AI decision-making needs clear human boundaries, contributors say

The commentaries, published by TechRadar Pro and Sifted, focused on decision intelligence: AI systems that move beyond retrospective analytics to recommend or take actions. An Infosys executive writing in TechRadar Pro said UK enterprises are already seeing results, citing a recent study that found 66% of UK enterprises are experiencing significant AI-driven productivity improvements, with leaders saying AI is freeing time for strategic decision-making, not just automating routine tasks.

The Infosys executive, identified as AVP and Unit Technology Officer for Data Analytics and AI, said traditional analytics has been reactive, slow and dependent on deterministic rules and static workflows. Agentic AI systems working in squads, by contrast, can analyze data from varied sources and make chains of complex decisions. The article gave the example of an AI underwriting agent that estimates a prospective borrower's creditworthiness, adjusts loan pricing, routes paperwork and alerts supervisors to compliance concerns. It said such systems need clear decision rights from the outset: an agent might price a loan or route paperwork on its own, but a compliance concern or an applicant outside its data constraints should go to a human rather than be resolved by the agent guessing.

The opportunity is sizable, the article said, estimating Europe's decision intelligence market at $4.35 billion in 2025 and projecting $20.61 billion by 2035. It said UK organizations could capture a meaningful share if they move deliberately. The article also cited beauty and fragrance retailers that implemented an AI decision engine to determine the right time and proposition for engaging each customer and earned double-digit ROI. It warned that AI-based decision intelligence carries cybersecurity risk, compliance failure, lack of transparency and accountability and occasional improper outcomes, and that full implementation needs operating model changes, deep integrations, specialized talent and governance.

A second TechRadar Pro article, by Spendesk's chief product officer, made a similar point about finance. It cited Deloitte's CFO Signals as showing close to 93% of CFOs now use AI across key operations, and FTI Consulting's Global CFO Survey as finding nearly 70% of CFOs rely on AI-driven insight alongside experience-based judgement rather than in place of it. The article said AI can highlight anomalies but cannot intuit organizational nuance, such as a supplier cost increase that reflects a strategic shift agreed in a leadership meeting rather than a contract issue. It said trust needs to be built into systems, with finance teams able to see how an output was reached and clear boundaries on what an AI system can do independently.

The Spendesk article also argued that the biggest obstacle to AI in finance is rarely the model but the data underneath it. Fragmented systems, duplicate suppliers, inconsistent coding and outdated information can make AI accelerate existing problems, it said. Clean, consistent and accessible financial data should come before more ambitious projects, and focused use cases such as invoice processing, expense categorization and early spend-anomaly detection are a more realistic route than trying to transform the entire finance function at once.

A Sifted commentary focused on leadership, saying that as AI agents take over routine decisions, leadership will become harder rather than easier. The easy decisions will be delegated, leaving leaders with ambiguous, unquantifiable decisions that have no clean precedent. The commentary said leaders' errors will become more visible because the human decision portfolio will contain harder calls, human misjudgements will stand out against consistent automated execution and accountability will be more traceable.

The Sifted article said every consequential AI-assisted decision should have a named accountable human, and that saying AI recommended it is not an answer. It also advised leaders to decide what they are optimizing for before automating, tell teams in advance that bolder decisions will sometimes fail, and for investors to reward quick, honest reversals rather than treat public U-turns as weakness. The article said a CEO signing off on financial statements remains accountable for every number even when others prepared and checked them, and that not checking what AI says in one's name may undermine trust.

Editor's Summary

The commentaries describe AI moving from analytics and routine automation into decision intelligence, with adoption already high in UK enterprises and finance functions. They agree that organizations need explicit decision rights, visible accountability and clean data before giving agents more autonomy. The potential payoff cited is faster decisions and more strategic human work, while the risk is misplaced trust, compliance failures and harder leadership calls.