Where source-to-pay automation breaks down
Source-to-pay automation fails at the seams: the handoffs between process stages and the gaps between systems that no single tool can see. Automate one stage in isolation and the bottleneck moves to whichever stage you left alone, so the program buys speed in one place and builds a backlog in another. If your S2P investment is underdelivering, this is almost certainly why. Each tool you bought sees one slice of a process that runs from sourcing through payment, and the problems live in the whole.
Faster matching just finds the exceptions sooner
Take the most common first purchase, automated invoice matching. If upstream PO data is incomplete, reqs converted to POs without line detail, change orders agreed by phone and never entered anywhere, all the matching engine can do is discover the exception pile faster. Every mismatch it flags in seconds still lands on an analyst's desk, and the queue now grows at machine speed while resolution runs at human speed. In Ardent Partners' 2025 AP benchmarking research, invoice exceptions (53%) ranked as the top challenge holding AP back, the first time exceptions have topped the list in the study's 19 years. That finding arrives after years of heavy spend on AP tooling, which should tell you what the spend actually bought: faster processing for clean invoices and a clearer view of the mess. Where AP automation stalls covers the invoice-side version in detail.
A bot on one ERP can't see the exceptions between two
The second failure is architectural. Large enterprises rarely run a single system of record. A manufacturer mid-acquisition might have two ERPs, an MES on the plant floor, a dozen supplier portals, and email and spreadsheets holding the edges together. RPA sits on exactly one of those. A bot scripted against your primary ERP will execute its rules without complaint and never notice the duplicate invoice in the ERP you inherited, or the change order that reached a buyer's inbox and nothing else. Suite automation has the same blind spot from the other side: it automates what happens inside the suite and treats everything outside it as out of scope. RPA vs agentic AI in procurement goes deeper on why scripted bots hit this wall.
Exceptions are a property of the whole process
This is the underlying reason point solutions and RPA fail on the same work. An invoice exception is usually the visible end of a defect that started two or three stages earlier: a PO cut without line-level pricing, a supplier record with a stale remit-to address, a receipt posted against the wrong line. Automate the stage where the defect surfaces and the upstream factory keeps producing defects on schedule. The same dynamic sits behind most P2P bottlenecks, and it explains why so many procurement transformation programs stall after a strong first year: the single-stage wins get harvested early, and what remains is cross-stage work the tools were never built to touch.
The next investment has to reason across systems
A practical test for whatever you are evaluating now: can it read the PO in one ERP, the receipt in the MES, and the supplier's email thread, then resolve the exception the way your best analyst would? If the answer is no, you are buying another slice, and the bottleneck will move again. This cross-system work is what Fragment's agents do, across the full source-to-pay lifecycle: req-to-PO conversion, change orders, RFQ administration, supplier master exceptions, three-way match, duplicate detection, credits and deductions, GL coding, maverick spend, tax holds. The agents reason across the ERP, MES, portals, email, and spreadsheets you already run, with nothing ripped out or replaced, a human approving anything consequential, and production in weeks rather than quarters. Your analysts get their capacity back for work that actually needs judgment. Browse the source-to-pay workflows Fragment covers, or book a demo to trace where your own process is leaking.