What is formative assessment?

Formative assessment is assessment done during learning to find out what people do and do not yet understand, so the teaching can change before the final test. For staff learning to oversee AI agents, it means short checks on real decisions that show where judgement is weak while there is still time to fix it.

Why does formative assessment matter for agent oversight?

Most agent training ends with a single test. If someone fails it, they retake it; if they pass narrowly, nobody looks at what they got wrong. The person then reviews live items with the same gaps.

Formative assessment moves the checking earlier. A reviewer who keeps confusing refused with withheld is caught in week one, not after they have spent a month escalating working-as-designed refusals to the agent owner.

How does formative assessment work in practice?

It is frequent, low-stakes and specific. Good formative checks for agent oversight look like this:

• Three held items from a sample queue: approve, edit or decline, with one line on why.
• A short extract from the record: which bound decided this verdict?
• A list of five facts with their provenance: which can authorise a payment?
• A one-minute question at the start of a team meeting about last week's refusals.

The results are not graded for a credential. They go to the trainer or team lead, who reteaches the concepts most people missed.

What does formative assessment look like in a real team?

Take a utility's billing team preparing for an agent that drafts payment plans. Before training starts, the team takes a short readiness check. Halfway through, the lead runs a ten-item formative check and finds that most people trust any account note, whatever its source. She spends the next session on provenance. The final test shows the gap closed.

The before-and-during pattern is described in running a readiness assessment in a utility and, for a different sector, in a telecoms operator.

How is formative assessment different from summative assessment?

Summative assessment comes at the end and decides an outcome: pass or fail, certified or not. Formative assessment comes during and decides what to teach next. The same question can serve either purpose; what differs is what you do with the answer.

Both are needed. Summative assessment gives the evidence an auditor wants. Formative assessment is what makes people pass it for the right reasons. A team-wide check before go-live sits between the two, and the argument that a readiness assessment is cheaper than a failed go-live applies here.

It is also different from an AI risk assessment, which assesses the system, not the people.

Which related terms should you know?

• Knowledge check: a short formative quiz inside a lesson.
• Readiness assessment: a team-level check before an agent goes live.
• Retention check: a later check that learning has stuck.

The topic guide on training teams for AI agents shows where each type of assessment fits.

Where the Academy fits

Module tests in Pryme Intelligence Academy are free and set in the learner's own industry and role, so they work as formative checks before the certificate exam. For teams of up to 25, a free readiness assessment uses the Foundations test and reports scores by concept and the gaps, which tells the lead what to reteach.