Air OPS AI: Procedures, Notices, Evidence

A crew notice on revised fuel figures for one route is still in the briefing pack four months after the OM amendment that replaced it. The dispatchers work from the amended manual. Half the crews read the notice first. Nobody has sampled the difference, because the notice has no expiry date and no owner.

Air OPS compliance is full of gaps like that. It covers manual control, procedure wording, crew and dispatch notices, training impact, compliance monitoring, safety links and evidence. A weak change in one place shows up somewhere else: a notice nobody withdrew, a manual paragraph nobody trained, a dispatch step nobody sampled, an audit pack with records in three systems.

AI can help with this work, but only in the right lane. For Air OPS work, that lane is preparation. AI drafts, checks, structures and organises material for human review. It should not approve an operations manual, certify compliance, replace a nominated person, close a finding, issue an official notice or make an operational decision.

That boundary is what makes AI usable in a regulated operation.

Air OPS wording needs care

Operators may work across ORO, SPA, CAT, NCC or NCO contexts, depending on the type of operation and the approvals involved. The exact rule path, guidance material and acceptable means of compliance must be checked against current official sources, starting with the Air OPS rules in the regulation library, and against the operator's own approval basis.

AI should not pretend every operator has the same obligation.

What it can do is prepare the work around the obligation. It can turn a requirement note into a checklist, draft plain-language procedure text, and show where a manual section may need a responsible role, a record, a training note or an escalation path. It can prepare questions for the compliance monitoring manager, the flight operations post-holder or the subject matter expert.

For ORO-style management and operational control topics, that means responsibility matrices, meeting-note structures and compliance monitoring prompts. For SPA-style approvals, it means controlled checklists around approval scope, training notes, procedure references and evidence location. For CAT, NCC or NCO-style procedures, it means clearer instructions and role-based briefing material.

This output is a draft, not a legal position. The final interpretation and approval stay with competent people using controlled sources. How the operations manual and management system fit together is set out in the guide to the operations manual and management system under Part-ORO.

Operations manual drafts should describe the real process

A manual section that copies formal language without explaining the company process may pass a quick review and still fail on the line.

AI can help prepare a better first draft from a non-confidential process description. A practical structure includes:

• purpose and scope;
• affected operation;
• responsible role;
• required action;
• interfaces with other departments;
• records created;
• training or briefing impact;
• escalation route;
• review and approval route.

If a draft says operations must monitor a limit but does not say who monitors it, what record is used, or what happens when the limit is exceeded, the draft is not ready. AI can flag that weakness. It cannot make the procedure approved.

For an OM change, the reviewer pack should also answer:

• why the change is needed;
• which OM section, flight operations process or form is affected;
• who has to use the instruction: crew, dispatchers, operations control, station staff;
• which systems or records are touched;
• what crew or dispatcher familiarisation is needed before the effective date;
• whether a temporary notice is covering the gap and when it should be withdrawn;
• which questions need the procedure owner's answer;
• who approves and publishes the final version.

The approved manual stays in the company document system, with version control, distribution control and the normal approval path. The amendment lifecycle around it is covered in AI for aviation manual amendments and procedure control.

Flight operations: dispatch, crew reporting and operational control

Flight operations procedures are used under schedule pressure, often at night and often by people reading them for the first time in months. That makes clarity a safety matter, not a style preference.

The areas that generate most flight operations change work:

• Dispatch coordination. Who confirms the flight plan, fuel figures and weight data, and how a late change reaches the crew.
• Crew reporting. Report times, briefing content, what the crew must check before departure and where they record it.
• Operational control. Who can delay, divert or cancel, who must be informed, and how the decision is logged.
• MEL communication. How a deferred defect reaches dispatch and the crew, and how the operational procedure attached to it is briefed.
• Route briefing. Route and aerodrome information, NOTAM highlights, company-specific notes for an unfamiliar destination.

AI is useful here in two ways. It can make an instruction short enough to read at the crew room table, and it can show the interfaces a draft forgot. A dispatch procedure change that does not mention how operations control learns about it is a gap. A route briefing template that has no field for company notes will be filled in five different ways.

What AI does not do is decide. It does not release a flight, accept a defect under the MEL, decide on a diversion or sign a briefing record. Those stay with dispatch, the commander, operations control and maintenance under the operator's procedures.

Crew and dispatch notices should be short, controlled and traceable

Notices clarify a short-term issue, introduce a temporary control, prepare staff for a change or bridge the gap before a manual amendment.

They also create risk. A notice can stay alive after the reason has gone, as in the fuel example above. Staff may not know whether it is guidance or instruction. Training may not know it affects competence. Compliance may not know where acknowledgement evidence is stored.

AI can prepare notice drafts that are short, direct and traceable. A good notice package includes:

• reason for issue;
• affected staff, fleet, route or operation;
• required action;
• effective date;
• expiry or review date;
• related manual section;
• required acknowledgement;
• owner for follow-up;
• whether a permanent procedure change may be needed.

Those fields make edge cases visible before release and make withdrawal easier. The company still controls the notice. AI should not issue it, distribute it, mark it acknowledged or decide that it can replace a controlled manual amendment.

Training impact must appear early

Procedure changes create training impact. Sometimes it is small. Sometimes it affects flight crew, cabin crew, dispatchers, operations controllers, station managers, instructors or contracted providers.

The problem is that training impact is often noticed late. The manual changes first. The briefing comes later. Dispatch familiarisation is left to word of mouth. The evidence trail is built after somebody asks for it.

AI can prepare a training impact note whenever a procedure draft, notice or compliance action is created. It can ask:

• which roles are affected;
• whether the change is awareness, task training or competence-critical;
• whether recurrent training material needs an update;
• whether a briefing or acknowledgement is enough;
• what evidence should be retained;
• who confirms completion.

Official training records, competence records and staff data stay in the approved company system. A personal AI layer can hold reusable training templates, public learning notes and non-confidential briefing structures. It should not become a hidden training record. The wider training picture is covered in AI for aviation training compliance.

Audit evidence should be prepared before the auditor asks

Air OPS audits test both procedure and evidence. The operator may have a process, but the auditor needs to see that it is controlled, used and reviewed.

AI can prepare evidence packs before the audit, structured by:

• requirement or topic;
• manual reference;
• procedure owner;
• sample evidence;
• record location;
• training link;
• open gap;
• responsible follow-up person.

If the evidence location is unknown, the pack should say so before the audit. If the procedure is current but acknowledgement evidence is missing, the owner should see the gap early. For flight operations, samples might be two dispatch release records after a late fuel change, three crew acknowledgements for a notice, one route briefing for a new destination and one closed corrective action.

After an audit, AI can prepare a draft response that separates the finding, requirement, evidence sampled, immediate containment, root cause questions, corrective action proposal and closure evidence plan. It should not close findings. Closure belongs to the compliance monitoring process and the authorised reviewer.

Ramp inspection findings follow their own route. What a SAFA or SACA category 1, 2 or 3 finding means, and how the operator replies, is in the SAFA findings guide.

Occurrence follow-up in flight operations

A dispatch communication issue, a fuel coordination problem or a documentation error may need action from flight operations even when the safety team owns the report.

AI can help prepare the working structure: a timeline from the narrative, facts separated from assumptions, missing information listed, interview prompts drafted and actions grouped by owner. For a fuel coordination event, that means the planned figure, the uplift requested, the uplift delivered, the point where the difference appeared and who noticed it.

The official occurrence record, confidential statements, safety analysis and action closure stay in the safety management system. AI should not decide cause, assign blame, close the report or approve the safety action.

Keep sensitive material in official systems

Controlled procedures, operations manual content, safety reports, flight data, customer information, crew or staff data, proprietary methods, contracted provider records and official audit evidence must stay in company systems.

A personal AI layer is for safe learning and preparation. It can hold public sources, generic templates, personal certificates, reusable structures, non-confidential notes and draft methods. It helps a professional prepare a cleaner question or a better review pack without storing the company's controlled record.

The value is practical, not dramatic

AI is not a shortcut around Air OPS compliance. It is not a substitute for current EASA material, competent interpretation, approved manuals or accountable review.

Its value is more ordinary. Procedure drafts get clearer. Crew and dispatch notices get an owner and an expiry date. Manual changes arrive with their training impact attached. Audit evidence lists exist before the auditor asks. Missing owners, missing records and weak wording show up before they become findings.

In Air OPS compliance, that boundary is the control. Where a draft stops and a person decides is in assistance versus decision-making.