← Back

The Steering Room #8: AI in Aerospace and Defence

Nov 5, 2026

Jaime Andreo for The Steering Room podcast, episode 8.

Aerospace and defence run on documents. Specifications, contracts, compliance matrices, programme reports, change requests. On paper, an industry like that should be AI’s best customer. In practice, it is one of the slowest to get value from it.

In this episode of The AI Reality Check, Jaime Andreo, Senior Finance Consultant at Centida with finance experience in the industry at Eurofighter, explains why and what would change it.

The documents are the problem, not the opportunity

Most documents in the sector are controlled. They are restricted, owned by the customer, or covered by export licenses. People often don’t know which ones they are allowed to use, so they use none.

The industry is also built on signing procedures: every number and every document can be traced to someone who signed it. Jaime describes how finding the latest valid version of a signed document can take longer than the analysis itself. The knowledge is spread across the ERP, emails, Excel files and personal drives, places a standard AI license doesn’t see. A tool that can’t show where its answers come from clashes with that culture.

A capacity question

At the same time, the pressure is rising. Defence budgets are growing, and suppliers are being asked to ramp up. Bids can double, production rates can double — but no company can double its engineers or controllers within a year or two. Jaime sees the bottleneck in the back office: bids, contract reviews, programme reporting.

If AI takes over part of that administrative load, it effectively adds capacity a company can’t hire right now. The saved hours should go where they matter most: variance analysis, programme reporting and decision making.

Buying licenses is the easy part

Many companies have rolled out AI licenses and written guidelines over the past two years. Jaime is clear that these are the easiest steps: buying licenses is procurement, writing guidelines is compliance. Neither changes how a single report gets made. The result is a recurring cost without visible value — exactly what a CFO will question.

Usage then stalls. Nobody owns the process, senior managers don’t use the tools themselves, and teams treat AI as optional. People aren’t sure which documents they may use, and guidelines usually list what not to do, without examples of what to do. Jaime’s fix is simple: a one-page guide with real examples.

Who pays for AI?

By default, AI costs land in the IT budget as overhead. Jaime considers that a mistake: in IT, AI is seen as a cost. Charged to the programmes that use it, it becomes an investment someone has to justify — with a clear owner and visible value.

Customers will demand it, and restrict it

Customers such as Airbus and defence procurement agencies are likely to do both, in Jaime’s view: restrict anything involving technical or customer data, while expecting faster quotes and better quality and traceability. A clear AI setup may become a requirement for winning contracts — and most defence companies are unlikely to choose a cloud system for it.

Plan the process, not the tool

Programmes run for decades; AI models change every few months. Jaime’s answer is to plan around the process rather than the tool. What lasts is the data model, the programme structures, the contracts and the controls. AI tools should be treated as tools — replaceable — while the processes are designed to outlive them.

He also points to Europe’s dependency on US and Chinese AI providers. His advice: don’t put all your eggs in one basket. Keep data centres and models in-house where it matters, and use outside providers for part of the work.

The full conversation

Watch the full episode for Jaime’s view on what a realistic AI setup looks like in a sector where rules, security and long programmes shape every decision.