What you can ask it today
Not "tell me about my business". Specific questions about a specific record, phrased the way somebody on your team would actually ask them. These are the prompts Ask AERONEXUM accepts today — substitute your own records for the identifiers shown.
Understand one sales order completely
You already know which order is the problem. These answer why, without opening everything behind it.
What is the current operational state of Sales Order 1234?Why is Sales Order 1234 blocked?What dependencies are holding up Sales Order 1234?What needs attention on Sales Order 1234?What guidance does AERONEXUM have for Sales Order 1234?
Follow a single unit through its life
The history attached to one serialized part instance, assembled rather than pieced together.
Show me Part Instance 1234Show me the lifecycle of Part Instance 1234What is the life status of Part Instance 1234?
Identify a unit from part number and serial
Resolve the physical identifiers in front of you to the specific instance in the system.
Resolve part number "622-123-001", serial "SN456"Resolve part number "622-1233-001", serial "SN789"
Ten prompts covering eight supported operations, and every one of them is a read. Nothing here changes a record. That is not a gap in the roadmap — it is the first stage of a deliberate sequence, and it is where a capability belongs until it has earned anything more.
Two practical notes. Record references currently take internal record IDs rather than the document numbers printed on your paperwork. And every answer is bounded by the permissions of the person asking — you can only ask about records you could already open yourself.
Every answer carries its evidence
This is the part that matters, and it is the difference between something you can use in an aviation business and something you cannot.
A general-purpose assistant produces fluent text. If it is wrong about a condition code, a core due date or an airworthiness document, it is wrong confidently and you find out later. The cost of that in this industry is not an awkward sentence — it is a shipment that should not have left, or one that did not leave when it should have.
So an answer here is never just prose. It comes back with the records it was drawn from: the specific orders, the serial numbers, the documents, the dates. You can check it in one click, and you are meant to. The system is built on the assumption that it has to show its work, because an answer nobody can verify is an answer nobody should act on.
What it does when the question is vague
Ask it something loose — "show me my open orders" — and it does not guess what you meant. It asks which order type and which owner you are talking about, and it reads nothing until you have said. Once the scope is explicit, it can return the matching orders along with their authoritative blockers.
That exchange is deliberate. A system that quietly picks an interpretation will eventually pick the wrong one, and the answer will look exactly as confident as a correct one. Asking costs you a sentence; guessing costs you a decision made on the wrong set of records.
The same discipline applies to what an answer is allowed to claim. Two distinctions the system makes explicitly rather than glossing over:
- A blocker is not a predictionIt can explain what is holding a sales order. That is not the same as forecasting whether the order will miss its promised date, and it does not pretend otherwise.
- Recorded is not derivedIt reports the shelf-life expiration the stock authority actually holds. It does not manufacture a cure-date forecast the record does not contain.
If an answer silently leaves out what it could not determine, the absence of a finding reads as an all-clear — and an all-clear that nobody actually issued is the most expensive kind of wrong answer in this industry. So where the data does not support a conclusion, the answer says so, in those words. You should be able to tell the difference between "nothing is wrong" and "I could not tell you whether anything is wrong".
What changes on a Tuesday morning
The honest pitch is not that AI transforms your business. It is narrower and more useful than that.
A lot of what an operations team does each day is retrieval: the same lookups, across the same systems, to rebuild a picture the system could have assembled itself. It is necessary work, it is largely mechanical, and it consumes the attention of the people best placed to handle the things that are not mechanical — the awkward customer conversation, the repair that came back with discovered scope, the part nobody can source.
Moving those lookups to software does not replace anyone. It gives the experienced people their mornings back, and it means the exceptions get found at nine o'clock by a system that looked, rather than at four o'clock by a customer who called.
- Answering "why" stops costing an afternoonThe reason an order is stuck arrives with the dependencies attached, instead of being reconstructed screen by screen.
- Answers stop depending on one personThe picture that lived in an experienced coordinator's head is something the business can ask for directly.
- The history is already assembledA unit's lifecycle and life status come back as one picture rather than being pieced together from several records.
- Attention goes where judgement is worth somethingPeople spend their time on the exceptions, which is the only part that actually needed them.
Why you can trust what it tells you
It works inside your existing permissions. It operates in your tenant, under the access rules you already enforce, and it cannot see data the person asking could not see themselves. It has no direct database access and no private route to your records.
It starts read-only. A new capability observes and reports before it recommends, recommends before it is allowed to act, and acts only on explicit authorization for that specific class of action — with the result verified afterwards by an independent read. Autonomy is earned against evidence, one step at a time, and plenty of capabilities should never move past reporting.
It also stays out of decisions that are not its to make. It does not make airworthiness determinations or sign releases. Those are certificate-holder responsibilities and they remain with the certificate holder.
The five-stage model behind all of this — outcomes, operational identity, orchestration, bounded domain capabilities, and trust earned in stages — is written up in full on how aviation operational AI works. If you are the person who will be asked whether this is safe to put near your data, start there.
Where this starts
Not with a platform-wide rollout. It starts with one question your team answers manually every day, and making the system answer it reliably instead.
That is a short conversation, and a specific one. The useful thing to bring to it is not a list of AI requirements — it is the thing somebody on your team checks every morning and wishes they did not have to.