When Intelligence Is Not the Problem

AI, Traceability, and the Reality of Planning Work in Ireland

Anyone who has prepared a planning application in Ireland will recognise the pattern. The proposal itself may be straightforward, even well aligned with policy. The real work begins elsewhere: checking that every notice is dated correctly, that every plan uses the right colour code, that no obscure objective buried in a county development plan has been missed. Much of this effort is invisible in the final decision, yet it determines whether an application is even allowed to enter the system.

This is not a story of poor planning or irrational regulation. On the contrary, Irish planning professionals already operate within dense but defensible chains of logic. Decisions are grounded in statute, policy hierarchies, precedent and procedural safeguards. What is failing is not intelligence, human or artificial, but visibility. The system demands traceability, while many emerging AI tools offer abstraction without explanation.

That mismatch matters.

A System Under Pressure, Not in Crisis

The Irish planning system is handling scale as well as complexity. In 2022, local authorities made 33,592 planning decisions from just over 32,300 valid applications, with similar volumes persisting year on year (Office of the Planning Regulator, 2024a). At the same time, statutory planning services cost approximately €161 million to operate in 2021, while fee income covered barely one sixth of that amount (Office of the Planning Regulator, 2024a). These are not signs of a lightly loaded system.

What is striking is where the pressure shows up. National invalidation rates rose from 15 percent in 2021 to 17.5 percent in 2022, meaning nearly one in six applications failed before substantive assessment (Office of the Planning Regulator, 2024a). In some authorities, the rate was far higher. Dublin City invalidated over a third of submissions in the same year, while other counties invalidated fewer than 3 percent.

These are not conceptual failures. County guidance makes clear that many invalidations stem from minor omissions: mislabelled maps, missing north points, incorrect notices, unsigned forms. The rules are explicit. The difficulty lies in navigating them consistently across authorities, formats and evolving guidance. Even local authorities acknowledge that officers may spend minutes per application simply scanning for completeness, multiplying into significant administrative burden at scale (Office of the Planning Regulator, 2024a).

Complexity Without a Single Point of Truth

Planning professionals do not struggle because policy is unknowable. They struggle because it is fragmented.

Every application sits at the intersection of EU law, national policy, regional strategies, county development plans, local area plans and supplementary guidance. The Planning and Development Act 2024 alone runs to more than a thousand pages, consolidating and amending decades of legislation (Government of Ireland, 2024). Across 31 local authorities, those statutory foundations are layered with locally specific objectives and interpretations.

The information exists, but it is scattered. Different councils publish registers in different formats, operate on different IT systems and interpret similar requirements with subtle variation (Office of the Planning Regulator, 2024a). Consultants routinely cross reference multiple local authority archives, national policy documents and appeal decisions to answer what should be simple questions: What height applies here? Which assessments are required for this site? How have similar schemes been treated?

The result is a culture of defensive documentation. Reports are padded not because the issues are contentious, but because uncertainty about completeness carries unacceptable risk.

Time Lost to Rework and Delay

Statutory timelines promise certainty. Most planning applications are meant to be decided within eight weeks. Appeals have formal objectives. In practice, delay is common and costly.

By mid 2024, nearly 1,000 planning appeals had exceeded the statutory objective period, with hundreds pending for more than a year and some for over two years (Office of the Planning Regulator, 2024b). In 2023, only 26 percent of appeals were resolved within the target timeframe, and the average decision time stretched to 48 weeks (Office of the Planning Regulator, 2024b).

These delays are not neutral. Research on renewable energy projects shows that planning and regulatory delay materially increases project costs and risks missing climate targets (Lynch, 2023). In the UK, preparation costs for major applications routinely run into hundreds of thousands of pounds before construction begins, magnifying the financial impact of rework and resubmission (Strategic Land Group, 2020; Home Builders Federation, 2023).

Yet only a small minority of Irish applications ever reach appeal. In 2022, appeals accounted for roughly 7.4 percent of decisions, and around 88 to 89 percent of applications nationally were granted (Office of the Planning Regulator, 2024a). The system is not paralysed by conflict. It is slowed by administration and uncertainty.

Why Black Box AI Fails in Regulated Domains

Against this backdrop, it is tempting to see AI as a shortcut. Large language models can summarise documents, draft text and answer questions fluently. But fluency is not the same as fitness for purpose.

In planning, every conclusion must be defensible. Professionals are accountable not just for outcomes, but for the reasoning that leads to them. A tool that produces an answer without showing where it came from is not merely unhelpful. It is unusable.

This is where many AI tools falter. They optimise for speed and abstraction, when the domain demands traceability. Without visibility into sources, assumptions and logic, AI introduces risk rather than reducing it.

Making Logic Chains Visible

A genuinely useful AI system must surface sources explicitly, integrate precedent in a structured way, and support early validation. The value lies not in automation for its own sake, but in reducing rework and uncertainty by making the system’s expectations legible.

Intelligence as Infrastructure, Not Decision Maker

The most productive way to think about AI in planning is not as a decision maker, but as infrastructure. Its role is to organise, connect and expose information that already exists, allowing professionals to reason more efficiently within established frameworks.

Conclusion

The core limitation of many AI tools in regulated domains is not intelligence, but opacity. Planning professionals do not need faster guesses. They need clearer logic chains.

Explainable AI is not an ethical aspiration. It is a practical necessity for accountability, trust and professional responsibility.