Training that sticks: spacing, retrieval and feedback

Why one session rarely holds

A common pattern in workplace training is the single event: a briefing or a course, a pass mark, a certificate and nothing after. When the same mistakes reappear months later, the usual conclusion is that training does not work. The research below suggests a narrower reading. What is often missing is not the content but the practice that follows it, and the check on whether it was retained.

Two findings explain why this matters. Roediger and Karpicke (2006) found that after a delay, students who had been tested on a passage remembered more than students who had restudied it the same number of times, even though the restudying group felt more confident about their memory.^[1] Karpicke and Roediger (2008) reported that students' predictions of their own performance were uncorrelated with how they actually did.^[2] Neither result says that people are poor judges of everything. They say that how confident a learner feels after a session is not a dependable measure of what will still be there later.

Dunlosky and colleagues (2013) reviewed ten learning techniques. Rereading and highlighting were rated low in utility because they did not consistently boost performance, while practice testing and distributed practice were rated high because they benefit learners of different ages and abilities.^[3] A slide deck read twice, with a signature at the end, sits closer to rereading than to either of the two techniques that rated highest.

Retrieval: build the questions in

In the Roediger and Karpicke experiments, students studied prose passages and then either took free-recall tests without feedback or restudied. On a test five minutes later, restudying did better. On tests after two days and after one week, the tested group did substantially better.^[1] The benefit appears when retention is measured later, which is the point of most training.

Karpicke and Roediger (2008) taught foreign-language vocabulary and varied what happened once an item had been recalled correctly. Repeated studying after success had no effect on delayed recall, and repeated testing had a large positive effect.^[2] The materials were word lists, so the study shows the mechanism clearly but says little about complex job tasks.

Pooled studies point the same way. Rowland (2014) meta-analysed experiments comparing testing with restudy and reported that initial recall tests produced larger testing benefits than recognition tests.^[4] Adesope and colleagues (2017) found practice tests more beneficial than restudying and than all other comparison conditions, with the size of the effect depending on the tests, participants, outcomes and methods.^[5]

A design reading, offered as inference and not as a finding: where a rule or a procedure must be recalled at work, questions that ask the person to produce the answer are better supported than questions that only ask them to recognise it from a list. A rule change, for example, can be followed by a short question that asks the person to state the new step, not by a second reading of the notice.

Spacing: the gap should fit the horizon

Cepeda and colleagues (2006) synthesised 839 assessments of distributed practice from 317 experiments in 184 articles. They compared massed with spaced study and compared shorter and longer gaps. Their analyses suggest that the gap between sessions and the time before the final test work together, and that the gap giving the best retention grew as the interval before the final test grew.^[6]

That finding cautions against a fixed formula. A gap that suits a test next week may not suit knowledge that must last for months. The review was of verbal recall tasks, so its evidence concerns remembering words and facts, not complex skills. Dunlosky and colleagues still rated distributed practice highly because it helped many kinds of learner across many tasks.^[3]

For a team, the practical question is how long the knowledge must last and how soon the next real use will come. A procedure used daily is practised by the work itself and may need less repetition than a rule used once a quarter, though the sources do not test that. As an illustration only, a rule that changes might be followed by a question the next day, another a week later and another a month later. The research does not specify those intervals. They are a starting point to adjust against what people actually forget.

Feedback: specific, and on real work

Hattie and Timperley (2007) described feedback as one of the major influences on learning but stressed that the type of feedback and the way it is given can be differentially effective.^[7] Wisniewski, Zierer and Hattie (2020) later meta-analysed 435 studies with 994 effects and more than 61,000 participants. They found a medium average effect (d = 0.48) and significant heterogeneity: feedback is not one consistent treatment. Impact depended on the information the feedback carried, and was higher for cognitive and motor skills than for motivational and behavioural outcomes.^[8]

The reading for a manager is that feedback works through its content. Telling someone that a task was done well carries less than showing which part of a listing description, a follow-up message or a filed form met the standard and which did not. This is an interpretation of the pattern across studies, not a quoted result. It also implies that the manager's time is part of the training, which a course on its own cannot supply.

Worked examples are a related tool. Barbieri and colleagues (2023) screened 8,033 abstracts and kept 43 articles reporting 55 studies and 181 effect sizes. The average effect of worked examples on mathematics performance was medium (g = 0.48). Studies using correct examples alone showed larger effects than those using incorrect examples alone or combined with correct ones, and adding self-explanation prompts moderated the effect negatively.^[9] The outcome was mathematics performance from elementary grades to postsecondary settings, so it should not be assumed to hold for every subject.

Putting the pieces together

Taken together, the sources support a sequence rather than an event: teach the material, ask people to recall it soon after, ask again after a longer gap, and give specific feedback on real work that uses it. None of the sources tests that sequence on employees, so it should be treated as a reasoned design and checked against the work itself. The guide on measuring whether training worked sets out how.

A manager with a calendar and a list of questions can run this for a small team. A learning management system can help by holding questions and sending reminders at set intervals, if it offers those functions. The Works LMS, which we make, issues a certificate of completion only when every assessment is passed, with a pass mark of 75 percent and two attempts per assessment, and a timed final exam drawn from a question bank. Those are properties of one product. They do not by themselves provide spacing or workplace feedback, which a programme has to arrange around them.

What the evidence does not show

Most of the studies cited here used university students, school-age learners or laboratory tasks. The abstracts read do not report trials with employees on job-related training, so applying the results to workplace learning is an inference. The Dunlosky review also notes that some promising techniques had not been adequately evaluated in educational settings. Effect sizes differ across test formats, populations and outcomes, so each number is an average. A sensible reading is that the principles are well supported for recall and comprehension, and that their effect on performance at work needs separate evidence. This page is general information and not training advice.^[3][9]

Questions