AI has moved from vendor slides to board agendas, and asset-intensive organisations are under real pressure to be seen doing something.
In SFG20's State of Facilities Management Report 2026, 51% of respondents named AI and automation as the top trend of the next five years, but only 24% rated their organisation AI-ready. Ambition is running well ahead of readiness.
One question often goes unasked, and it decides whether any of this works: can the data underneath support it? For many organisations the answer is not yet, for reasons that have little to do with AI.
AI does not fix bad data. It industrialises it.
AI works with whatever it is given. Feed a model duplicates, gaps and unverified records and it is unlikely to flag them. It will return fast, confident answers at scale, carrying the authority of a technology people tend to trust. The output looks like insight, but it is the same unreliable data, now harder to challenge.
That can be riskier than having no AI. Data known to be unreliable gets double-checked. Data that has passed through a model tends to be believed. A 2025 global study by KPMG and the University of Melbourne found that 66% of employees who use AI at work rely on its output without checking its accuracy, and 56% have made mistakes in their work because of it.
The failure rates point upstream
When an AI initiative stalls, the review often blames the tool, the vendor or the budget. The independent evidence points elsewhere.
RAND asked 65 experienced data scientists and engineers why AI projects fail. Its report cites estimates that more than 80% do, twice the rate of IT projects without AI. Two of the five main causes came down to data: organisations lack either the data to train an effective model or the infrastructure to manage it.
Gartner reaches the same conclusion. By the end of 2025, at least half of generative AI projects had been abandoned after proof of concept, with poor data quality among the reasons. It also found that 63% of organisations lack the right data management practices for AI, or are unsure whether they have them. It expects 60% of AI projects without AI-ready data to be abandoned through 2026.
In facilities management, SFG20 found that the share of respondents citing integration with existing systems as a barrier to new technology rose from 20% to 52% in a year. Systems that look connected but do not work together rarely show it in a vendor demonstration, where the data is curated and the use case simple. The gaps surface once something important depends on the output.
The barrier is usually not a lack of AI. It is the lack of a data foundation AI can safely use, and a new tool does not supply that.
Getting the sequence right
AI readiness is often framed as choosing a platform, enabling the feature and switching it on. In practice the work starts a step earlier, with the data.
- Trace. Establish where the data came from and verify each record against evidence.
- Resolve. Remove duplicates, close gaps and settle the contradictions that would distort any analysis.
- Categorise. Align records to recognised classification standards, such as Uniclass for built assets or UNSPSC for procurement, so each asset, supplier or cost is described the same way in every system. Consistent codes are what make data machine-readable.
- Connect. Make the systems work together in practice, not only on paper, using open exchange formats such as IFC and COBie where BIM is involved.
This foundation work lasts only if someone owns it, the rules for creating and changing records are agreed and people are trained to apply them. It decides whether anything built on top will hold.
The foundation pays for itself with or without AI
The return does not depend on a future that may not arrive. Validated, well-structured data pays back straight away, whether or not a model is switched on, through better decisions and less time spent checking the data behind them.
AI readiness and operational performance do not need to compete for budget. They rest on the same foundation. Adding AI without it automates the uncertainty you already have.
The organisations that get value from AI will not necessarily be the fastest adopters. More often, they will be the ones whose data can support it. That work can start now, wherever the roadmap stands.
Before choosing a tool, go back to the question at the start: can your data support it?