AI Procurement Guides
Our AI governance toolkits are designed to equip businesses across New Zealand with the ability to address the most critical issues around governing AI systems. These procurement guides build on the governance toolkits and are designed to support organisations to go through the steps necessary to identify and make their own key decisions when adopting or procuring AI solutions.
We adopted three lenses for these guides:
- Evaluating AI applications, which includes model selection and data protection, but also involves platform selection, integration, and lifecycle management
- Considerations around the data that is used to train (or ground) an AI solution, as well as how to protect the data that goes into and comes out of AI solutions
- Decisions related to the models themselves to inform how an organisation needs to be thinking about model selection
These guides are structured around the key procurement phases of planning, sourcing, managing, and decommissioning. For each phase, the guide includes key questions to ask, the key AI principles that apply, and potential mitigation strategies to address inherent risk. Cutting across all three lenses, we also provide recommendations around key stakeholders to involve in each procurement phase, as well as further external reference material that readers may find useful.
AI Application Procurement Guide
Procuring AI applications involves identifying and selecting end-user-facing solutions that address specific business needs. Applications embedded with AI can bring powerful data processing and decision-making capabilities to business operations. During procurement, organisations must ensure alignment between the chosen AI solution and their overall business strategy, addressing key questions such as adaptability, flexibility, and alignment with long-term goals. Additionally, it is essential to assess a vendor’s ability to provide ongoing support, manage potential gaps in the solution, and ensure accountability for AI system outcomes.
AI Data Procurement Guide
The procurement of AI-related data is a critical phase that underpins the overall success of AI solutions. The data that enterprise AI solutions rely on needs to be representative, unbiased, and aligned with their business objectives. Key considerations include data security, compliance with regulatory standards, international standards, proper licensing agreements to define the use and sharing of the data, and a clear understanding of any intellectual property implications. Ethical issues such as data ownership, consent, and privacy must also be addressed to protect internal data and avoid potential legal or reputational risks. Vendor reliability, ongoing data management, and the need for regular data updates are additional factors that play a pivotal role in maintaining the integrity and accuracy of AI models and systems over time.
AI Model Procurement Guide
Acquiring and using AI models, whether open-source or proprietary, requires careful evaluation of both the technical and ethical implications at both the development and deployment levels: both developers and deployment teams need to understand organisational policies around responsible AI, especially when systems are producing outputs that impact decisions or outcomes. Organisations must first define the scope and intended use of the model and ensure compatibility with existing infrastructure. Considerations such as data privacy, model reliability, and the ethical use of AI should be thoroughly assessed. Vendor responsibilities in terms of continuous support, performance guarantees, and compliance with international industry standards are essential for ensuring the model’s ongoing effectiveness. Organisations should also plan for potential decommissioning, ensuring that data ownership, intellectual property, and long-term sustainability are properly managed throughout the lifecycle of the AI model.
AI Procurement
Successful AI procurement and adoption requires collaboration across various roles and expertise, with each phase demanding specific skills and stakeholder involvement. In this section, we outline key contributors at each stage and provide references for each phase to support further exploration and understanding of the practices and frameworks discussed.
These guides offer considerations across procurement of AI models, data, and applications, helping organisations set up their AI systems for success and ensuring these solutions are aligned with organisational needs and ethical standards.
As is the case with our governance toolkits, these lenses are most informative when viewed together. For example, while the procurement of AI applications may be a natural starting point for many organisations, it is also important to carefully consider the model and data that power these applications.
These guides offer a comprehensive overview of the key considerations and potential risks associated with procuring AI models, data, and applications. By addressing key questions and inherent risks proactively and adopting a holistic, lifecycle-focused approach, organisations can harness the power of AI effectively and responsibly, ensuring alignment with their strategic objectives and ethical standards.