AeronauticsAI in Engineering: why adoption hasn’t translated into value
AI investment in industry has never been higher. Neither has the gap between spending and results.
Global corporate investment in AI more than doubled in 2025, reaching $581.7 billion (Stanford HAI AI Index, 2026). Among manufacturers, 97% say AI is now embedded in at least one core process. Yet fewer than 30% of executives are satisfied with the return they’re getting (Gartner, 2025), and Stanford’s 2026 AI Index finds that while 88% of organisations now use AI in at least one business function, the resulting productivity gains stay concentrated in a narrow leading cohort. Adoption is no longer the question. Value is.
How much value is AI in engineering actually creating?
The technology itself is no longer in question. Airbus has documented measurable gains from digital twins across its assembly lines. Schneider Electric has cut energy consumption by 26% at its smart factories. AI-driven predictive maintenance, correctly implemented, delivers 30 to 50% reductions in unplanned downtime and productivity gains of 15 to 30%, usually within six to twelve months. Stanford’s 2026 AI Index finds a close parallel in software development, where documented productivity gains reach 26%, among the strongest of any function it tracks.
The divide is in who captures that value. McKinsey Global Institute’s long-running simulation of AI adoption puts numbers on it: manufacturers that absorb AI within the next five to seven years are on track for a cumulative 122% increase in cash flow by 2030, against just 10% for slower followers, and an estimated 23% decline for companies that never adopt at all. Stanford’s 2026 AI Index confirms the same pattern at a broader scale: 88% of organisations now use AI in at least one business function, but most deployments remain layered onto existing processes, running as pilots, proofs of concept and steering committees, rather than transforming how the business runs. None of that shows up on the income statement.
Why do most AI in engineering projects stall at the pilot stage?
The technology is rarely the bottleneck. What’s missing is a direct line between what the AI does and what the business needs. Many projects launch under a broad mandate, “become an AI company”, without answering a simpler question first: which outcome, by how much, and by when? Without that anchor, the project produces a demo instead of a result. Stanford’s 2026 AI Index points to a compounding factor: 74% of respondents now cite inaccuracy as their top AI risk, up 14 points in a single year, ahead of cybersecurity and regulatory compliance.
France illustrates the pattern. 62% of French manufacturers are running concrete AI projects, backed by public initiatives such as France 2030. But success stays concentrated in large groups. SMEs, which employ the majority of industrial workers, have largely not crossed the line from experiment to operational system.
What’s behind the AI in engineering skills gap?
A structural skills shortage compounds the problem. 60,000 manufacturing roles are currently unfilled in France alone (French Ministry of Labour, Q3 2025). Across Europe, hybrid profiles, augmented engineers who can also work with AI systems, remain in short supply. AI cannot compensate for a workforce that hasn’t yet been trained to use it. The demand signal extends well beyond France: Stanford’s 2026 AI Index, drawing on Lightcast labour-market data, finds AI-related skills now explicitly requested in 2.5% of all US job postings, a 297% increase over the past decade. According to Deloitte’s TMT Predictions 2026, most AI will soon be consumed indirectly, embedded in existing enterprise tools rather than deployed as standalone platforms. The era of the isolated AI project is ending.
What will define industrial competitiveness over the next 18 months?
Three forces are compounding the gap between leaders and laggards.
Regulation. The timeline just moved. Brussels’ Digital Omnibus, in force since late July 2026, pushed back the EU AI Act’s high-risk obligations, covering critical infrastructure, recruitment, healthcare and financial services, from 2 August 2026 to 2 December 2027. Transparency rules and the AI literacy obligation still apply on the original schedule, and a compliance programme still takes 12 to 18 months to build. European companies have already invested €4 to €6 billion collectively; for those who stay the course, that investment still becomes a barrier to entry, just on a longer runway.
Reindustrialisation. 81% of manufacturing executives want to expand domestic production in 2026, and 74% are nearshoring or considering it (Xometry, 2026). New facilities can build AI-native data infrastructure in from day one, a structural advantage over retrofitting.
Competitive divergence. 82% of manufacturing executives now see AI as an essential growth driver, and nearly half report significant ROI already (Xometry, 2026), a pattern Stanford’s 2026 AI Index confirms more broadly: 88% organisational adoption, but productivity gains concentrated in a narrow leading cohort. The middle ground, companies that have neither committed nor opted out, is disappearing.
Where does ALTEN see the gap actually closing?
Engineering and R&D functions show the pattern in its most concentrated form. Two different numbers are easy to confuse here. The first is about scale: how many organisations have reached AI maturity that actually changes how a programme runs. ALTEN’s own client work suggests engineering-heavy environments sit below the 88% headline adoption figure once the bar moves from AI being used somewhere to AI being embedded in how the programme runs. The second is about magnitude: how much the deployments that do work actually deliver. Vendors routinely claim task-level productivity gains of 30 to 40%. Measured across the full total cost of ownership of an engineering programme, ALTEN’s own figure lands closer to 12 to 15%, still meaningful, but a different conversation than the one most procurement decisions are built on. ALTEN’s own framework sorts AI-in-engineering maturity into three stages: personal productivity tools such as coding assistants and prompt platforms, now table stakes rather than a differentiator; process-level AI embedded across an entire workflow, from requirements engineering to predictive maintenance and digital twins, where ALTEN has measured time-to-market reductions of 25 to 30% on specific programmes; and AI-defined products and services, where AI creates capabilities that did not exist before. Most organisations are still consolidating the first stage.
Across more than 300 AI deployments with industrial clients, ALTEN sees three traits consistently separate the projects that scale from the ones that stall: a named, measurable business outcome agreed before deployment, not a vague ambition to improve efficiency; an integrated team where ALTEN’s own AI specialists work inside the client’s environment from day one rather than delivering from a distance, a model close to what the industry has started calling forward deployed engineering; and a data governance agreement settled before implementation starts. In regulated sectors, from aerospace to pharmaceuticals to financial services, ALTEN has built its own AI management system around explicit alignment with the EU AI Act and ISO 42001, on the view that auditability is now a contractual requirement rather than a preference.
That delivery model rests on more than governance. ALTEN’s approach to AI in engineering is built on three reinforcing pillars: strategic technology partnerships, with Mistral, Google, Nvidia, Anthropic, Microsoft, OpenAI and Dataiku among others, which keep architecture choices driven by the problem rather than a single vendor’s roadmap; a set of proprietary accelerators, built as configurable frameworks rather than off-the-shelf products, covering AI in engineering, application lifecycle management and data-driven decision-making; and more than 5,000 AI and data specialists organised across 20 global Centres of Excellence. With Mistral specifically, the partnership extends to on-premise deployment and purpose-built Small Language Models tailored to engineering domains, an increasingly explicit requirement where strategic autonomy over sensitive data now weighs as heavily as model performance.
How can companies in engineering-heavy industries turn AI investment into measurable returns?
Companies that generate returns share a consistent set of practices: prioritising use cases with documented payback periods, since predictive maintenance, automated quality control and technical data management have the strongest track record; setting measurable KPIs before deployment rather than after the fact; investing in data platforms before selecting models; and building teams where engineers and AI specialists work together from the outset, rather than in separate workstreams that converge at the end.
At ALTEN, this is a conviction as much as a practice: AI augments the skills of engineers, it does not replace them, provided the data foundations are in place. The technology’s value lies in what it frees engineers to focus on, not in what it automates away.
In 2026, the question for industrial companies is no longer whether to use AI. It’s whether their AI investments are tied to a clear business outcome, built on the right data, and governed well enough to scale. Getting this right isn’t just a technology advantage. It’s the operational model that will define competitive industry for the next decade.
