AI in apprenticeship software, without the hype.

The useful question is not "does it have AI?" but "what does the AI read, what can it change, and can I check it?"

What AI is good for

• “What needs my attention today?” Start with a supported operational question, then inspect the available facts and next-step links.
• Drafting for a person to review. A draft progress-review summary from the logged hours and evidence, a draft feedback note for an assessor to edit, proposed KSB links for an upload. All of it is a suggestion until a named person accepts it.
• Spotting patterns early. Off-the-job hours drifting behind plan across a cohort, an employer whose reviews are always late, funding records that would fail the next return. The pattern is surfaced; the action is taken by a person.

What AI must never decide

Funding eligibility and price. Gateway readiness. End-point assessment marks and grades. Learner status changes, including breaks in learning and withdrawals. Quality judgements on a tutor's work. Safeguarding decisions. Each of these is a consequential action with a person accountable for it, and each is recorded in the audit trail with that person's name. A supplier that lets a model take any of them has removed the accountability the funding rules and the inspection framework assume.

Seven questions to ask any supplier

1. Where do the answers come from: my records, other providers' records, or the open web?
2. Can the AI change any record on its own? Which ones?
3. Does every answer show the records it read, with timestamps and a version?
4. Is learner data sent to a third-party model provider? Which one, under what terms, and is it retained?
5. How is usage metered, reported and priced? Is it included or billed per query?
6. Has the supplier published a responsible-AI statement and completed a data protection impact assessment?
7. What happens when the AI is wrong? Where is the correction recorded?

How Journey does it

Supported answers show source references and an as-of timestamp, with links to the relevant work. Open the source and check the recorded facts before making a decision. Journey records AI usage and audit metadata; an answer is not an immutable snapshot of every source record.

A one-line test

Ask the supplier's AI a question about a learner, then ask it to change something about that learner. The first should return an answer with sources. The second should return a refusal and the name of the person who can. If either answer is different, keep looking.

Questions providers ask

• Can AI make funding or gateway decisions in apprenticeship software?

  It should not, and in Journey it cannot. Funding, gateway, EPA, learner-status and quality decisions are consequential actions that stay with a named person and are recorded in the audit trail. AI can read the record and answer questions about it; people decide.

• Where should an AI feature get its information?

  From the provider's own records, scoped to that provider, with every answer listing the records it read. An AI that answers from the open web, or from other providers' data, cannot be checked and should not be near a learner record.

• Is AI in apprenticeship software safe for learner data?

  It depends on governance, not on the model. Ask whether learner data leaves the tenant, whether prompts and outputs are stored and for how long, which sub-processors are involved, and whether the provider has completed a data protection impact assessment. Journey Advanced answers from the provider's own record inside the signed-in app and lists sources; details are on the Responsible AI page.

• What questions can Journey AI answer?

  Questions the record can answer: learners behind plan on off-the-job hours, reviews overdue, funding records that would fail the next ILR return, learners approaching gateway with an item outstanding, actions overdue and their owners, and cohort or employer comparisons of any of those.