For many decision-makers in mid-sized companies, AI in procurement is no longer a question of if, but of where to start. What remains unclear is where artificial intelligence actually pays off in purchasing and where a pilot simply ties up attention. The gap is real: in 2024, the UK National Audit Office found that 70 percent of the public bodies it surveyed were piloting or planning AI, while only 37 percent had actually deployed it (NAO).
Below we assess seven realistic use cases and explain why success depends less on the technology than on scope, stakeholders and expectations.
Seven AI use cases in procurement
Effort is a relative estimate. It depends heavily on your data and system landscape.
- Supplier onboarding and document checks: AI reads submitted documents and checks that certificates and evidence are complete and valid. Prerequisites: clear criteria and a leading master data system. Effort: low to medium.
- Free-text purchase requisitions: Employees describe what they need in their own words; AI suggests the item, category and account assignment. Prerequisites: maintained catalogs and categories. Effort: medium.
- Quote comparison: Differently structured quotes become comparable, and deviations in price, lead time and terms are flagged. Prerequisites: consistent evaluation criteria. Effort: low.
- Contract analysis: AI locates clauses on term, liability, price adjustment and termination and compares them with your standards. Prerequisites: centrally stored contracts and defined template clauses. Effort: medium.
- Invoice verification: Invoices are matched against purchase orders and goods receipts, and exceptions are prioritized. With Germany's e-invoicing mandate phasing in for issuing invoices from 2027, invoice data increasingly arrives in structured form (IHK). Prerequisites: clean PO and goods receipt data. Effort: medium.
- Spend classification: Order and invoice lines are assigned to categories, making bundling opportunities visible. Prerequisites: an agreed category taxonomy. Effort: low to medium.
- Negotiation preparation: AI condenses supplier history, terms and market information into a briefing. Prerequisites: reliable data and experienced buyers who can interpret the results. Effort: low.
Scope: one case, one process, one metric
The most common mistake is not the wrong technology but an overly broad scope. Three use cases at once spread data, budget and attention across too many fronts. A better start is a case that occurs frequently, is clearly bounded and has a measurable baseline, such as processing time per supplier onboarding or the share of invoices that need no manual rework.
The evaluation of the Australian Government's Copilot trial supports this: overall use was moderate, and participants saw the greatest value in well-defined tasks such as summarizing and drafting (DTA). Our takeaway: start narrow, measure, then expand. How an AI application connects to SAP as the core system is covered in our article on the target architecture with SAP at the core and an AI app.
Get stakeholders around the table early
AI in procurement touches many functions. Finance owns invoice verification, IT owns interfaces and permissions, business units create the demand, and suppliers provide the documents. Without a sponsor who sets priorities and a steering committee that decides on rollout, a pilot remains an isolated project. Alongside data quality and ageing IT, the UK National Audit Office explicitly named unclear accountability and missing performance metrics as barriers. The Australian trial adds a second lesson: training should reflect people's actual tasks, and questions about data security and permissions must be settled before launch.
This matches our own experience: in a strategy assessment for a mid-sized energy company, we looked at organization, processes and systems together. The results included automated C-parts purchasing, an optimized ordering process and a roadmap for the S/4HANA transformation.
Guardrail: AI prepares, people decide
In all seven cases, the decision stays with people. Approvals, awards and contract signatures continue to follow your control and approval rules. This also makes regulatory sense: following the Digital Omnibus amendment that entered into force in July 2026, the EU AI Act obligations for Annex III high-risk systems apply from 2 December 2027 (Hunton). In our assessment, typical procurement applications usually do not fall into that category. A case-by-case review is still necessary.
Next steps for decision-makers
- Identify processes with a lot of routine work and measure baseline values.
- Choose one use case and set a target metric and time frame.
- Name a sponsor, a steering committee and owners from procurement, finance and IT.
- Clarify data, interfaces and permissions before launch.
- After the first phase, decide based on the metric: expand, adjust or stop.
Digitalizing procurement does not mean starting with all seven cases. Learn more about our approach to AI strategy for procurement. Our procurement assessment shows which case suits you best as a starting point, or you can book an initial consultation right away.