China is putting AI inside public procurement itself

On February 12, China's National Development and Reform Commission (NDRC) issued a roadmap to embed AI across public procurement. The planned functions cover the full cycle: irregularity detection, bid evaluation assistance, collusion detection, market analysis, document checking, and tools to help bidders navigate the process from their side. Provinces are expected to implement parts of this by end of 2026. Nationwide rollout is targeted for 2028.

This is a meaningful distinction. Procurement is where a huge share of public money either gets spent well or doesn't. Globally, governments spend roughly 12–15% of GDP on public contracts. In that space, the dominant pathologies are bid rigging, opaque evaluation criteria, and information asymmetries that favor incumbents. If AI can detect collusion patterns that human reviewers miss, that's genuinely useful. If it can flag irregularities before a contract is signed rather than after an audit three years later, even more so.

The design principle the roadmap reportedly includes is the right one: AI conclusions do not replace independent judgment. That clause is doing a lot of work. It's the difference between a decision-support tool and an accountability black box. Whether that principle survives contact with implementation pressure, procurement deadlines, and overstretched officials is the real question.

The replicability question is where this gets hard. For any of this to work, procurement data needs to be structured, complete, and consistently coded. Most governments, including high-income ones, don't have that. And if you train a collusion-detection model on historical procurement data that already reflects past biases, you're not cleaning the system. You're encoding its pathologies at higher speed and with greater apparent legitimacy.

The bidder-side assistance function is the one I'd watch. It could genuinely lower barriers for smaller firms and new entrants who struggle with the documentation burden. Or it could advantage whoever has the better prompt engineers. Both are plausible outcomes from the same feature.

China has the data infrastructure, state capacity, and political will to move fast on this. That combination is rare. Whether the model generalizes to contexts where procurement records are fragmented, institutional oversight is thin, and the pressure to produce a procurement outcome fast outweighs the pressure to produce a clean one, that's a different question.

The honest answer is that we don't know yet. The roadmap is the easy part.