10. Public Procurement
How Agentic AI redefines how governments buy and what they get.
How It (Doesn’t) Work Today
Procurement is one of the biggest barriers to the modernisation and improvement of public sector performance. It is often criticised for being slow, inefficient and overly formal. Although it was designed to prevent corruption, promote competition and deliver value for taxpayers, the opposite can sometimes be the result.
Simplistic decision criteria prevent informed buyers from making nuanced or holistic decisions. Newer approaches, such as innovation procurement, outcomes-based contracting and as-a-service models, remain niche. Public administration is struggling with a structural mismatch: on the one hand, traditional budgeting cycles and funding rules favour one-time projects and capital expenditure; on the other hand, cloud-based software works best with flexible, ongoing pricing and contract models.
The procurement of technology-based products and services is particularly hindered by outdated processes, cultural inertia and systemic fragmentation. Tenders often specify legacy solutions rather than functional requirements, effectively locking out innovative solutions often provided by startups. The risk-averse nature of public institutions, the complexity of compliance demands, and the lack of digital expertise among procurement officers mean that it is almost impossible for new technologies to gain a foothold.
A Vision for Agent-Driven Outcomes and Procuring Agents
Agentic AI transforms public procurement in two fundamental ways. First, by serving as a core tool to modernise and automate the procurement process itself. And second, by becoming a product governments increasingly need to procure.
Agentic AI for the Procurement Process
Throughout the procurement lifecycle, AI agents augment and, in some cases, replace traditional processes. The system becomes faster, more responsive and more transparent, eliminating the bureaucracy and fragmentation. This transformation plays out across the following phases in the procurement process:
Needs and problem analysis: Rather than starting with a predefined solution, agentic systems define procurement needs based on a clear articulation of the underlying problem. By drawing on structured and unstructured data from various agencies, agents can identify inefficiencies, performance gaps or service failures, and present them as procurement opportunities, regardless of specific vendors or technologies.
Market scanning and intelligence: Agents continuously monitor the landscape of suppliers and technological developments, mapping emerging capabilities against the needs of the public sector. This enables them to identify new market entrants and solutions in real time.
Cross-departmental coordination: Agentic systems can detect similar needs across departments, jurisdictions or agencies, and will automatically flag opportunities for joint tenders or consolidated purchasing. This improves volume leverage and reduces duplication.
Tender design and matching: Once a problem has been identified and the market mapped, AI agents will recommend the most appropriate procurement strategy, whether that be a standard open procedure, a dynamic purchasing system, an innovation partnership or an outcome-based contract. They match procurement demand with supply conditions and regulatory pathways in real time.
Vendor dialogue and bidding: AI assistants support government buyers and vendors throughout the tender process. They answer questions to clarify details, generate customised bid templates and help vendors to frame their proposals so that they meet both formal criteria and real needs. This makes it easier for smaller suppliers and newcomers who are unfamiliar with government jargon to participate.
Contracting and negotiation: Agents autonomously conduct procurement discussions and negotiation on behalf of governments. Much like tools used in the private sector, these agents are given clear parameters, such as price ceilings, risk tolerances and service levels, and negotiate directly with supplier agents to reach mutually acceptable terms. The result is faster, cheaper and often more balanced deals with built-in audit trails and far less room for human bias or misconduct.
The procurement of Agentic AI
Traditional public procurement still centres on input, such as staff-hours, bespoke software development, buildings, and fleets, However, pockets of performance-based and design-build-operate contracts do exist in sectors such as highways and energy. What traditional procurement rarely buys is the end-to-end capability that delivers a public service.
Agentic AI changes that calculus. Agents will plan, execute and self-audit entire workflows that currently require human teams (whether in-house or outsourced). Because the object of procurement increasingly bundles decision-logic, delivery and assurance in a single service layer, whole tracts of government operations become contestable where they never were before. The result is a rapid shift from paying for tools, software integration or personnel to paying for capabilities and outcomes. This can be for example, “process each building permit within five days at ≥97 percent statutory accuracy,” or “answer 95 percent of citizen queries in under 30 seconds with a minimum of 90 percent satisfaction”.
Public procurement, at 11-12 percent of global GDP (approx. USD 13 trillion), already constitutes the world’s largest market, surpassing the automobile and food industries. However, even this figure represents only one third of public sector expenditure. If even half of routine decision-making and frontline service labour shifts to agentic AI outcome-based contracts, the share of public spending open to procurement could grow from one-third to well over half.
Agentic AI productises internal and external services, alongside novel payment triggers:
Frontline service delivery: Whether In-house or outsourced, staff hours turn into a metered service, enabling any public agency to source capacity on demand. Pay per inquiry solved.
Permits and licensing: Case officers and system monitoring become a decision engine. They are paid per compliant permit issued, with bonuses for low error or noncompliance rates.
Monitoring and accountability: Periodic audits and static dashboards are replaced by always-on risk detection service, with part of the cost paid per incident or violation detected.\
Policy implementation: ‘Pay and pray’ policymaking is replaced by outcome contracts with payment tied to verifiable impact metrics.
This in turn contributes to significant wins:
Efficiency and speed: Faster procurement cycles and quicker time-to-outcome.
Value for money: You only pay when value is delivered), helping bend the cost curve in traditionally low-productivity areas like healthcare, education, and justice.
Access to new capabilities: Even small municipalities or agencies can now ‘rent’ intelligence, for instance AI services for legal drafting, citizen communication, or service routing, that used to require elite human teams.
Flexibility and resilience: Governments can test, iterate, and replace underperforming services with minimal disruption.
Transparency and auditability: Procurement shifts from paperwork to performance — with live dashboards, KPIs, and full traceability.
As governments leverage agentic AI in both of the above described ways, they can move effectively towards outcomes while maintaining the unique constraints of public services (e.g. universality, ethical and legal safeguards). This unlocks new forms of public-private partnership; delivering public value no longer requires that services be operated or delivered by the public sector itself.
Key Questions
How can governments safely begin experimenting with agentic procurement within current legal frameworks? What use cases, such as mini-competitions, outcome sandboxes, or low-value micro-services, offer the highest learning value with the lowest risk?
How do we pay for outcomes delivered by autonomous agents without perverse incentives? What contracting structures ensure alignment between public value and automated performance, especially when human oversight is minimal?
What certification, monitoring and update mechanisms are needed to ensure vendor-supplied agents remain safe and reliable over time? Can governments develop continuous-competition models that keep systems up to date without locking in single providers?
Could a government-owned agent run procurement processes autonomously? If so, what governance guardrails are required to guarantee transparency, fairness, and contestability and to prevent institutional bias or capture?
As procurement becomes more automated, how do we retain strategic control over public priorities? What metrics, audit trails, and decision records are needed to preserve legitimacy when contracts are negotiated at machine speed?