The Real Barrier to AI Value in Capital Projects: Why Foundations Must Come Before Features
Construction has cycled through waves of transformative technology promises — digital twins, advanced analytics, sophisticated PMIS platforms, and now artificial intelligence. Each arrived with bold claims of faster decisions, lower risk, and dramatically better outcomes. Each settled into a more modest reality: useful in pockets, but rarely delivering the systemic shift the industry needs.
The current AI wave is no exception. While the tools show genuine capability in document comparison, markup consolidation, and visual progress assessment, broad disappointment persists. The root cause is rarely the models themselves. It is the persistent failure to build the data quality, process discipline, and governance foundations that any analytical capability requires — foundations that mature PMIS implementations are uniquely positioned to provide. Without them, AI simply inherits and amplifies the fragmentation that has limited every previous technology wave in capital projects.
Construction has never lacked for the next big idea. Over the past two decades, the industry has been told that digital twins, advanced analytics, blockchain, and sophisticated PMIS platforms would fundamentally solve its most persistent problems. Each wave arrived with confident predictions and impressive demonstrations. Each eventually settled into a more modest reality — useful in parts, disappointing in others, and rarely transformative on its own.
We are now in the middle of another loud cycle, this time centered on artificial intelligence. The promises are familiar: faster decisions, reduced risk, automated oversight, and dramatically improved project outcomes. Much of the current conversation, however, is being driven by voices with limited experience actually delivering complex capital projects or implementing the systems that support them. The result is a widening gap between what is being sold and what is realistically achievable on live programs.
Where These Tools Can Already Add Value
It is important to be clear about where current AI capabilities, particularly large language models, can deliver practical benefits today. These systems are already useful for comparing large sets of documents and drawings, consolidating markups and comments across multiple reviewers, and supporting visual progress assessment through image analysis. In a well-structured PMIS environment, they can reduce the time spent on repetitive coordination tasks and help surface inconsistencies that might otherwise be missed.
These are meaningful improvements. They do not, however, represent a fundamental break from the challenges that have limited previous technology waves. The value of any analytical capability remains tightly linked to the quality of the underlying data and the discipline of the processes that feed it.
The Data Problem No One Wants to Fix
The most common barrier I see in major project environments is not a lack of technology. It is a lack of investment in the foundations those technologies require.
The most common barrier I see in major project environments is not a lack of technology. It is a lack of investment in the foundations those technologies require.
Many owners still choose to adopt out-of-the-box PMIS solutions or rely heavily on external vendors to configure and manage their systems. These generic implementations are rarely tailored to the specific contract structures, delivery models (including P3/AFP), risk allocations, or commercial mechanics of the programs they are meant to support. As a result, the data captured in the system is often incomplete, inconsistently structured, or misaligned with how decisions are actually made.
When organizations later attempt to layer advanced analytics or AI capabilities on top of these foundations, the models inherit the same limitations. They can process large volumes of information quickly, but they cannot compensate for data that was never properly defined, governed, or connected to the commercial and contractual realities of the project. The output may look sophisticated, but its reliability is constrained by what was built underneath.
Recent industry analysis reinforces this pattern. RAND Corporation research from 2024 found that more than 80% of AI projects fail to reach meaningful production deployment — roughly twice the failure rate of traditional IT projects — with inadequate data quality and accessibility among the primary root causes. S&P Global’s 2025 survey showed the share of companies abandoning most AI initiatives rising to 42%, up from 17% the previous year, with data issues frequently cited as a top obstacle. Gartner has similarly noted that poor data quality or lack of relevant data drives the majority of AI project failures.
In construction specifically, project data is characteristically unstructured, fragmented across multiple environments (often a median of 11 in recent regional surveys), and lacking standardization, precisely the conditions that undermine AI effectiveness.
Mature implementations take a different approach. They treat the PMIS not as a software installation, but as an extension of organizational process and commercial intent. This requires clarity about how information should flow, who owns which data, and how the system needs to reflect specific contract terms. Without that clarity, even the most advanced analytical tools will continue to produce outputs that require extensive manual verification — often by the same experienced people the organization hoped to free up.
The Verification Burden and Accountability Challenge
Increased analytical capability does not reduce human effort in a straight line. In practice, it often shifts the workload. Teams spend less time gathering information and more time reviewing and validating what the models have produced. This verification burden grows as the volume of analyzed data increases, and it tends to fall on the people best equipped to catch errors — precisely the senior resources organizations can least afford to tie up in checking outputs.
This dynamic also creates a subtle but real accountability problem. When models are used to support high-stakes activities, change order evaluation, progress validation, risk identification, or commercial recommendations - it becomes easier for responsibility to become diffused. Someone still needs to own the final decision and stand behind it when outcomes differ from expectations. Treating model output as a starting point rather than a conclusion is essential, but this discipline is not automatic. It must be deliberately designed into workflows and governance.
Why Transparency Remains Difficult
Even when the technical foundations are sound, organizational behavior often gets in the way. Many project environments continue to keep selected processes, delays, and performance data outside formal systems. This is rarely an oversight. It is frequently a deliberate choice that protects against uncomfortable visibility — who is holding up approvals, where overruns are occurring, and which parts of the organization are consistently underperforming.
Modern PMIS platforms, and the analytical tools layered on top of them, make this kind of opacity harder to maintain. For organizations that have not yet developed the maturity or political willingness to operate with that level of transparency, the introduction of more powerful analytical capabilities can feel threatening rather than enabling. Until leadership is prepared to accept the visibility these tools create, adoption will remain partial and the benefits will stay limited.
McKinsey’s ongoing analysis of construction productivity underscores the broader stakes: the sector has seen productivity growth of roughly 0.4% annually for two decades while facing a projected need for $22 trillion in output by 2040. Technology alone has not closed the gap because it has not been deployed against disciplined, standardized processes and data architectures.
What Responsible Adoption Actually Requires
The organizations that will extract real value from these capabilities are not the ones chasing the most advanced features. They are the ones willing to do the less glamorous work that precedes them.
Responsible adoption starts with honest assessment of data quality and process discipline. It requires clear decisions about where model output will be trusted, where it will be routinely validated, and who remains accountable when recommendations prove incorrect. It also demands leadership that is prepared to accept greater transparency into how work actually gets done and where problems originate.
These are not primarily technology problems. They are organizational and commercial problems that technology can expose or accelerate, but cannot solve on its own. Treating AI adoption as just another technology transformation — with the same attention to data foundations, process clarity, change management, and governance — is the most reliable path to results that justify the investment.
In practice, this means PMIS implementations that enforce consistent attribute definitions, maintain end-to-end requirements traceability, standardize workflow semantics, and create governed environments where AI agents can operate deterministically within defined policy boundaries. It means moving beyond generic configurations to systems explicitly aligned with the contractual and commercial realities of large capital programs. And it means designing human-in-the-loop verification into every high-stakes workflow from the outset.
The current wave of interest in AI for construction is not going away. The tools will continue to improve. What will determine whether they deliver meaningful impact is whether owners and delivery teams are willing to build the conditions those tools actually need to perform. Without that foundation, we will simply add another chapter to construction’s long history of promising technologies that failed to live up to their billing.
The choice is no longer whether to engage with these capabilities. It is whether to build the PMIS and organizational discipline that turns capability into consistent, defensible results on complex programs.
About Kahua
Kahua is a leading provider of capital program and construction project management software. We are enabling innovation that is changing the way that capital programs are planned and delivered. The world’s leading construction organizations use Kahua’s collaborative construction management solutions to improve efficiency, lower costs and reduce project risk throughout the entire lifecycle of their capital programs. Our purpose-built solutions for owners, program managers and contractors enable rapid implementation that minimizes time-to-value and enhances user adoption. And with Kahua's low-code application platform our customers can easily customize existing Kahua apps or even build their own new apps to run their business at peak efficiency today and rapidly adapt as business conditions dictate. To learn more, visit www.kahua.com.