Every manufacturer now has an AI mandate. Far fewer have an AI-ready organization. In 2025, MIT’s NANDA initiative analyzed 300 enterprise AI deployments and found that 95% of generative AI pilots delivered no measurable return. Those pilots didn’t stall because the models were weak. They stalled because the organizations around them weren’t ready to feed, trust, or use them. For engineering and IT leaders under pressure to show an AI story, that’s the uncomfortable truth and also the opening. This is what AI readiness actually means in product development, why it’s organizational rather than technical, and the two disciplines that decide whether AI pays off.
Key Takeaways
- Most AI initiatives in product development fail for organizational and data reasons, not technical ones. MIT found 95% of enterprise GenAI pilots return nothing measurable (2025).
- AI learns from your BOMs, your engineering change orders, and your metadata. Feed it inconsistent product data and it scales the inconsistency rather than fixing it.
- AI readiness is two disciplines run together: trusted, governed product data and adopted new behavior. Neither one alone is enough.
The industry agrees AI fails for organizational reasons. Most stop there.
A quiet consensus has formed across research and industry: AI initiatives fail for behavioral, cultural, and data reasons far more often than technical ones. In 2024, the RAND Corporation interviewed 65 experienced data scientists and engineers and concluded that AI projects fail at roughly twice the rate of non-AI IT projects, with the root causes sitting in data and organization rather than the algorithm.
That framing is now common ground, and it’s correct as far as it goes. In 2025, BCG put a ratio on the same idea: only about 10% of what makes AI work is the algorithms and models, 20% is the data and technology, and 70% is people and process. When seven out of ten parts of the problem are human, calling AI a technology project is the first mistake.
Here’s where most of the conversation stops, though. It lands on “people and culture, so do some change management” and goes no further. It rarely reaches the specific engineering data that AI actually consumes inside a manufacturer. That’s exactly where the real readiness gap lives, and it’s where the rest of this article spends its time.
AI does not fail in the model. It fails in the organization.
The headline numbers are blunt. In 2025, MIT found 95% of enterprise generative AI pilots produced no measurable profit impact, and the report was explicit that the failure was almost never the model itself. Gartner reached the same conclusion from the data side: through 2026, organizations will abandon 60% of AI projects that aren’t supported by AI-ready data, and 63% of organizations either lack or are unsure they have the data-management practices AI requires.
The problem is getting worse, not better. In 2025, S&P Global found the share of companies abandoning most of their AI initiatives had jumped from 17% to 42% in a single year. Read those studies together and the named causes repeat: undefined outcomes, broken workflow integration, and data that was never ready. None of those is a technology problem. Each one is a decision, a habit, or a gap in how the organization works.
Your AI learns from your BOMs, your ECOs, and your metadata
When a manufacturer says “our data isn’t ready for AI,” this is what that means in practice. AI does not learn from your data in the abstract. It learns from your bill of materials structure, your engineering change history, your part numbering, your metadata, and your effectivity rules. Gartner’s finding that 63% of organizations lack AI-ready data practices isn’t about storage or pipelines. It’s about the daily discipline of how product data gets entered and maintained.
Picture a few familiar examples. If engineers enter the same attribute three different ways across three sites, an AI model reading that data doesn’t reconcile the difference. It treats the inconsistency as signal and recommends part reuse you’d never approve. If your ECN and ECO traceability has gaps, a model asked to predict change impact will produce confident answers built on broken links. If documents never flowed cleanly through the lifecycle, the model has no reliable history to learn from. Multiply that across a multi-CAD environment and the picture gets worse.
The principle underneath all of it is simple. AI is a multiplier, not a repair tool. Feed it clean, connected, governed product data and it compounds the value of good engineering. Feed it chaos and it scales the chaos, now with a confidence score attached.
That’s the part of AI readiness no platform demo will show you, and it’s the part a manufacturer has to own before any model can help.
Everyone is pursuing AI. Few are capturing value from it.
Usage is nearly universal, value capture is rare, and the gap between them is organizational. In 2025, McKinsey’s State of AI research found that 88% of organizations use AI in at least one function, yet only about 6% described themselves as capturing significant value, and roughly 39% reported any measurable enterprise-level impact. The constraint is no longer access to AI. It’s the organization’s ability to put it to work.
Why does the gap exist? The organizations that capture value are the ones that redesigned how work gets done around the tool. McKinsey found that relatively few had fundamentally redesigned their workflows, and that redesign was the single biggest driver of whether AI delivered. BCG’s 70% people-and-process figure says the same thing from another angle. For a manufacturer, the differentiator is no longer whether you use AI. It’s whether your product data and your people are ready for it. Budget is not the constraint. Data discipline and real adoption are. For the wider context, this is the same shift driving manufacturers to rebuild the product-data backbone, the work at the center of our PLM transformation practice.
In regulated industries, compliance discipline is AI readiness
For aerospace, defense, and medical-device manufacturers, there’s an advantage hiding inside the compliance burden. The traceability, configuration management, and change control required to pass an ITAR, AS9100, or ISO 13485 audit is the same rigor that makes product data AI-ready. If you can prove the full chain of changes to an auditor, you can also feed that data to a model with confidence. Audit-grade product data is, almost by definition, AI-ready data.
The reverse is just as true. A manufacturer that has let traceability slip has a compliance problem and a readiness problem, and they turn out to be the same problem. The fix for both is governance and adoption discipline, not a new tool.
As one quality leader we work with put it, “If we can’t prove the chain of changes during an audit, the platform isn’t doing its job.” That same chain of changes is exactly what an AI model needs to reason about change impact. The companies furthest along on regulatory traceability have, without naming it that way, been building an AI-ready foundation for years.
Even quality-focused organizations still stall at scale, and the reasons are familiar. Capgemini’s 2025 World Quality Report found that around 90% of organizations are pursuing generative AI in quality engineering, but only about 15% have reached enterprise scale, with data and integration cited among the top barriers. The tooling is rarely the wall. The readiness is.
Technology adoption is the multiplier, not the model
Clean data gets you a capable model. It doesn’t get you a used one. The discipline that turns a working AI capability into realized value is change management, and the numbers are stark. Prosci’s benchmarking shows projects with excellent change management meet or exceed their objectives 88% of the time, against just 13% for those with poor change management, and effective executive sponsorship moves the same needle, from 27% to 79%. Most transformations that fail, fail here: McKinsey finds roughly 70% fall short of their objectives, and 72% of the failures point to employee resistance or management behavior rather than technology.
In engineering organizations, the resistance has a specific shape. AI gets read as judging my roadmap, replacing my judgment, or disrupting how planning has always worked. Trust is the gating factor, and it’s fragile. A 2025 global study by KPMG and the University of Melbourne found that while 66% of people use AI regularly, only 46% are willing to trust it, and trust is falling even as use rises. You can’t train your way past that with a one-day rollout.
This is the work ArcherGrey’s change practice is built for. Our approach surfaces and neutralizes the cultural anchors that quietly derail adoption, paces the rollout to how people actually learn rather than how the project plan assumes they do, and wires the new behaviors into daily workflows so the change holds after go-live. Done well, it frames AI as decision support that sharpens engineering judgment, not a replacement for it. That’s the difference between a model that technically works and one your team actually uses.
What an AI-readiness program actually looks like
AI readiness is not a model-selection exercise. It’s two disciplines run in parallel: getting product data trusted, connected, and governed, and getting people ready to adopt and trust the output. Neither alone is enough. Clean data with no adoption is a pilot that dies after the demo. Enthusiastic adoption on chaotic data is confident, scaled error. The manufacturers who get value from AI are the ones who treat both as a single program.
The data pillar is concrete and measurable. It means BOM accuracy, ECN and ECO traceability, consistent metadata and part numbering, controlled effectivity, and a digital thread complete enough for a model to follow. The adoption pillar is just as concrete: executive sponsorship, role-based enablement, resistance worked rather than wished away, and feedback loops that let engineers question and improve AI output. None of this is tied to a particular platform. The work is the same whether you run Windchill or anything else, which is exactly why it belongs to strategy, not to a software install.
The honest part is the part worth saying out loud. AI will not fix what your organization has not. A model trained on undisciplined data and dropped on a skeptical team will fail in ways that have nothing to do with how good the model is. Start with outcomes and readiness, sequence the data and the adoption work together, and the technology finally has something solid to stand on. If you’re weighing where to begin, our perspective on de-risking a PLM initiative before you start covers the readiness questions worth asking first.
Frequently asked questions
Why do most AI projects fail?
Most AI projects fail for organizational and data reasons, not technical ones. MIT found 95% of enterprise generative AI pilots returned nothing measurable in 2025, and Gartner expects 60% of AI projects to be abandoned through 2026 without AI-ready data. Fragmented product data and false adoption, not the model, are the usual causes.
What is AI readiness in manufacturing?
AI readiness in manufacturing is the combination of two things: product data that’s trusted, connected, and governed, and an organization prepared to adopt and trust AI output. In 2025, McKinsey found 88% of organizations use AI but only about 6% capture significant value, and the gap is almost entirely organizational rather than technical.
Does AI fix bad product data?
No. AI is a multiplier, not a repair tool. It scales whatever discipline or chaos already exists in your PLM. If your BOMs, engineering change orders, and metadata are inconsistent, AI will produce confident recommendations built on that inconsistency. In 2025, Gartner found 63% of organizations lack the data practices AI requires, which is where readiness has to start.
How do you prepare an organization for AI?
Start before the tool. Assess your data discipline across BOM accuracy, ECN and ECO traceability, metadata standards, and the digital thread, then run structured change management to build adoption and trust. Prosci’s benchmarking shows projects with excellent change management meet objectives 88% of the time, against 13% with poor change management.
Is compliance data already AI-ready?
Largely, yes. The traceability and configuration management required for ITAR, AS9100, or ISO 13485 audits are the same disciplines AI depends on. Regulated manufacturers who can prove a full chain of changes to an auditor often have a stronger AI-ready data foundation than they realize, and a real head start on readiness.
The bottom line
The future of AI in product development won’t be decided by which models a manufacturer buys. It will be decided by which organizations are ready to use them.
- AI in product development fails for organizational and data reasons, not technical ones.
- AI learns from your BOMs, ECOs, and metadata, and it scales their current state rather than fixing it.
- The readiness gap is about data discipline and adoption, not budget.
- For regulated manufacturers, compliance discipline is an AI-readiness head start.
- Real readiness is two pillars, trusted data and adopted behavior, run as one program.
ArcherGrey helps engineering and IT leaders get both halves of AI readiness right: the product-data discipline AI learns from and the adoption that makes it stick. It starts with an honest read of where your organization actually stands. To talk through that, explore our organizational change and adoption practice or start with the warning signs in 8 signs your PLM system is holding your team back.