AI and automation
AI that helps work move forward — with people remaining accountable.
The useful part of AI in trust operations is not judgement. It is preparation: gathering what is relevant, drafting what is repetitive, and putting a clear decision in front of the right person.
Agentic, with controls
Interpret an event, prepare the next step, return to a person
Agentic does not mean uncontrolled. It means the platform can interpret an event, gather authorised information, propose the next actions, coordinate a workflow and return to the responsible person for approval.
- Step 1
Event detected
- Step 2
Context gathered
- Step 3
Sources checked
- Step 4
Action proposed
- Step 5
Human reviews
- Step 6
Approved work progresses
- Step 7
Evidence recorded
- Step 8
Outcome reported
AI agents do not replace APIs. APIs remain an important mechanism through which authorised systems exchange information. Any connection depends on supplier access, permissions, security review and technical approval.
Assistants
Eight focused assistants, not one general chatbot
Operations Assistant
Summarises the current operational position across schools.
Policy Review Assistant
Assembles the previous version, guidance and a suggested review summary.
Board Report Assistant
Prepares a draft assurance report from live records for human editing.
Contract Review Assistant
Summarises contract terms, notice dates and consolidation questions.
Digital Standards Assistant
Compares readiness across schools and highlights inconsistency.
Evidence Assistant
Identifies missing evidence and who is best placed to provide it.
Assessment Assistant
Converts assessment findings into proposed, owned actions.
Education Partner Service Assistant
Prepares client service summaries and review agendas.
Example capabilities
What the assistants are asked to do
- Summarise the current operational position.
- Identify missing evidence.
- Prepare draft reports.
- Convert assessment findings into proposed actions.
- Compare schools and identify inconsistencies.
- Find approaching deadlines.
- Summarise contract information.
- Suggest questions for a review meeting.
- Draft non-consequential internal communications.
- Explain why an item has been prioritised.
Every AI output carries
- Sources used: the specific records the draft was built from.
- Date generated: so nobody acts on a stale summary.
- Responsible reviewer: the named person who must check it.
- Controls: approve, amend or reject — nothing is applied without one of these.
If an assistant cannot evidence an answer from records the user is permitted to see, it says so rather than filling the gap.
Limits
What the AI cannot do on its own
These are deliberate design limits, not temporary restrictions.
- Make safeguarding decisions.
- Make employment decisions.
- Approve expenditure.
- Change official policies.
- Send consequential communications.
- Determine legal compliance.
- Override access permissions.
- Access information the user is not authorised to see.
AI features assist with administration, analysis and preparation. They do not determine legal, regulatory or safeguarding compliance, and they do not replace accountable human decision-making.
See the assistant working on fictional data
In the demonstration, every AI answer names the invented records it used and offers to create a draft action rather than acting alone.