Formative vs Summative Assessment: A Guide for Workplace Training

Go from storyboard to an interactive course in minutes

Start Free

A team lead builds a two-week onboarding course for new support agents, in three modules. Every module ends with a short quiz, and the course ends with a final test. All of them are set up the same way: graded, with a pass mark, and nobody looks at which questions people got wrong. Three module quizzes and one final test, and none of them changes anything about the training.

Formative assessment uses evidence to improve ongoing learning or teaching. Summative assessment judges achievement against a standard at a defined point, such as the end of a unit or a sign-off.

The format does not decide which one a check is, because a quiz can be either. What decides it is what happens to the result: whether it counts, who reads it, and what changes because of it. I'm an instructional designer at Coursebox, which makes an AI quiz generator. This guide is written for workplace training, not school. It covers both definitions, the differences, examples of each, one onboarding course worked through with both kinds of check, and what happened when I built that course with an AI course tool.

What is formative assessment?

Formative assessment uses evidence to improve ongoing learning or teaching. It is commonly low stakes and can be ungraded or carry a small grade. The defining feature is that the learner or trainer uses the information to decide what to work on next.

Paul Black and Dylan Wiliam, who reviewed the classroom research on it, define it by what happens next. In Inside the Black Box they write that assessment "becomes formative assessment when the evidence is actually used to adapt the teaching to meet student needs." The check is only half of it. The other half is somebody acting on the answers.

What is summative assessment?

Summative assessment judges achievement against a standard at a defined point, such as a module, course or certification milestone. It may produce a grade or a pass decision. Whether someone is authorised to do a job depends on all required evidence and approvals, not the label or a quiz score alone.

In workplace training it usually takes one of these forms:

• A final graded quiz: a pass grade the learner needs before moving on.
• An observed task sign-off: a supervisor watches the person do the work against a checklist.
• A certification exam: set by an external body or an internal standards team.
• A renewal check: a licence or compliance record that expires and has to be passed again.

Each one compares the person with a standard decided in advance, not with how the rest of the group did.

Formative vs summative assessment: the key differences

Both kinds can use the same methods: multiple choice, short answers, observation or a role play. The method does not decide the type.

The test I use: what happens to the result?

When I review a course, I ask three questions of every check in it. The answers sort each check into formative or summative, and they show up the checks that are neither.

• What does the result decide? If it contributes to a grade, progression or sign-off, that is a summative use. If it guides feedback and the next attempt, that is a formative use. It may do both.
• Who uses it, and when? A learner may use an answer explanation immediately; a trainer may adjust the next session; an assessor may judge achievement at a milestone. Name the person and the intended action.
• What changes because of it? Formative: the training, or what the learner does next. Summative: whether the person is cleared.

A mid-course quiz with an enforced pass mark is serving a summative progression decision even if nobody reviews individual answers. Without usable feedback it supplies little formative guidance. If neither the result nor the answer information is used, its assessment purpose is unclear, although answering can still provide retrieval practice.

A self-check can be formative when the learner uses the answer to revise, even if nothing reaches a trainer report. In the worked course below, Priya also needs item-level results for coaching, so an unrecorded accordion does not meet that particular design.

Course briefs rarely say whether a check should count. I had an AI model classify a sample of the requests people typed into the Coursebox course builder over the last 90 days, then spot-checked the results.

A brief that says "add a quiz after each module" leaves the most important decision to whoever builds it. Say for each check whether it counts, who reads it and what they will do with the answers.

Can one assessment be both?

Yes. A graded task can judge achievement and provide feedback for the learner's next attempt. Results can also inform redesign before the next cohort, which is a formative use for course improvement rather than evidence that the first learner acted on feedback. Carnegie Mellon's assessment guidance describes how summative information can guide later learning.

What formative checks do for learning

A formative check can guide what the learner or trainer changes. Answering can also support memory through retrieval practice, which is a separate benefit. Henry Roediger and Jeffrey Karpicke showed the second effect in a 2006 study of the testing effect: students who were tested on a passage remembered more of it days later than students who spent the same time restudying it, even though the restudy group felt more confident. Their conclusion was that testing is a powerful way to improve learning, not only to assess it. A later review of study techniques by John Dunlosky and colleagues gave practice testing and distributed practice its highest utility rating, because they help learners of different ages and abilities.

The evidence for the first job comes from schools. In Inside the Black Box, Black and Wiliam reviewed studies in which teachers changed their teaching on the strength of formative evidence.

The classroom effect range is not a forecast for a workplace course. The transferable design idea is to collect relevant evidence early enough to act on it, give useful feedback and provide another opportunity to practise. Low stakes often help that purpose.

Case study: Testing as part of a resuscitation skills course in Copenhagen

Researchers at Copenhagen University Hospital wanted to know whether the testing effect holds for a practical skill, not only for remembering text. Their setting was a mandatory 4-hour in-hospital resuscitation course for medical students in the seventh semester of their degree.

• What they did: In Kromann, Jensen and Ringsted's 2009 study, 140 students were randomised to two versions of the course. One had 3.5 hours of instruction and training followed by 30 minutes of testing; the other spent the full 4 hours on instruction and training.
• Evidence and result: Two weeks later, 81 students volunteered for an assessment in a simulated scenario, scored against a checklist by assessors who did not know which group they were in. The group whose course ended with testing scored a mean of 82.8% against 73.3% for the group that practised for longer, an effect size of 0.93.
• Limitation: Only 81 of the 140 took part in the follow-up, all were medical students on one course, and the outcome was a simulation two weeks later, not performance with a real patient.
• Lesson: Spending the last half hour on a test instead of more practice led to higher checklist scores in a simulation two weeks later. A test inside the course time is part of the teaching, so budget time for it instead of treating it as overhead.

Formative assessment examples in workplace training

Most examples use a fictional support onboarding course. Its contractual policy allows a full refund within 30 days of renewal, offers unused-month account credit from day 31, and requires team-lead approval for refunds or credit over $500. Statutory rights are handled separately. Each example names both the check and how the trainer or learner uses its information.

1. A scenario question after a module

• The check: "A customer renewed 34 days ago and asks for their money back. What do you offer?"
• What the trainer does with it: reads which option each person chose and goes over the 30-day rule again if several picked the refund.

2. Explain it back

• The check: "In two sentences, tell a new colleague when a refund needs approval."
• What the trainer does with it: reads the answers for the gap people share, such as forgetting that credit needs approval too.

3. A practice role play with a peer checklist

• The check: pairs act out a difficult call, and the partner ticks off each step on a short checklist.
• What the trainer does with it: collects the checklists and coaches the step most pairs left out.

4. A draft work product reviewed before the final

• The check: each new starter writes their first incident report and submits it as a draft.
• What the trainer does with it: returns comments on the draft before the person writes the one that counts.

5. A one-question pulse at the start of a live session

• The check: "Which of the four rules are you least sure about?"

This confidence pulse helps choose a discussion topic. It does not establish whether agents can apply the rule; use their answers or performance for that.

• What the trainer does with it: spends the first part of the session on the rule most people chose.

6. An ungraded practice run of the final

• The check: a short practice quiz in the same style as the final, with different questions and answers shown.
• What the trainer does with it: tells each learner which rule to revise, and checks the practice questions are not misleading anyone.

For writing the questions themselves, our guide to e-learning knowledge checks covers question types and feedback.

Summative assessment examples in workplace training

Each example names the check and the standard it is measured against.

1. An end-of-course graded quiz

• The check: scenario questions that cover every rule the course teaches.
• The standard: a pass grade, such as 80%, set before anyone sits it.

2. An observed task sign-off

• The check: a supervisor watches the person do the task on the job.
• The standard: every item on an observation checklist completed correctly.

3. A scenario marked against a rubric

• The check: a written response to a realistic case, such as a complaint that needs a refund decision and a reply.
• The standard: a marking rubric that says what a pass looks like for each criterion.

4. A compliance certificate with a renewal date

• The check: an annual assessment on a policy such as privacy or workplace safety.
• The standard: the pass mark the policy owner sets, recorded with an expiry date.

5. A competency assessment in vocational training

• The check: in Australian vocational education and training, an assessor judges whether a learner is competent in a unit.
• The standard: the regulator's practice guide on assessment says skills are assessed by observing the person carry out the task in a suitable environment, and that the judgement rests on a range of evidence.

The last example shows how a regulator thinks about summative judgement, and it applies well beyond registered training: if the job is handling a refund call, the evidence has to include the person handling one. The validity of an assessment depends on it measuring the job it claims to.

Worked example: one onboarding course, both kinds of check

This is a hypothetical course, built to show where each kind of check goes and what happens to its results. A payroll software company has 9 new support agents starting together. Their team lead, Priya, runs a two-week onboarding course called "Refund and credit requests", in 3 modules. The agents must learn the company's four refund rules well enough to handle requests on their own:

1. Refund in full if the request arrives within 30 days of the renewal date.
2. From day 31, no refund: offer account credit for the unused months.
3. Any refund or credit over $500 needs team lead approval before you confirm it to the customer.
4. Record the reason code in the ticket before you close it.

The formative checks

There are 3 formative checks, one after each module. Each has 2 short scenario questions, is ungraded, shows the answer and feedback after the attempt, and allows one retry. Two of the items:

• Item: "A customer renewed 34 days ago and asks for their money back. What do you offer?"
• Feedback: "Day 34 is past the 30-day window, so no refund. Offer credit for the unused months (rule 2)."
• Item: "The customer's annual plan cost $1,200 and they renewed 12 days ago. They want a full refund. What do you do before you reply?"
• Feedback: "Within 30 days, so a refund is allowed, but $1,200 is over $500: get Priya's approval before you confirm (rule 3)."

Priya reads the results question by question on the Tuesday and Thursday of week 1. If most of the 9 miss the approval item, she runs a 15-minute huddle on approvals before Module 3 starts. The check changes the training that week, which is what makes it formative.

The summative checks

At the end of week 2 there are two summative checks. The first is a final of 6 new scenario questions covering all four rules, none of them repeated from the practice checks:

• Item: "A customer asks for a $500 refund 9 days after renewal. What do you do?"
• Expected answer: "Refund in full. It is within 30 days, and $500 is not over $500, so no approval is needed (rules 1 and 3)."

The other five follow the same pattern at the edges of the rules: day 30 at $450, day 31 at $320, day 14 at $501, day 70 at $480, and a $900 credit on day 47 that was approved but has no reason code in the ticket.

For this worked plan, use six equally weighted items marked correct or incorrect; 80% requires at least five correct. Every essential rule must also be correct in a further check, so the average cannot conceal a missed approval or required record. Priya reviews an initial five live tickets against all four rules and extends the sample if evidence is missing. Failed and not observed are separate. Independent handling follows complete evidence; approval above $500 always remains required. This illustrative rule differs from the 60-point assignment produced in the test below.

After the first cohort, Priya reads the final's results question by question. If the day 9 item at exactly $500 is the one most people missed, rule 3's wording ("over $500") goes into Module 2 with an example at the boundary for the next cohort. That is the summative result being used formatively, for the course rather than the person.

When to use formative and when to use summative

Use a formative check when the training can still change: during a module to see whether a rule landed, before a live session to choose what to spend it on, or before the final so learners find their gaps while it costs nothing.

Use a summative check when someone has to make a decision: signing a person off to work unsupervised, granting access to a system or customer account, issuing or renewing a licence, or keeping a compliance record.

A third kind, diagnostic assessment, happens before training starts and shows what each person already knows. Our guide to pre-training assessment covers how to set one up, and our guide to types of assessment sets out the other types, such as comparing each person with their own earlier results or with a fixed standard.

How formative and summative assessment work together

The two kinds should test the same skill with different questions. The formative items rehearse each rule, and the summative items check it again in new situations, so a pass says something about the next customer who calls on day 31.

The results flow in two directions. Formative results change the training this week: a huddle, a re-explained rule, a coaching conversation. Summative results, read item by item, change the course for the next cohort. When you've actually had to maintain a course library, you start to care about that second loop, because it is one source of evidence of which module is not doing its job.

Setting up each type in a course: my test in Coursebox

I wanted to see whether the Coursebox course agent would set the two kinds of check up differently when I asked for them, so I gave it the refund course in a test workspace.

What I asked for. I gave the agent the four rules and asked for a course called "Refund and credit requests" for 9 new support agents: 2 weeks, 3 modules, an ungraded formative check after each module with 2 scenario questions and feedback that names the rule, and a graded final of 6 different scenario questions with a pass grade of 80%, marked by a facilitator rather than AI. I also asked it to describe, not quiz, Priya's 5-ticket sign-off.

What it produced. The plan came back without clarifying questions: 3 sections and 11 pages, with a practice page at the end of each module, a page explaining the live-ticket sign-off and a final assessment page. The final was a rubric-marked assignment worth 10 points per question, and its page told learners that the pass mark is 80%, or 48 out of 60, and that a facilitator, not AI, would mark their answers. Its six items were the boundary cases in the worked example, and none reused a practice item. I found no invented rules or exceptions.

What was wrong. The three practice pages were accordions with answers and rule-based feedback. They could support learner self-correction, but did not collect the committed answers Priya needed for Tuesday and Thursday coaching. I also could not confirm facilitator marking as a setting on the final assignment. Its page text alone did not establish that workflow.

The fix, and how I checked it. I sent the AI editor one message on the Module 1 practice page: replace the accordion with a quiz using the same two scenarios, feedback naming the rule, Graded quiz off, one retry. It replied that it had built a two-question multiple-choice quiz but could not turn grading off or set a retry from the page editor.

Some things are still missing. The conversion dropped the rule-naming feedback, I did not set a retry, and Modules 2 and 3 are still accordions. I also did not check what the results view would show Priya, and no learners were enrolled, so nothing here shows anyone learned more.

After reload, Module 1 held a two-question native quiz and the accordion was gone. Graded quiz was off, matching this brief, despite the editor saying it could not change that setting. The formative purpose still depends on learners receiving useful feedback and Priya using the answers; those parts were not verified with learners.

Common mistakes

These are the three I see most when I review a course's assessments:

• Grading every quiz. A pass mark on a mid-course check turns practice into a test, so a wrong answer costs the learner something when it should only teach them.
• Showing no feedback on formative checks. A learner who gets it wrong and is not told why repeats the mistake in the final.
• Reusing practice items word for word in the final. The final then measures memory of the practice, not the rule.

One more applies to tools with AI marking: do not let AI mark a summative short answer with no human review, because a sign-off is a decision a person should own.

Formative vs summative evaluation (judging the course, not the learner)

The same two words are used for evaluating training programmes, and that is where they started. Michael Scriven coined formative and summative evaluation in 1967. Formative evaluation improves a course while it is being built or piloted; summative evaluation judges whether the finished course worked. Our guides to the CIPP model, the Dick and Carey model and the Kirkpatrick model cover evaluation in that sense.

Putting formative and summative checks to work

Formative and summative describe the use of assessment evidence. Use feedback to improve the next attempt; use explicit criteria when judging achievement. A task can serve both purposes. Where the job requires performance, collect performance evidence rather than relying only on a course score.

Take one course and list every check. Name what its result decides, who uses it and what happens next. Keep practice low stakes where useful, add actionable feedback, and use sufficient new cases for the final judgement. Review required performance and approvals separately from average scores.