What it means to buy artificial intelligence, and how not to get burned doing it
Unlike traditional software, AI is slippery. It evolves over time, often learns from its own use, and may behave differently in one context versus another. This makes procurement of AI systems, particularly generative or predictive models, profoundly different from buying a printer, CRM system, or roads maintenance service. And yet, public bodies across Ireland, the UK, and Europe are increasingly under pressure to integrate AI into service delivery, contract monitoring, planning, and risk modelling.
But how do you buy something that can’t always be explained, benchmarked, or guaranteed to behave the same way twice?
What Are You Really Buying?
Procurement of AI can involve several models:
- Off-the-shelf AI tools (e.g., ChatGPT, Jasper, Midjourney) where licensing, privacy, and user limitations are key.
- Custom-built AI systems, often developed by tech firms using client data to train proprietary models.
- Embedded AI, tucked inside other systems (e.g., AI-driven fraud detection inside a financial management tool).
Crucially, AI is rarely a “product” in the traditional sense. It’s a system, a service, or an evolving process. And that requires a different approach to specification, evaluation, contract management, and even termination.
Key Procurement Considerations
Data Rights & Privacy
AI systems depend on data, often sensitive, personal, or confidential. Contracts must:
- Define who owns the training data and outputs.
- Prevent unauthorised reuse or commercial exploitation.
- Align with GDPR and local data legislation.
Explainability & Transparency
AI systems must be explainable, particularly in the public sector where decisions affect lives. Evaluation should score bidders on:
- Their ability to provide model documentation.
- Clarity on decision-making processes.
- Auditability of outputs.
Bias and Fairness
All AI models reflect the data they were trained on. If you train on biased data, you get biased decisions. Procurement teams should demand:
- Testing protocols to detect bias.
- Regular fairness audits.
- Independent oversight.
Ongoing Monitoring and Updates
Unlike static software, AI models degrade over time. Procurement strategies should include:
- Model retraining cycles.
- Maintenance contracts that cover performance drift.
- KPIs for continued relevance and ethical performance.
Standards Are Still Catching Up
One of the major challenges in buying AI is the lack of harmonised standards. ISO/IEC is working on AI-specific procurement guidance, but most jurisdictions currently rely on patchworks of IT procurement rules, innovation partnerships, and outcome-based specifications.
Until clear technical standards emerge, contracting authorities should:
- Involve multidisciplinary teams (procurement, legal, IT, ethics).
- Use pre-commercial procurement for high-risk or exploratory AI.
- Insist on pilot phases before full-scale rollouts.
- Risk-Based Buying: AI Is Not “Plug and Play”
Contracting authorities need to adopt a risk-based mindset:
- High-risk AI (e.g. decision support in health, justice, or housing): Requires deep scrutiny, ethical vetting, independent testing.
- Moderate-risk AI (e.g. chatbot for planning queries): Should still be explainable and auditable.
- Low-risk AI (e.g. text generators for internal reports): Can be procured with lighter touch, but staff training is key.
Contracting authorities must also think ahead: what happens if it fails? This includes:
- Robust exit strategies.
- Clear intellectual property clauses.
Contingency for data portability and retraining if switching providers.
AI as Assistive Technology: Don’t Just Buy It, Buy Into It
In many cases, the best use of AI in the public sector isn’t automation, but augmentation. AI should support humans, not replace them. Procuring AI that empowers case workers, analysts, or policy teams means:
- Selecting tools that are intuitive, trainable, and flexible.
- Prioritising user adoption and literacy alongside tech performance.
Avoiding systems that create dependency without understanding.
Procurement Skills Need an Upgrade
To buy AI well, procurement professionals must:
- Understand the lifecycle of AI systems.
- Learn to write functional, ethical, and outcome-based specifications.
- Collaborate with technical, legal, and ethical leads from the outset.
But that’s only the beginning. Procuring AI requires a shift from traditional transactional mindsets to adaptive, agile thinking. Procurement professionals must not only interpret complex, evolving technologies but also translate them into clear, enforceable contractual terms. This means developing new competencies in:
- Risk assessment frameworks specific to algorithmic systems, including the capacity to distinguish between low-impact tools and those that carry societal or legal risks.
- Market engagement techniques, such as innovation dialogues or competitive dialogues, to ensure early-stage input from tech suppliers without compromising probity.
- Contract design that anticipates algorithmic drift, retraining requirements, and changing data contexts, clauses must accommodate ongoing oversight, not just one-off delivery.
But teams must also grow their understanding of:
- Ethics in procurement, including how public values (fairness, equity, transparency) are operationalised in AI systems.
- Cross-functional governance, ensuring that procurement sits not just with commercial officers, but alongside AI engineers, data stewards, digital ethicists, and frontline users.
- Supplier capability assessment, with tools and scorecards to evaluate not only pricing and timelines, but explainability, model governance, and post-deployment accountability.
Finally, training should be practical, not theoretical, real-world case studies, simulation-based learning, and procurement labs that model AI scenarios in contract form. The next generation of public procurement isn’t about understanding code. It’s about mastering the art of asking the right questions, defining the right outcomes, and knowing when to walk away.
Final Word: Think Big, Buy Carefully
The potential for AI in the public sector is immense, from detecting fraud to forecasting costs and enhancing inclusion. But its procurement must be deliberate, ethical, and grounded in real understanding. AI may be invisible, but the consequences of buying it badly are all too tangible.
Sources:
European Commission – Artificial Intelligence Act Proposal (2021)
OECD – Procurement of AI Systems: Good Practices and Principles (2023)
ISO/IEC JTC 1/SC 42 – AI Procurement Guidelines (draft standards)
