Missiles, UAVs, submarines, soldiers. This is what comes to mind when the public thinks about defence capabilities. However, all that equipment and firepower begins life in a more mundane environment; someone far more familiar to the civilian: the procurement office.
Here, requirements become tenders, tenders become contracts, and contracts lead to end products ranging from ammunition to water purification tablets.
While the equipment and materials used by nations for defence have reached incredible technological capabilities, the procurement process is comparatively rudimentary. It is reliant on brilliant minds, yes, but also pen and paper and tedious step by step analysis. There are so many sensitivities and intellectual property at stake, that it must happen in a controlled environment.
This dynamic has led to criticism. In the UK, Parliament’s Defence Sub-Committee has called the Ministry of Defence’s approach “well and truly broken…siloed and slow-moving”. It references recent studies of how improper decision-making has resulted in costly errors, such as technical issues with the Type 26 frigate and supply chain issues with the E7 Wedgetail underline the operational consequences.
While current procurement approaches across international defence sectors may be bureaucratic, procurers often arrive at the right decision. They avoid the catastrophes that can come with improper due-diligence. The issue isn’t spending too much time to then be wrong, but not getting to the right decision quick enough. A blindspot exists in how secure, agentic AI can transform their processes.
How Agentic AI has transformed procurement
Defence is not the only sector where procurement teams underestimate how far AI has come. Systems have moved far beyond mere generative functions. Agentic systems – whereby humans deploy task-driven ‘agents’ to carry out multiple levels of specific analysis with controlled and informed reasoning – are now pivotal. It is the next wave of significant transformation ready for regulated, highly process-driven decision making.
For procurement teams, the opportunity does not lie in replacing judgement, but gaining speed, scale, and accuracy. Agentic AI accelerates the conversion of information into decision-grade artefacts. Procurement is dominated by reading and cross-referencing vast amounts of specification data and supplier information, supply chains, and historic relationships. Much of that information is unstructured and scattered across separate systems. Making sense of it all still relies too heavily on human expertise having to grind through administrative tasks.
Agentic AI systems can maintain control, and protect sensitive information by assessing data confidentially before the results are seen by human eyes. With the proper set up, systems can ingest data, extract meaning, contrast and compare, and create outputs, all while maintaining a chain of evidence. Once experts have these outputs, then is the time when their expertise should be applied.
So what is “AI-enabled procurement” that truly transforms?
It’s agentic, multi-model, internally managed, and human orchestrated.
As there’s no finite limit to how agentic AI models can be deployed, procurement teams can work on multiple briefs simultaneously. Agents can interact with various workstreams, reducing the complexity of interlocked decision-making. In short, if one procurement decision influences another, say due to soldiers deployed in terrain that has a variety of weather or climate differences, agents understand this. This enables procurement teams to work with the full breadth of knowledge within the decision-making ecosystem, removing the arduous nature of making choices one at a time.
AI systems should also be model agnostic, meaning they aren’t reliant on a singular model. Using multiple AI models removes single supplier dependence. Each AI model has its own strengths, weaknesses and biases. Some handle reasoning or summarisation better, while others excel at coding or visual tasks. Procurement should take advantage of unique strengths, while negating weaknesses or blind spots.
Research from LMsys and IBM shows that routing and ensembling – intelligently sending each task to the most appropriate or cost-effective model – can maintain around 95% of a top model’s quality while significantly reducing cost. In other words, multi-model approaches bring both performance and cost effectiveness. There’s also a governance benefit. Multi-model setups provide natural checks and balances: independent models can verify one another’s outputs, reducing the risk of hallucinations and bias.
To maintain the pace of analysis, control, and security, AI systems should be managed by the procurement workforce and data teams. Using a system that requires an external workforce is slower, and another step in the delivery chain brings natural risk of data mishandling. While there can be sharp learning curves in how to properly implement agentic systems, management in-house, made possible through intuitive, design-first platforms, is optimal.
Finally, keep humans as accountable decision-makers. The best use of AI in procurement is to reduce drudge work and surface what matters so evaluators can make better, faster decisions. Human review is not a “step backwards”; it is how regulated procurement remains defensible.