Spaced Repetition: How to Schedule Reviews in Training

Go from storyboard to an interactive course in minutes

Start Free

A team does its annual refresher in February. Everyone passes the quiz at the end, the completion report goes to the manager, and the next time anyone thinks about the spill procedure is when something spills in September. The course taught the steps once, and nothing brought them back in the seven months between.

Spaced repetition addresses that gap. At work, it becomes a scheduling job for a whole team. I'm an instructional designer at Coursebox, where we make an AI quiz generator and course tools, so I'll say where our product fits and where it does not. This guide is about the workplace version of the question: how far apart to set reviews when people must remember something for a week, a quarter or a whole year, and how to make those reviews happen. I cover what spaced repetition is, how to choose the gap, the 2357 method, and a full year of reviews for one course, with three real cases from emergency and hospital training, a product test and a review calendar you can download.

What is spaced repetition?

Spaced repetition means returning to the same material in separate study or practice sessions, rather than doing all the repetitions together.

Spaced practice and distributed practice describe repetitions separated in time. Spaced retrieval is the version in which learners try to recall an answer before seeing it. Spacing and retrieval are separate choices: rereading can be spaced too. This guide combines them because answering a question or doing a task makes the review more useful than simply reopening the lesson. The opposite of spaced practice is massed practice, with repetitions packed together.

The evidence for spaced repetition

Repeating something five times in one sitting is massed practice, however many repetitions it has. It also feels effective, because it works for a test the same day. Dunlosky and colleagues' 2013 review of study techniques found that the benefit of spacing "is often stronger on delayed tests than immediate ones, with massed practice (cramming) actually benefitting performance on immediate tests" (p. 38). A training course that ends with a quiz straight after the content is measuring the thing cramming is good at.

The idea is old. In 1885 Hermann Ebbinghaus reported that, for relearning a 12-syllable series, 38 repetitions spread over the three days before matched the effect of 68 repetitions made the day before. He was the only subject, he used nonsense syllables, and he said the result was "found here for only very limited conditions". Spacing has been studied in hundreds of experiments since, and our guide to cognitive learning theory examples covers why it works and why recall beats rereading, with a spaced retrieval planner for writing the questions themselves. This guide is about the schedule: when each review happens, and who makes it happen.

For the workplace schedule in this guide, combine two design choices:

• There is a gap. The same content returns in a separate session. Here the gaps are days or weeks; research also studies shorter gaps.
• It asks for recall. Learners attempt a question or task, then receive feedback. Keep any job aid that the real task permits. Kang's 2016 policy review sums up the research as showing that "incorporating tests into spaced practice amplifies the benefits".

Spaced repetition apps and where their schedules come from

Many published spaced repetition schedules come from flashcard systems. In 1972 Sebastian Leitner described a box system in So lernt man lernen. Cards you answer correctly move to a box you review less often, and cards you miss move back to a box you review more often. Versions differ on how far a missed card drops and how long each box waits, so treat any "1, 2, 4, 8 days" list as someone's variant, not Leitner's rule.

Piotr Wozniak turned the same idea into a formula. His SM-2 algorithm, first used in December 1987, sets the first interval at 1 day, the second at 6, and multiplies each later interval by an "E-Factor" that starts at 2.5 and falls for harder items, never below 1.3. With the factor left at 2.5, the formula gives reviews 1, 6, 15, 38 and 95 days apart, rounding each interval up before calculating the next. Wozniak built it by "trial-and-error" on his own learning. Anki's manual says its scheduler "is based on an older version of the SuperMemo algorithm called SM-2", and the app also offers an alternative scheduler called FSRS. If you are choosing between flashcard apps for your own study, our Quizlet vs Anki comparison covers that.

Two cautions carry over to training. An app schedules each card for each person, while a training course usually schedules a whole group on the same dates. And the evidence behind growing intervals is thinner than the software suggests: Cepeda and colleagues' 2006 review notes that the lack of data "has not stopped some software developers from assuming that expanding study intervals work better than fixed intervals" (p. 370).

How far apart should spaced reviews be?

The gap that works best depends on how long people need to remember the material. Knowledge needed next week wants a short gap. Knowledge needed in eleven months wants a much longer one.

The clearest test of this is a 2008 study of review timing by Cepeda, Vul, Rohrer, Wixted and Pashler in

Those are interpolated figures from the study's curves (p. 1099). The best gap gets longer as the test gets further away, but it shrinks as a share of the wait: about 43% of a one-week delay, and about 8% of a delay of almost a year. For a fixed amount of study time, the best gap gave a 64% relative increase in final recall compared with a zero-day gap, where the review comes straight after learning. Cepeda's earlier review of distributed practice in verbal recall found the same pattern: "at retention intervals of 6 months or more, ISIs of at least 1 month maximized retention" (p. 366), where an ISI is the gap between study sessions.

In that experiment, recall declined more gradually beyond the estimated best gap than it rose before it. That is useful when planning for a distant test; it is not a reason to postpone an early check of something people must use tomorrow.

Here is a planning guide for workplace content, with the study results kept separate from suggested schedules. Its test delays were measured from the review, not from the original lesson. They tell us about recall on a later test, not whether a person could perform safely on every intervening day.

The study used trivia facts learned online, one review and one final test. It did not test procedures, physical skills or a year of repeated reviews. The proposed workplace timings need checking against the task, prior knowledge and delayed performance. If staff need the material immediately, teach it and check it before use, then space later practice.

An annual compliance course sits in the year row, where the choice is one long refresher or short sessions spread across the year. An ambulance service in Germany tested exactly that choice.

Case study: Paramedics practising monthly instead of yearly in Germany

Resuscitation skills fade between refreshers, and emergency staff have little time to train. A team from University Hospital Tuebingen, working with the German Red Cross ambulance service in Reutlingen, tested whether short, frequent sessions could replace the usual annual training day for paramedics and emergency medical technicians.

• What they did: 35 staff were randomly split into two groups. One group had about two hours of guideline teaching and scenario practice at the start, then a single cardiac arrest scenario every four weeks, lasting 10 to 15 minutes with brief feedback from the trainers. The other group had one full day of training of the kind given on the first day of a standard advanced life support course. Both groups were filmed running a simulated cardiac arrest at the start and 12 months later, and six experts who did not know the groups rated the videos.
• Evidence and result: In Moll and colleagues' 2026 trial in BMC Medical Education, the monthly group scored 4.7 out of 5 overall at 12 months against 4.2 for the annual group, with the biggest gap in the non-technical skills part of the rating (4.8 against 3.9). They also started CPR sooner, at 13 seconds against 18. The annual group's confidence was no lower: it rated itself slightly higher than the monthly group did.
• Limitation: 35 people in one ambulance service, rated on a simulation rather than real patients, and the monthly group practised the same type of scenario it was tested on. Monthly sessions also meant more contact with trainers, so spacing is bundled with extra feedback. The authors note that they made many comparisons without adjusting for them.
• Lesson: For a skill that must be ready on any day of the year, a short session every four weeks did better than one long day, in less total time. Do not judge the schedule by how confident people feel.

Fewer, wider reviews can do the same job

Spacing does not have to mean more training time. A long-running study by four members of the Bahrick family, published in Psychological Science in 1993, compared relearning schedules for 300 English and foreign-language word pairs, with recall tested up to five years after training ended.

The four people were the authors, a family of psychologists who knew the memory research, learning vocabulary on their own. It is a small, unusual study and no basis for a workplace saving. What it shows is the trade-off: the longer gaps "slowed down acquisition slightly", so the wide schedule was slower during training and paid off later. Expect a wide review schedule at work to look slower in the first weeks too, and judge it by recall months later.

The 2357 method, and whether gaps should grow

Two questions come up whenever someone sets a schedule: should I use a named method such as 2357, and should each gap be longer than the last?

What is the 2357 method?

The 2357 method is a student revision plan that schedules study sessions counted back from an exam. Birmingham City University's revision guide to spaced repetition and the 2357 method describes a session the day before the exam, then sessions two, three, five and seven days further back, and says it suits revision "as little as 18 days away from your exams". The same page calls it "just a framework" and tells students to adjust the intervals.

If you count it out, the sessions fall 18, 11, 6, 3 and 1 days before the exam. Seen from the exam, the gaps get longer the further back you go; seen from day one, they shrink: 7, 5, 3 and 2 days. I found no study behind those numbers. It is a way to organise revision before a known exam date, rather than a tested application of Cepeda's single-review findings. For knowledge that must last a year it bunches every review into the first 18 days and leaves the rest of the year empty.

Expanding or equal gaps?

An expanding schedule makes each gap longer than the last; an equal schedule keeps the gaps the same. The idea that expanding is best goes back to flashcard systems and to early name-learning studies, and the research since then is mixed. Cepeda's 2006 review found expanding schedules ahead on average across 18 studies, but called the conclusions "necessarily tentative" because results varied so much between studies (p. 364).

• For equal gaps: In Karpicke and Roediger's 2007 experiments, where all the practice happened within one session, expanding gaps helped 10 minutes after learning, but equally spaced retrieval "produced superior retention 2 days later". They also found that delaying the first test improved long-term retention, whatever the later pattern.
• For expanding gaps: Dobson's 2012 study of 250 people learning physiology, with no feedback after each test, found expanding schedules gave higher scores 29 days after learning, especially when the first test came straight after the lesson.
• No winner: Karpicke and Bauernschmidt's 2011 study found that long gaps helped a great deal, but that the relative schedule of repeated tests "had no discernible impact".
• A useful middle result: in Kang, Lindsey, Mozer and Pashler's 2014 study, 37 adults learned Japanese words online with reviews on days 3, 9 and 28 or days 10, 19 and 28. Each person did both schedules. Final recall eight weeks later was about the same, but the expanding schedule gave higher average recall during the 28 days of training.

For a training team, begin with a workable spread of practice and check what learners remember between sessions. The 2014 vocabulary study suggests that expanding gaps can help maintain recall during learning, but it does not establish the best schedule for every workplace task. Equal gaps may be easier to organise around existing meetings.

How to schedule spaced reviews in a workplace course

To show the steps on one real-looking course, I'll use a hypothetical example; the company, its rules and its dates are illustrative.

A commercial cleaning contractor in Brisbane has about 140 cleaners working night shifts across 60 office and medical-centre sites. It is replacing ready-to-use bottles with a wall-mounted dispensing system and four colour-coded products. Many of the cleaners are in their first year with the company and some speak English as a second language. They already know how to clean; what is new is the dispenser, the colours and the contractor's own chemical rules. Every cleaner does a 30-minute course on a phone, "Safe chemical use on our sites", at the start of a paid shift on Monday 1 February 2027, the day the dispensers go live.

The course teaches the contractor's six rules. They are its own procedure, not a legal standard, and anything about first aid or spills points to the product's safety data sheet (SDS) and the supervisor:

1. Use the colour for the job: green for general surfaces, blue for glass, red for bathrooms, yellow for the medical-centre disinfectant.
2. Fill spray bottles only from the wall dispenser. Never decant into a drink bottle or an unmarked container.
3. Label every bottle with the product name and colour before it leaves the cleaners' room.
4. Never mix products. The red bathroom cleaner and the yellow disinfectant must never go in the same bucket, bottle or drain at the same time.
5. Gloves for every product; gloves and safety glasses for red and yellow.
6. If there is a spill or a splash: stop work, keep others away and alert the supervisor immediately. Follow the product's SDS and the site emergency procedure. Use a spill kit only if you are trained, equipped and it is safe to do so.

Before anyone uses the new products, the supervisor checks that they can select, fill and label a bottle, choose the required protective gear, and locate the SDS and spill kit. Any missed safety step is corrected before independent work. The later reviews reinforce that starting standard; they do not replace it. The colour rules get daily practice, while the spill and mixing rules may go unused for months.

1. Sort the content by how long it must last

List what the course teaches and, for each item, write how long people must remember it before the job or a check next tests it. For the cleaning course that gives three horizons:

2. Spend the reviews on what the job does not practise

Daily work supplies practice only when the procedure is being followed correctly. Keep supervisor observation for routine steps, and spend more of the delayed question time on rare events such as mixing or spills. This is a design choice for the example, rather than an interval tested in the research.

3. Set the first gap from the horizon

Use the study as a reason to spread practice, rather than a formula for this cleaning course. Check every safety rule before people start independent work. The day 2 review revisits colours; the first-quarter reviews revisit rules 2 to 6. The 26- to 28-day gaps fit the existing team meetings and are a proposed schedule to test.

The quarterly audit is an additional workplace check, not the first chance to catch incorrect filling, labelling or protective gear. After the first quarter, the planned gaps widen to 91 days as a workload choice. Bring a missed rule back sooner, correct it immediately when found, and retain the site procedure and supervisor support throughout.

4. Attach every review to an event the team already holds

A review with no time and no owner does not happen. Dunlosky's review describes a study where testing only once every three weeks led to study "peaking right before the exam", while daily testing spread it out, and concludes that students "will not necessarily engage in distributed study unless the situation forces them to do so" (p. 39). There is little reason to expect staff to behave differently, so each review in the cleaning course belongs to something the team already does: a shift start, the monthly team meeting, or the quarterly toolbox talk where the supervisor already covers safety. The supervisor runs it and knows the date.

Each review is a two- to five-minute task on a phone, which makes it a small piece of microlearning. Our microlearning examples show how to write one short enough to fit before a shift starts.

One of the clearest examples of a review tied to an existing routine comes from a hospital labour ward in Tanzania.

Case study: Weekly three-minute practice on a labour ward in rural Tanzania

Haydom Lutheran Hospital is a rural referral hospital in northern Tanzania with about 13 births a day. An earlier one-day course in newborn resuscitation lifted staff pass rates on a simulator from 18% to 74%, but what staff did with real newborns in the delivery room did not change, and no retraining followed.

• What they did: From February 2011, five local midwives became trainers and a newborn simulator was kept on the labour ward. Every Thursday morning, after the morning report, one of them ran a short session in which every labour ward staff member on duty practised on the simulator for about three minutes. There was also a monthly session of about 40 minutes, and a second one-day course for all labour ward staff in May 2011. The timing was agreed with the local midwives.
• Evidence and result: Research assistants observed every birth for a year before and a year after: 4,894 and 4,814 births. In Mduma and colleagues' 2015 study in Resuscitation, the share of newborns needing bag-mask ventilation fell from 7.3% to 5.9%, and deaths within 24 hours fell from 11.1 to 7.2 per 1,000 births (p = 0.040).
• Limitation: A before-and-after study at one hospital with no comparison ward, so the authors say causal relationships cannot be drawn. The extra one-day course ran in the same year, and practice outside the Thursday sessions was not recorded.
• Lesson: A one-day course changed test scores but not what staff did on the ward. Short practice at a fixed moment when the team already gathers coincided with better practice and fewer early deaths.

5. Write new questions for every review

Repeating an item can strengthen recall, but it can also leave you unsure whether someone can apply the rule in a different situation. Use the same rule in a new situation: a different site, a different product, a different person. For the cleaning course, keep enough scenarios per rule that none repeats within the year: at least three for most rules, and more for rules 4 and 6, which come up in every review. Give every answer one line of feedback that names the rule.

6. Check the results and adjust the next gap

After each review, look at which questions most people missed. Correct a missed rule now and bring it back sooner. A rule everyone gets right may be ready for a longer gap, but check that the questions test the intended action. That is the logic of Leitner's boxes, and you can run it with a spreadsheet. A 2012 trial by Kerfoot and Baker with 1,470 physicians in 63 countries used a version of it: questions answered wrongly came back after 12 days, questions answered correctly after 24, and a question was retired after two correct answers in a row. In the same trial, doctors sent 4 questions every 4 days scored lower than doctors sent 2 questions every 2 days.

Also check whether the reviews happened at all. A schedule on paper is not the schedule people get, as a hospital in Norway found.

Case study: Delivery unit staff asked to practise twice a month in Stavanger, Norway

Stavanger University Hospital has about 4,500 births a year. Most delivery unit staff did yearly off-site training in newborn resuscitation, but many rarely ventilate a newborn for real, so the skill fades between courses.

• What they did: From 2019 to 2021, midwives, nurses, anaesthetists, obstetricians and paediatricians had a personal teaching session of two to three hours, then were randomly assigned for nine months either to practise twice a month or to practise as often as they wanted, on a newborn simulator kept on the unit that gives feedback in a few minutes.
• Evidence and result: In Haynes and colleagues' 2021 study in Children, 187 staff completed the nine months. The twice-a-month group did a mean of 8 sessions, against a target of 18, and the free-choice group did 2.8. Scores at nine months did not differ significantly between the two groups. Staff who practised nine or more times scored significantly higher on one of the two test scenarios than those who never practised. The authors give two reasons for the missed sessions: heavy clinical workload, and the pandemic, which led managers to deprioritise simulation training.
• Limitation: The main comparison found no difference; the benefit appears only when staff are grouped by how much they chose to practise, which is not a randomised comparison. Scores were on a simulator, and the pandemic disrupted the schedule.
• Lesson: Twice a month on paper became roughly once a month in practice. Keep each review a few minutes long, put it where people work, give managers the completion numbers, and plan for busy months.

All three cases are health and emergency training, where the source search found measured review schedules. The Tanzania and Norway studies share researchers but concern different hospitals and study designs; neither supplies an independent replication of the other. I found no controlled non-health workplace study with a clear sample and a directly tested review interval in the sources checked. These cases inform the cleaning example, but they do not validate its dates.

The review calendar for the cleaning course

Here is the full year for the cleaning course, from the launch on Monday 1 February 2027 to the refresher a year later. All gaps are in days.

That is one 30-minute course, then six reviews of 2 to 5 minutes each, about 27 minutes of review in 12 months. The proposed 10-minute annual refresher needs to fit the organisation's requirements and observed performance; the calendar does not authorise shortening required training.

Here is what two of the reviews look like when written out.

The day 2 check

• When: Wednesday 3 February 2027, at the start of the shift.
• Gap: 2 days after the course.
• Questions: "You are cleaning the glass doors at reception. Which colour do you fill your bottle with?" (Blue.) "You are in a staff bathroom. Which colour?" (Red.) "The site procedure requires the medical-centre disinfectant for this bench. Which colour?" (Yellow.)
• Carried by: the shift start, on the phone, about 2 minutes.

Review 1

• When: Monday 1 March 2027.
• Gap: 26 days after the day 2 check, 28 days after the course.
• Questions: five scenarios, such as "A colleague says the bathroom will clean faster if you add some yellow disinfectant to the red bucket. What do you say?" (No. Never mix products. Rule 4.) and "You knock over a full bottle of yellow disinfectant in a corridor. What are your first two steps?" (Stop work and keep people away; alert the supervisor immediately. Follow the SDS and site procedure, using a spill kit only if trained, equipped and safe. Rule 6.)
• Carried by: the monthly team meeting, run by the supervisor, about 5 minutes.

The other three Review 1 questions cover rules 2, 3 and 5: an unlabelled bottle on the trolley, a drink bottle offered when the dispenser room is locked, and which products need safety glasses.

The review calendar download has this calendar filled in for the cleaning course, a blank version for your own, and a short check to run after each review.

Setting up spaced reviews in an AI course tool

When customers ask me about timing in support, they do not use the words spaced repetition. In threads I read from the past year, found by keyword search, six customers asked how to release modules on set dates, four asked how to bring learners back to a course, and two asked how to make a course come round again each year. None asked for spaced repetition by name. That is a handful of conversations, not a measure of how common the need is. In each of them, the scheduling question arrived as a question about the platform.

So on 23 October 2026 I tested the scheduling question on our own product. I asked the Coursebox course agent, part of our instructional design software, to build the cleaning course. Coursebox is our product, so weigh my account with that in mind. The brief gave the six rules in their exact words and asked for 13 sections: a short "Before you start", four teaching sections, a final check, six review sections labelled with their timing (day 2, week 4, week 8, week 12, month 6 and month 9) and a wrap-up last. It asked for real questions in every review, new scenarios rather than repeats, and feedback that names the rule.

The first outline had all 13 sections in the briefed order with the briefed titles, each review in its own section and the wrap-up last. Before it built the pages, the agent asked one question it could not answer from the brief: what the final page should say about where the SDS folder and spill kit are kept. I chose the option that says locations vary and learners must ask the supervisor to show them both before work, rather than letting it invent locations. On the rendered pages, every review had its questions written out. The day 2 review asked about jobs not used in the teaching section, such as "You are cleaning a lunchroom table. Which colour do you use?" Review 1 had five scenarios labelled Rule 2 to Rule 6, each with a feedback line naming the rule, such as "At an office site, you need to fill a green spray bottle. A coworker offers you chemical in another bottle." with the feedback "Rule 2: Fill only from the wall dispenser." Reviews 4 and 5 asked only about mixing and spills. In an earlier test I ran, spaced checks had come back with no questions in them; that did not happen here.

Then I looked at what stops a cleaner doing all six reviews on day 0, and nothing in the course did. The section menu offers Edit, Down, Delete, Archive and Locked, and Locked means a section opens once the previous one is finished, not on a date. Advanced Settings includes grading, pass grade, points, a restart button, duration and enrolment length, with no release date or delay. The agent's own hand-over note said so: "The review sections are labelled for the requested dates, but scheduling, due dates and reminders are not configured in this chat." A keen learner who kept clicking Next could finish "Review 5: month 9" on the first day, and six spaced reviews would become one massed session.

So I tied each review to a meeting the team already holds. In the page editor I added a "When:" line to the top of each of the six review pages. The day 2 page now says "Do this review at the start of your shift on day 2, on your phone. It takes about 2 minutes." The week 4 page says "Your supervisor runs this review with the team at the monthly team meeting in week 4. Do not do it early: the gap is what makes it work." The week 8, week 12, month 6 and month 9 pages name the monthly team meeting or the quarterly toolbox talk in the same way. After the fix, I reloaded the learner view and opened all six reviews. Each started with its "When:" line, above the questions.

What I checked was the outline, the pages and the settings I opened, not a year of reviews with real cleaners. A "When:" line asks people to wait; it does not stop them, so the schedule lives in the supervisor's calendar, not in the course.

The product test used an earlier version of rule 6. The current guide requires immediate supervisor notification and limits spill-kit use to trained, equipped staff when it is safe. That wording needs updating in the generated course; it was not rechecked in the recorded run.

If you already have a course and only need review questions for it, ask our AI quiz generator for new scenarios on each rule rather than the course's own questions, then check them and plan the dates in your team calendar.

What spaced repetition does not fix

Spacing changes when people practise. It does not change what they practise, or whether they can use it on the job.

• Most of the evidence is about facts. The gap studies used trivia, word lists and vocabulary, mostly with students. Workplace evidence on review schedules comes almost entirely from health and emergency training, mostly measured on simulators and knowledge tests.
• Remembering is not the same as performing. In a 2007 trial with urology residents in the US and Canada, daily emailed questions spread over 27 weeks kept test scores steady while a group that got all the questions at once declined, but the gains "did not generalize to higher scores on the In-Service Examination". A skill still needs practice on the real task.
• It feels slower. Wide gaps slowed learning slightly in the Bahrick study, and in the paramedic trial the annual group rated its confidence slightly higher than the monthly group, which outscored it. People may feel more confident after the one-off version, so check the review results instead.
• It does not guarantee permanent recall. Spacing does not make knowledge permanent, and some learning may remain after a review programme ends. Plan the reviews for the whole period people need the knowledge, and budget for them after launch, not only for the course.
• It cannot rescue the wrong content. A well-spaced review of a rule nobody follows on site keeps the wrong rule fresh. Fix the course first.

Putting spaced repetition to work

Spaced repetition works when the gap fits how long people need the knowledge and when someone makes each review happen. For training at work, that means sorting content by how long it must last, checking initial performance before use and choosing a workable first review, spending most reviews on what the job never practises, and attaching every review to a meeting or shift start that already happens. Treat named schedules such as 2357 as revision frameworks for a known test date, and do not let the pattern of the gaps hold up a plan.

Your next step: choose one course, identify the rules people rarely use and check that they can apply them before independent work. Then book a short delayed review with a named supervisor and use the answers to decide when those rules return.