AI-powered centre and malpractice checks for awarding organisations
AOProGuard uses AI to read everything your organisation holds, from applications to A8.7(b) notifications to years of old records, and connects it, so the pattern behind a centre and the people running it is in front of your team when it matters. Your team sees what deserves a look, and every finding and decision is kept, so it holds up at appeal and stays when people move on.
AI assists. The analyst decides.
Why it matters
The cost of a missed centre is not the hour the check would have taken. It is the certificates issued, the learners affected and the decision you then have to defend.
How deep a check goes today depends on who does it and what else is on their desk. AOProGuard runs the same checks on every application: past directors, shared addresses, links to centres you have already withdrawn, and public records such as Companies House. Your analyst starts from what it found, not from a blank search.
When a centre appeals, you need to show what you knew, when you knew it and why you acted. Every alert, piece of evidence and decision is kept with who added it and where it came from. A report is a fixed copy of the case, so it reads the same at an appeal next year.
Much of what makes a good check lives in experienced people’s heads: the name that came up two years ago, the address that keeps reappearing. AOProGuard keeps it in the records, so a new colleague sees what your most experienced analyst would have spotted.
What happens to everything you give it
People who abuse the approval process rarely come back under the same name. AOProGuard takes everything your organisation holds, works out what it means and how it connects, and keeps checking it, so a pattern that would take an analyst days to find is in front of them the moment it appears.
Applications from your portal, A8.7(b) notifications from other awarding organisations, your old spreadsheets and records, and your team’s notes. One route for all of it, so nothing is checked differently because of where it came from.
AI reads emails and free text and pulls out the details in Ofqual’s terms. It also records what each piece of text means, so it can be found later by what it describes, not only by the words it uses.
Every record is compared with everything you already hold: people, company numbers, addresses, phone numbers, emails, websites, and stories that sound the same.
Directors are checked against Companies House, including the register of disqualified directors. Duplicate records are combined without losing any history.
Anything serious becomes an alert, ranked low to critical, with the evidence attached. It stays open until a person deals with it.
Two notifications about two different centres, in different words, describing the same method. Matching by meaning finds them where a keyword search finds nothing.
Every contact detail is held once and linked to everyone who uses it, so each record shows which other centres share its address, phone number, email or website.
People are matched on surname and month and year of birth, the details Companies House publishes. A different first name, an initial or a nickname does not hide them.
Anyone who ran a centre you rejected, suspended or withdrew is flagged the moment they appear on a new application, whatever the new centre is called.
Every director is checked against the Companies House register, with a link to the entry so you can confirm it yourself.
A name spelt differently, transliterated or written another way is still found, and marked as a possible match for a person to confirm.
“Portfolio evidence for six candidates contained identical witness statements with the same typographical errors.”
Different words, same method
“Assessor feedback on several learners’ work was word for word the same, down to the spelling mistakes.”
No shared names and almost no shared words. Found because they mean the same thing. Your analyst decides whether the two centres are connected.
Same surname, same month and year of birth
Raised as an alert with both records attached. Nothing is merged and nobody is labelled: your analyst decides.
In the product today
Being built next
With other awarding organisations
How it works
Six screens from the working tool, in the order your team would use them. Every screen shows example data: the centres and people are made up.
Step 1 of 6 · Alerts
When a check finds something that matters, such as a director who ran a centre you withdrew or one on the disqualified register, it becomes an alert. Each is ranked from low to critical, so you know where to start.
Nothing closes itself. An alert stays open until a person deals with it, and dismissing one needs a reason.
Step 2 of 6 · Connections
Northgate Skills Academy looks like a new applicant. One step out, its director John Smith was a director of Northgate Skills Ltd, whose approval was withdrawn for malpractice.
Every line shows what it rests on and how strong the evidence is. The suggested reading order is a starting point, never a judgement about anybody.
Step 3 of 6 · Search
Search for a centre or a person by name and it allows for spelling differences. “Nortgate Skils” still finds both Northgate centres, with their standing and risk beside them.
When nothing comes back, that tells you something too: you hold no record of them.
Step 4 of 6 · Notifications
When another awarding organisation emails you about malpractice, AI reads the email and fills in the details in Ofqual’s terms. The email sits beside its reading, and AI says how sure it was.
Nothing is saved to a record until a person confirms it.
Step 5 of 6 · Cases
A case gathers the centres, people, alerts and evidence an investigation concerns. Here, Harbour Point opened eleven weeks after its director left a suspended centre, and the case holds the Companies House records and a note of a call.
Evidence stays evidence until someone deliberately puts it on the record, and that step is recorded too.
Step 6 of 6 · Decide and report
Record what the organisation decided and what was then done. Create a report, a fixed copy of the case that reads the same at an appeal next year.
When malpractice needs reporting under A8.7(b), it drafts the notification in Ofqual’s own form and keeps a copy. Sending it stays with you.
Built for regulation
Anything you bring into malpractice work has to stand up to the same scrutiny as your own processes. These are the questions your compliance lead and your auditors will ask.
Every search, every record opened and every decision is logged with who did it and when. Opening a record counts, because looking at someone’s personal data deserves a record of its own. The log can be added to and is never edited.
Administrators and auditors can read the audit trail. Analysts cannot, so casework and oversight stay separate. An auditor can see everything and change nothing, and a refused request is recorded too.
AI reads, sorts and suggests. It never approves a centre, rejects one or closes an alert. Anything AI drafted is marked as such until a person confirms it.
Built around General Condition A8. Notifications follow Ofqual’s A8.7(b) template and keep its distinction between malpractice and maladministration.
Your records are stored in London, and the service will not run outside the UK. Matching by meaning uses a model we run ourselves in London. Reading notification emails uses a specialist AI service that does not train on what it reads, and nothing it suggests goes on the record until a person confirms it. Every user signs in with a password and a code from an authenticator app.
Ask in writing and you get all of your records back within 30 days, in formats any spreadsheet opens, at no charge.
What an analyst sees when they ask for the audit trail. The request itself is logged.
Next steps
Questions before we meet, or want to bring a colleague? Email Ian Bell.
ops@aoproguard.co.uk
