Fix processes before AI, says Celonis

Public sector organisations risk embedding inefficiency into digital services if they deploy artificial intelligence (AI) before addressing the underlying way work is organised, according to a senior exec at process intelligence specialist Celonis.

Rupal Karia, SVP and general manager for the UK, Ireland, Northern Europe and MEA, argues that while government and the NHS continue to increase investment in AI, technology alone will not deliver the productivity improvements public sector leaders are seeking.

“Because productivity problems in the public sector are rarely caused by a lack of technology. They are caused by the way work is organised and how decisions are made,” he told Think Digital Partners.

“Across government and the public sector, services have grown more complex over time. Processes span departments, agencies and legacy systems, with manual handoffs and workarounds filling the gaps. In that environment, AI does not magically create efficiency. It simply speeds up whatever already exists.”

Understanding work before automating it

According to Karia, many transformation programmes fail because organisations prioritise technology decisions before understanding how their processes actually operate.

“Too often, organisations default to large, centralised technology approaches before fully understanding their own processes and data. AI is deployed before leaders have a clear view of where delays occur, why cases get stuck, or which steps add little or no value.

“Real transformation starts by understanding the operational context of how work actually flows today. Once leaders have that visibility, AI becomes far more powerful. It can be targeted at the points that truly constrain performance, rather than layered across the organisation in the hope that something improves.”

Karia said the challenge extends across the public sector rather than being confined to healthcare.

“This is not just an NHS issue. The same patterns appear in tax, benefits, immigration, regulation and local government. When processes are unclear or fragmented, technology spend becomes a blunt instrument.”

AI is ‘an amplifier, not a cure’

Karia believes the greatest risk is that organisations automate inefficient processes instead of redesigning them.

“The risk is not just wasted investment. The risk is that public services hard-wire inefficiency into their operating model.

“If a process includes unnecessary steps, unclear ownership or repeated rework, AI will not fix that. It will execute those flaws faster and at a greater scale. In citizen-facing services, that can mean quicker backlogs, more incorrect decisions and a poorer experience delivered more efficiently.”

He also warned that automation can make poor processes harder to challenge.

“Once an inefficient process is automated, it becomes harder to challenge. People trust the system because it is digital, even if the outcomes are no better.

“That is why AI needs to be treated as an amplifier, not a cure.”

Decisions should be driven by operational evidence

Karia argues that most organisations already hold the data needed to understand how services operate, but rarely analyse it across entire processes.

“Most public sector organisations already have the data they need. The problem is that it sits across dozens of systems and is rarely looked at end to end.

“On paper, processes appear linear and controlled. In reality, work loops, stalls and jumps between teams in ways that are invisible to senior leaders. No single group sees the full picture, which is why debates about ‘where the problem is’ often go unresolved.”

He said analysing operational context allows organisations to identify where delays and inefficiencies occur.

“It exposes where cases wait, where variation creeps in and where capacity is being consumed without improving outcomes. Importantly, it replaces anecdotes with facts.

“Once leaders have that shared view, priorities become clearer. Teams stop optimising in isolation and start fixing constraints that affect the whole service.”

Focus on outcomes, not automation

Rather than asking which tasks can be automated, Karia believes public sector organisations should focus on where complexity causes the greatest harm to citizens.

“Public value starts with outcomes, not throughput. Efficiency gains are easy to measure, but they do not automatically translate into better services.

“The right question is not ‘Where can we automate?’ but ‘Where does delay, error or complexity cause the most harm?’ That could be long waiting times, inconsistent decisions or staff spending time on tasks that add little value.”

Looking ahead, Karia expects AI to become more deeply embedded across public services, but says success will depend on organisational change rather than technology alone.

“This year, AI will be far more embedded, but the real change will be cultural rather than technical.

“The organisations that succeed will be those that treat AI as part of a broader operating model, not a bolt-on. They will also prioritise building ecosystems that are open, interoperable and resilient – avoiding over-reliance on any single provider and strengthening domestic and European innovation in the process.”