A ministry of health in East Africa releases a tender for medical supplies worth several million dollars. Three hundred vendors submit bids. The procurement team, under pressure to complete the evaluation within a tight deadline, deploys an AI system to analyse the bids, score the vendors against predefined criteria and recommend an award. The system produces a ranked list. The procurement officer signs off. The contract is awarded.

Six months later, an audit reveals that the winning vendor had a conflict of interest that the AI system did not detect. The second-ranked vendor challenges the decision, arguing that the evaluation criteria were applied inconsistently. The procurement officer cannot explain why the system ranked Vendor A above Vendor B, because the system's reasoning is not transparent. The vendor community cannot audit the decision. The public cannot verify whether the process was fair.

This is not a hypothetical scenario. It is the operational reality of AI deployment in public procurement across Africa, where the technology is being adopted faster than the governance frameworks needed to make it accountable.

AI in public procurement is not a technology problem; it is an accountability problem. The question is not whether AI can improve efficiency, reduce corruption or accelerate decision-making. The question is whether the organisation can explain, audit and defend the AI's decisions when they are challenged.

The procurement challenge in African public services

Public procurement in Africa faces a familiar set of operational constraints: limited staff capacity, tight deadlines, complex regulatory requirements, fragmented data systems and persistent risks of corruption, favouritism and inefficiency. A 2025 analysis by the African Development Bank found that procurement delays and inefficiencies cost African governments an estimated 5–10% of project value annually, with the greatest impact on health, education and infrastructure programmes.

AI offers a compelling proposition: automate the repetitive, data-intensive parts of procurement—bid analysis, vendor scoring, compliance checking, contract monitoring—and free procurement officers to focus on judgement, negotiation and relationship management. The technology exists. Natural language processing can extract information from bid documents. Machine learning can score vendors against technical and financial criteria. Predictive analytics can flag risks such as supplier default, price manipulation or conflict of interest.

But the technology does not solve the governance problem. If an AI system recommends a vendor and the recommendation is wrong, biased or corrupt, who is accountable? If the system's reasoning cannot be explained to a challenger, an auditor or the public, how can the process be trusted? If the system is trained on historical procurement data that reflects past biases, how can it avoid reproducing those biases?

The accountability gap

Most AI deployment frameworks in public procurement focus on the technology: which algorithms to use, which data sources to integrate, which performance metrics to optimise. The accountability model is an afterthought. Organisations assume that if the AI system performs well on historical data, it will perform well in production. They do not design for the scenario in which the system's decision is challenged, audited or found to be wrong.

The Economist Intelligence Unit's analysis of Africa's procurement revolution highlights this gap. AI will not replace procurement professionals, the analysis argues, but it will elevate them—provided that organisations establish clear governance over how much decision-making authority is delegated to AI systems and who remains accountable when AI-generated recommendations lead to adverse outcomes.

That is a governance design problem, not a technology problem. It requires answers to questions such as:

• Can the AI system's reasoning be explained in terms a vendor, auditor or citizen can understand?
• Who is accountable when the system's recommendation is wrong: the procurement officer, the system vendor, the data provider or the organisation?
• How are conflicts of interest, bias and corruption risks detected and managed in the AI system itself?
• What is the process for challenging, auditing or correcting an AI-generated procurement decision?
• How is the system's performance monitored over time, and what triggers a review or recalibration?

Most African public procurement organisations do not have clear answers to these questions. They are deploying AI systems without the accountability infrastructure needed to make them trustworthy.

What a governance-first approach looks like

A governance-first approach to AI in public procurement does not slow down deployment; it makes deployment sustainable. The starting point is not the technology; it is the decision that the AI system will influence and the accountability model that will govern it.

Discover. Map the procurement decisions that the AI system will influence: vendor pre-qualification, bid evaluation, contract award, contract monitoring, supplier performance assessment. For each decision, identify the stakeholders who need to understand, audit or challenge the decision: procurement officers, vendors, auditors, citizens, regulators.

Design. Define the minimum transparency, explainability and auditability requirements for each decision. Specify who is accountable when the system's recommendation is wrong. Establish the process for challenging, correcting or escalating AI-generated decisions. Design the system to produce human-readable reasoning, not just scores and rankings.

Build and test. Integrate the AI system with existing procurement workflows, data systems and audit processes. Test the system against historical procurement decisions to identify biases, errors or gaps. Involve procurement officers, vendors and auditors in the testing process. Validate that the system's reasoning can be explained to a non-technical audience.

Enable and monitor. Deploy the system with clear documentation of its capabilities, limitations and accountability model. Train procurement officers not only to use the system but to understand its reasoning, identify its limitations and escalate concerns. Monitor the system's performance over time, including accuracy, fairness, bias and user trust. Establish a process for regular review and recalibration.

The African context

Africa's procurement landscape is distinct in several ways that affect AI deployment. Public procurement systems are often fragmented across ministries, agencies and levels of government, with inconsistent data standards, limited digital infrastructure and varying levels of capacity. Procurement regulations differ across countries, but most emphasise transparency, fairness, value for money and accountability. Corruption risks are well-documented, and public trust in procurement processes is often low.

These conditions mean that AI deployment in African public procurement cannot simply replicate approaches from high-income countries. The technology must be adapted to local languages, data availability, regulatory requirements and institutional capacity. The governance model must be designed for the African context, where procurement decisions are often politically sensitive, vendor communities are vocal and public scrutiny is high.

The Kenya Artificial Intelligence Strategy 2025–2030 recognises this challenge. It emphasises the need for AI systems that are transparent, accountable and aligned with national values and regulatory requirements. But the strategy is a policy document, not an operational framework. The gap between policy ambition and operational reality remains wide.

A practical starting point

Organisations considering AI deployment in public procurement should start with a narrow, high-value use case: a specific procurement decision that is repetitive, data-intensive and currently slow or error-prone. Examples include vendor pre-qualification for recurring supplies, bid evaluation for standard goods, or contract monitoring for service delivery.

The goal is not to automate the entire procurement process. The goal is to demonstrate that AI can improve a specific decision while maintaining the accountability, transparency and trust that public procurement requires. Once that is proven, the organisation can expand the use case, refine the governance model and build the institutional capacity needed for broader deployment.

Xelius supports public-sector organisations working on AI deployment in procurement through governance design, system evaluation, workflow integration and implementation support. The starting point is not a technology purchase; it is the procurement decision whose inefficiency, opacity or risk is currently holding back service delivery.

The standard is explainability, not efficiency

AI in public procurement will not succeed on the basis of efficiency gains alone. It will succeed when procurement officers, vendors, auditors and citizens can understand, trust and challenge the AI's decisions. That requires a governance model that makes accountability explicit, transparency non-negotiable and explainability a design requirement.

African governments have an opportunity to set a global standard for AI in public procurement: not the most advanced technology, but the most accountable. The question is whether they will seize it.