Public‑sector buyers urged to scrutinise agentic AI methods, not just vendor branding

agentic ai in public sector is gaining attention as governments consider how to evaluate these systems beyond marketing claims.

Defining the technology

Elsewhen describes agentic AI as “organisational workflows, agents operating behind the scenes to orchestrate tasks, coordinate systems, and carry out complex operational work that used to take hours of manual labour”. The report stresses that the technology can free civil servants to focus on judgment, empathy and strategy, rather than routine processing. This framing moves the conversation from personal assistants – such as email summarisers – to large‑scale public‑service automation.

Regulators map the risk landscape

The UK Information Commissioner’s Office (ICO) outlines the data‑protection implications of deploying agentic AI. In its Tech Futures paper the ICO notes that “understanding its capabilities and the associated risks is essential” and highlights four scenarios for adoption over the next two to five years. While the focus is on privacy and security, the document also signals that organisations will need to assess technical performance and compliance before procurement.

At the same time, ReedSmith reports a surge in regulatory activity. The firm notes that the Competition and Markets Authority (CMA) published a research paper on 9 March 2026 examining how agentic AI may affect consumers and how traders should mitigate risk. The CMA’s analysis flags “potential fines of up to 10 % of global annual turnover” for breaches, underscoring that existing consumer‑protection law will apply regardless of whether conduct is carried out by a human or an autonomous AI system.

The procurement gap

Despite the growing policy focus, none of the publicly available sources provide a concrete framework for public‑sector buyers to compare AI methods and results. The commission’s brief states that “buyers of agentic AI in the public sector have to compare method and results, not only the name on the brochure”, but the research packet itself acknowledges that the claim is not yet backed by specific evidence in the cited documents.

This absence of guidance creates a practical dilemma. Procurement teams typically rely on vendor‑provided specifications, pilot outcomes, and cost‑benefit analyses. Without independent benchmarks or standardized performance metrics, buyers risk selecting solutions based on brand reputation rather than demonstrable efficacy.

What buyers can do now

Given the current evidence base, public‑sector buyers can take three immediate steps:

• Map functional requirements. Identify the exact workflow steps that an agentic system is expected to automate – for example, processing asylum applications or managing hospital discharge summaries. This creates a concrete set of criteria against which any vendor’s claims can be measured.
• Demand transparent pilot data. Require vendors to share baseline performance figures, such as the number of records processed per hour, error rates, or time saved compared with manual handling. The Elsewhen excerpt cites “a hundred planning records per day, versus the usual average of five”, illustrating the kind of quantitative claim that should be independently verified.
• Engage regulators early. Use the ICO’s data‑protection guidance and the CMA’s consumer‑risk framework as starting points for contractual clauses on data handling, audit rights and compliance reporting.

These actions do not replace a sector‑wide benchmarking study, but they help close the information gap while the regulatory ecosystem catches up.

Looking ahead

Both the ICO and ReedSmith anticipate that the next two to five years will see “intense scrutiny from innovators, technology adopters and regulators worldwide”. As the technology matures, we can expect more detailed procurement guidelines from bodies such as the European Commission or national digital ministries. Until then, the onus remains on individual public‑sector buyers to build rigorous evaluation processes.

For a broader view of how emerging technologies intersect with public‑sector finance, see our recent analysis of Germany’s 2027 budget Germany’s 2027 budget earmarks €109.7 bn for defence and the upcoming crypto‑tax regime Germany to impose a flat 25 % tax on crypto gains from 2027.

In summary, while the buzz around agentic AI is growing, public‑sector procurement still lacks the hard data needed to move beyond brochure‑driven decisions. Buyers who proactively demand method‑level evidence and align with emerging regulator guidance will be better positioned to capture the promised efficiency gains without exposing their organisations to unforeseen risk.

Sources: ICO; ReedSmith.