Business

Public buyers must look beyond brochure names when selecting agentic AI solutions

Agentic AI can automate whole workflows in government, but regulators warn buyers need to assess methods and outcomes, not just marketing claims.

Sophie Tremblay
Paper Canadian household utility bill that includes a line‑item for an agentic AI government service fee

agentic state is reshaping how public services operate, but buyers face a new challenge in assessing what the technology actually delivers.

What agentic AI promises for public services

Elsewhen’s recent report notes that most public discussion still focuses on personal‑assistant style tools – chat‑bots that draft emails or schedule meetings. The firm argues the larger impact will come from “agents operating behind the scenes to orchestrate tasks, coordinate systems, and carry out complex operational work that used to take hours of manual labour.” It cites examples such as education planning, asylum processing and hospital discharge, where agents can handle “a hundred planning records per day, compared with the usual average of five.” The message is clear: the technology is positioned as a catalyst for freeing civil‑servant time for judgment, empathy and strategy.

Regulators start to focus on the technology

Regulatory attention has risen sharply. ReedSmith reports that on 9 March 2026 the UK Competition and Markets Authority (CMA) published a research paper on agentic AI, flagging risks such as misleading practices delivered through autonomous agents. The paper stresses that existing consumer‑protection law, including the CMA’s powers under the Digital Markets, Competition and Consumers Act 2024, applies regardless of whether conduct is carried out by a human or an AI system, with fines of up to 10 % of global annual turnover for breaches.

The CMA’s guidance, while aimed at traders, is directly relevant to public‑sector buyers because it highlights the need for clear evidence of how an AI system makes decisions and what outcomes it produces. Without that evidence, buyers could inadvertently expose themselves to compliance breaches.

Data‑protection and risk considerations

The ICO’s Tech Futures report on agentic AI adds another layer. It outlines data‑protection implications for organisations deploying the technology, noting that the systems combine generative AI with new tools that can “operate using natural human language and automate more open‑ended tasks.” The report presents four possible adoption scenarios for the next two to five years, underscoring uncertainty around risk management and the importance of early assessment of data‑handling practices.

Because agentic AI can process contextual information from multiple sources, the ICO warns that organisations must evaluate both the technical capabilities and the governance frameworks that control data flow, consent and accountability.

What this means for buyers

For ministries, municipalities and other public bodies, the combined messages from Elsewhen, the CMA and the ICO point to a practical buying problem: the need to compare methods and results rather than rely on a vendor’s brochure name. Buyers should ask for demonstrable performance metrics, pilot results and transparent documentation of how the AI makes decisions.

Key questions include:

  • What specific tasks does the agent automate, and how does that compare with current manual processes?
  • What evidence exists that the system can sustain the claimed throughput – for example, handling a hundred planning records per day versus the historical average of five?
  • How does the provider address data‑protection risks identified by the ICO, such as consent management and auditability?
  • What compliance assurances can be offered under the CMA’s consumer‑protection framework?

Answering these questions helps buyers avoid the pitfall of selecting a solution based solely on brand reputation. It also aligns procurement with emerging regulator expectations, reducing the risk of future fines or legal challenges.

Looking ahead

While the regulatory landscape is still evolving, the trend is clear: agencies will be expected to demonstrate due diligence in evaluating both the technical method and the measurable outcomes of any agentic AI system they adopt. Public‑sector buyers that build rigorous evaluation frameworks now will be better positioned to reap the efficiency gains highlighted by Elsewhen while staying on the right side of the CMA and ICO requirements.

For more on how public bodies are navigating AI procurement, see our earlier guide on AI regulation in the public sector and the analysis of technology adoption trends in government.

Sources: