Skip to content

Human-in-the-loop AI: why decisions in Aptem Enhance stay with the people accountable for them

human in the loop

Apprenticeship delivery answers to more than one regulator, and each looks at a different part of the picture. Ofsted inspects the quality of training and its impact on learners; it does not inspect AI tools as a standalone part of an inspection, and does not judge a provider on whether it uses AI. Ofqual regulates the end-point assessment most apprentices sit, and where a degree apprenticeship’s assessment is integrated into the degree itself, that assessment instead sits with the Office for Students. The Department for Work and Pensions (DWP), which took over funding policy from the Department for Education (DfE) in 2026, runs the audits that check learner evidence and delivery against what is being claimed.

None of these bodies ask whether a provider uses AI. They ask who made a decision, on what basis, and when it was recorded. That is the test any AI feature in an apprenticeship platform has to meet.

What is the role and value of AI if the decision still sits with a person?

If a tutor makes every call regardless, it is fair to ask what the AI is contributing. The answer sits before the decision, not in the decision itself.

A tutor’s oversight is limited by their own time: how many transcripts they can read closely, how many submissions they can mark in a session, how many questions they can put to a learner between sessions. Aptem Enhance does not take the tutor’s judgement out of that process. It gives that judgement more to work with, and lets it reach further across a caseload. A tutor reviewing an AI-drafted summary or a set of flagged feedback points is applying the same judgement they would use starting from nothing, only faster, and with a written record of the review itself.

Across all four features, the pattern is the same: AI produces a first draft, a set of results, or a summary; a person decides what happens with it; and it is the decision, not the AI output, that gets recorded.

The human-in-the-loop principle is built into Aptem Enhance

  • Virtual assistant: answers a learner’s routine questions about their progress, standard and KSBs around the clock, so a tutor’s time goes to the questions that genuinely need a person. Every exchange is logged in the tutor’s audit history.
  • Checkpoint: generates KSB-mapped questions and gives the tutor an impartial read on a learner’s understanding. The tutor decides what the result means and what to do next, whether that is a checkpoint run automatically or one triggered to address a specific set of topics.
  • Marking aid: maps feedback suggestions to KSB criteria so a tutor is marking against consistent criteria at speed. The tutor reviews, edits and approves every suggestion before a learner sees it; nothing goes out unchecked.
  • Enhanced reviews: turns a review meeting transcript into a structured summary and suggests trackable actions. A tutor reviews and edits both before they are saved, so the record reflects their judgement of the conversation, not a machine transcript of it.

This is not just a design preference. Ofqual, the qualifications regulator, states plainly that AI used as the sole mechanism for awarding a mark does not meet its requirement for human judgement in marking, a position it reaffirmed in January 2026. Marking aid’s tutor-approval step reflects that same principle, even though the on-programme exercises it supports sit outside Ofqual’s own regulated remit. In each case, the role of AI is to get a tutor to the point of decision faster, not to replace the decision. 

There is a compliance value here beyond accuracy. A review, a marking decision, or a checkpoint result is only useful as evidence if it is recorded close to when it happened, not reconstructed weeks later from memory. DWP audits specifically look for contemporaneous evidence of learning activity, evidence created at the time, not written up afterwards to fit the file. Because Aptem Enhance produces a summary or a set of results as soon as a review or submission is complete, the tutor’s sign-off happens promptly too, so the record shows a decision made at the time rather than one reconstructed later.

What this means for inspection and audit

Ofsted, Ofqual, the Office for Students or DWP do not need to know which parts of a provider’s workflow used AI. They need to see who made the decision that affected a learner, on what basis, and when. Because a person reviews, edits or approves every AI output in Aptem Enhance before it reaches a learner’s record, that trail already exists: it is the tutor’s own decision, evidenced, not a system-generated one that needs explaining after the fact.

That is what human-in-the-loop design means in this context: not an added safeguard, but the reason a person’s decision, and not the AI’s output, is what stands behind the record.

Download the brochure to learn more about Aptem Enhance or book a demo for a detailed look at the Aptem Enhance features.

Frequently asked questions

What does human-in-the-loop mean in an apprenticeship platform?

An AI feature produces a suggestion, a summary or a set of questions, and a person such as a tutor or verifier reviews, edits or approves the result before it affects a learner’s record or feedback.

If a tutor makes every decision anyway, what is the AI adding?

Time and consistency. The tutor’s judgement is the same either way; what changes is how much of their caseload that judgement can reach, and how quickly, because they are starting from a draft rather than from nothing.

Does Ofsted inspect AI tools directly?

No. Ofsted does not treat AI as a standalone part of an inspection and does not evaluate specific AI tools. Inspectors consider the impact any AI use has on learner outcomes and whether a provider’s decisions serve learners’ best interests.

Who regulates the assessment itself?

Ofqual regulates end-point assessment for most apprenticeship standards. For degree apprenticeships where the end-point assessment is integrated into the degree, that assessment is regulated by the Office for Students.

Does AI-generated feedback in marking aid go straight to the learner?

No. A tutor reviews, edits and applies the feedback before the learner sees it, and the workflow keeps a clean copy and a marked copy of the submission for compliance and portfolio evidence.

Does Ofqual allow AI to mark work on its own?

No. Ofqual’s published position is that AI as the sole mechanism for awarding a mark does not meet its requirement for human judgement in marking. Marking aid is built around a tutor reviewing and approving every suggestion for that reason, even though the on-programme exercises it supports fall outside Ofqual’s own regulated remit.

What is enhanced reviews?

The AI features inside a learner’s review: a summarisation feature that turns a review meeting transcript into a structured summary, and auto review actions, which suggests trackable goals a tutor can review, edit and save.

Additional resources

For more expert guidance on AI for apprenticeship delivery, visit our dedicated AI hub, including free on-demand webinar training.

Share this post with your friends