reading, the step performed when someone is watching, the complaint in the break room that quality keeps changing things for no reason.

Individual Differences in Resistance

People differ in how strongly they resist change. Shaul Oreg’s Resistance to Change Scale, developed and validated across a series of studies, identified four factors behind a person’s disposition to resist change: routine seeking, emotional reaction to imposed change, cognitive rigidity, and short-term focus.¹⁴ Leaders do not need to measure these in their staff. The four factors are useful as a design checklist, because each one suggests a different response.

Autonomy, Competence, and a Reason Why

Self-determination theory, developed by Richard Ryan and Edward Deci, proposes three basic psychological needs: competence, autonomy, and relatedness.¹⁵ In their 2000 summary of the research, Ryan and Deci report that threats, deadlines, directives, pressured evaluations, and imposed goals diminish intrinsic motivation, while choice, acknowledgment of feelings, and opportunities for self-direction enhance it. Their work on how people take in rules they did not choose finds that people internalize a requirement more fully when they feel competent to meet it, connected to the people asking for it, and able to see its value.

Compliance training cannot remove every deadline or directive. Some requirements are not negotiable. But the design can change how the requirement is presented, and the research gives specific guidance:

• Give the reason. Every critical step in training should come with the reason it exists, ideally a real event. “This check exists because two years ago a mislabeled carton reached a distributor” is more persuasive than “per SOP-QA-0147.”
• Offer choices where they exist. When to complete training, which session to attend, in what order to complete modules. Small choices help.
• Acknowledge the effort. Saying plainly that a change adds work, and why the company thinks it is worth it, reduces the sense of being managed.
• Build competence before assessment. People who practice before being tested feel capable, and people who feel capable are less defensive.
• Use supervisors and peers. Salas and colleagues list creating a suitable learning climate as one of the steps to take before training begins.⁷ Supervisors shape that climate more than anyone. A supervisor who treats training as paperwork teaches the team to do the same. Experienced operators who helped design the training are often the most credible people to deliver it.

Outcomes Leaders Can Track

None of this is worth doing unless it changes outcomes, and outcomes have to be measured. Our earlier article on training effectiveness covers the kinds of evidence that demonstrate effectiveness to an inspector, including structured observation and task qualification. This section focuses on a narrower question: which numbers a quality, operations, or clinical leader should watch to know whether a training redesign is working.

Move From Activity Metrics to Outcome Metrics

Completion rate and training hours measure activity. They are needed for compliance reporting, but they cannot tell you whether a redesign worked, because a redesign aims for the same or higher completion with fewer hours. The useful measures are closer to the task.

These are suggestions, not a fixed standard. Pick three or four that fit the task you are redesigning. Two of them should be outcome measures (error rate, right-first-time, or repeat deviations) and one should be an input measure that shows whether unneeded assignments are falling (assignment volume or retraining-only CAPA share). Our article on behavioral KPIs explains why measures of what people do tell you more than measures of what was delivered.

Baseline First, Then Redesign

The most common measurement failure is starting the redesign before capturing the baseline. Pull at least twelve months of deviation history for the task, run a month of structured observation using the old training, and record the current assignment volume. Without that, any later improvement is an anecdote.

Run a Contained Pilot

A practical sequence for a first redesign:

1. Months one and two: select one high-risk task with a deviation history. Capture the baseline. Run the five-step task discovery routine. Sort past deviations into the four gaps and send design gaps to change control.
2. Months two and three: build the new training with the operators who helped in discovery. Test it with a small group, observe, and revise.
3. Months three through six: deliver the new training to the affected role. Track the chosen metrics monthly. Keep the SOP read-and-sign for document awareness.
4. Month six: compare against baseline. Present the result to quality leadership with the effort involved, and decide which task to redesign next.

Be realistic about what a single pilot can show. Deviation counts on one task are often small, and a few months of data will not support strong statistical claims. Observed adherence and right-first-time rates usually move faster and give a clearer early result. The goal of the pilot is to show the organization what the new approach looks like and whether the direction is right, not to prove a precise effect size.

Keep the Compliance Record Intact

Redesigning training does not mean abandoning the training record. The training system should still record who was trained on what, when, and on which document version, because document control and inspection readiness depend on it. What changes is what the record contains: a practice session and an observed assessment instead of a signature alone. Make sure the training SOP describes the new methods before the pilot starts, so the redesign itself is under change control and the new records are recognized as valid training evidence.

Conclusion

Compliance training fails for predictable reasons. It relies on reading when recall and practice work better. It goes to everyone when only some people’s tasks changed. It treats human error as a knowledge problem when the cause is often the design of the work. And it presents requirements as directives, which invites the minimum effort that a signature allows. Human-centered compliance training reverses each of these: it starts from the task as people do it, writes short content for one role, builds practice and realistic errors into training, gives people the reason behind each critical step, and measures error rates and observed performance instead of completion.

The first step does not need a new platform or a large program. It needs one high-risk task, a baseline, a few days of watching the work, and the willingness to fix the design gaps before writing a single slide. Sakara Digital works with pharma and biotech organizations on quality systems and the people who run them. If you are rethinking how your teams are trained and want an independent view on where to start, we are happy to have that conversation.

For Further Reading