AI is shifting procurement from document work to constrained automation
AI tools are gaining ground in procurement and supply chains, but survey data shows most teams remain in pilots or limited workflow redesign.
By Ingrid Halvorsen · Venture Capital Reporter
· 3 min read
AI procurement supply chains projects are concentrating first on document-heavy analysis and bounded decisions, rather than end-to-end autonomous operations. In an HFS Research and Art of Procurement survey of 73 senior procurement executives, 38% said their teams were still running pilots or proofs of concept, while 26% had redesigned selected workflows and 15% had rebuilt procurement around AI.
The survey is a small, enterprise-heavy sample: 90% of respondents were vice-president level or above, and 78% worked at companies with at least $1 billion in annual revenue. Its results are a useful indicator of adoption maturity among large-company procurement leaders, not a market-wide rate.
How is AI changing procurement and supply chains?
Procurement’s near-term use case is turning dispersed information into something teams can act on. Contract creation and analysis was the most active use reported in the survey, at 26%, followed by category strategy and intelligence and spend analytics, each at 18%. Purchasing and transactions accounted for 12%.
That fits the data problem inside direct procurement. Robin Van Aeken, chief executive of automation company Magentic, told Sifted that legacy ERP systems may preserve the final price paid without retaining negotiation history or supplier-performance context. Magentic says its software extracts information from emails, documents, files and CAD drawings to check purchase-order pricing, compare contracted options and identify possible renegotiation opportunities. Those capabilities and the company’s reported customer results have not been independently verified; the Sifted feature was sponsored by venture firm N47.
Across supply-chain operations, vendor and consultancy material points to similar information-processing tasks: demand forecasting, inventory planning, supplier-risk monitoring, route adjustments, warehouse work and predictive maintenance. Amazon Business describes systems that combine internal data with signals such as weather, supplier performance and transport delays to support planning and sourcing decisions. These are descriptions of potential applications, not evidence that they improve results in every deployment.
Why are most teams still keeping people in the loop?
Automation becomes harder when a recommendation requires commercial context or carries material risk. Catalant says supplier monitoring can combine internal systems with public signals such as credit downgrades, litigation and executive departures, but indiscriminate alerts create false positives. Procurement staff still need to judge which supplier failure would threaten production and whether an alternative is viable.
The same applies to contracts. AI may surface renewal dates, pricing provisions and obligations from a repository, but deciding whether a clause creates negotiating leverage or operational exposure depends on category knowledge and the counterparty relationship, according to Catalant.
Data quality, organizational change and cross-functional integration remain obstacles, according to AcademyGlobal’s summary of third-party supply-chain research. The practical operating model is therefore constrained automation: let software retrieve, compare and flag information, while people approve consequential sourcing, contract and disruption decisions and retain manual capability for critical situations.
The evidence base is limited. Much of the available material is vendor, consultancy, sponsored or trade-publication content, and several cited statistics rely on underlying studies not included here. The clearest supported conclusion is that AI is being applied to specific workflows, while broad claims of autonomous supply chains remain ahead of demonstrated adoption.
This story draws on original reporting from Sifted.