Autonomous procurement is software that makes and executes procurement decisions on its own – converting a requisition to a purchase order, resolving an invoice exception, processing a change order – and escalates to a person only when confidence drops or policy requires sign-off. Two tests separate it from everything else sold under the label: which decisions the system owns by default rather than routing for approval, and whether it can act across the ERP, supplier portals, and email without a human re-keying data in between. A copilot that drafts a recommendation and waits for a click fails both tests.
The dividing line is delegation by default
Vendors have applied "autonomous" to guided buying catalogs, chatbots that answer questions about spend data, and approval workflows that move requests between inboxes faster. The blunt question to ask in any demo: at 2 a.m. on a Tuesday, with nobody watching, what does this system decide and do?
A genuinely autonomous system carries a delegation policy. Purchase orders under a threshold get created and sent. A supplier master data conflict gets investigated and corrected. A failed three-way match gets traced to its cause and cleared, or escalated with the evidence already assembled – which is what autonomous exception resolution actually means in practice. Autonomy describes default ownership of a decision. Human oversight stays in the design, through thresholds, escalation paths, and audit logs that people set and review.
The second test is cross-system action. Procurement work lives in the ERP, the MES, supplier portals, shared inboxes, and spreadsheets, and a single decision usually needs to land in several of them. Software that decides correctly but needs a person to carry the result from one system to the next has automated a recommendation while leaving the manual work intact. This is the boundary where rule-based RPA in procurement has always failed: it acts, but only inside one system, on inputs that match its script.
Production autonomy in procurement is still rare
Agents are spreading through enterprise software far faster than procurement teams are putting them into production. Gartner predicts that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5% today. Procurement lags that curve: in Deloitte's 2025 Global Chief Procurement Officer research, 92% of CPO respondents were planning and assessing generative AI capabilities, but only 37% were actively piloting or deploying it in procurement. Most of what gets called autonomous procurement today is an assessment or a pilot on one narrow workflow. When a vendor claims customers running autonomously, ask how many are in production and which decisions those systems own without a person clicking approve. Pricing is a quick tell in the same conversation: a vendor priced per seat is planning for people to keep working the queue, while outcome-based pricing is aligned with the autonomy being claimed.
The payoff from real autonomy is capacity: the same team absorbs more volume, and people move from routine queues to supplier strategy and root-cause work. A practical next step is to list which source-to-pay decisions you would delegate first – req-to-PO conversion, PO amendments, supplier data exceptions, invoice matching, GL coding, tax holds – and test vendor claims against that list. Fragment's workflow catalog covers that full lifecycle, with agents that reason across your existing ERP, portals, and inboxes, keep a human in the loop by default, and require nothing ripped out or replaced. A demo run on one of your own queues shows where the delegation line would fall for your team.