Procurement process mining: why your source-to-pay system is hiding more inefficiency than it reveals

Your procurement system reports cycle times, approval rates, and compliance percentages. The question is whether those numbers describe what actually happens — or what the system was designed to do. In most enterprises, the gap between designed process and executed process is larger than leadership assumes. Accenture reports a 75% reduction in procurement cycle time after using process mining to identify bottlenecks and non-conformance in its own operations — which implies the starting baseline was carrying three-quarters of waste that nobody saw.

The designed process is fiction

Organizations design procurement processes with the expectation that work flows smoothly from start to finish. In reality, everyday execution looks very different. SAP Signavio's research on process mining in procurement documents the gap: manual steps, exceptions, and system-level deviations slow down procure-to-pay cycles and create compliance risks that standard dashboards never capture because dashboards measure throughput, not conformance.

Process mining reconstructs the actual end-to-end flow from ERP event logs — requisition timestamps, purchase order creation, goods receipt, invoice arrival, payment. Instead of asking "what does the process manual say?", it asks "what did 10,000 transactions actually do?" The answer is almost always a set of variants: a few that match the designed process and many that do not. SAP Signavio's analysis of procurement operations across its customer base shows that the number of process variants is typically 3-5x higher than the number of formally documented workflows.

"Process mining ensures transformation starts from reality, not guesswork. It shows which workflows are stable enough for automation and where compliance gaps could undermine resilience."
— SAP Signavio, Why Process Mining Matters

Four things process mining finds that your S2P dashboard misses

Conformance violations. A purchase order created after the invoice arrives. A goods receipt recorded without a matching PO. A contract price override with no documented reason. These are not edge cases — SAP Signavio found that credit checks were bypassed in nearly 30% of orders in order-to-cash processes, and procurement equivalents are similar. Process mining runs conformance checking against every transaction, flagging each deviation against the target model. It does not sample. It audits the entire population.

Maverick spend at the transaction level. Most maverick spend measurement looks at total value by supplier category. Process mining traces each maverick requisition back to its source — the business unit, the requester, the vendor that fulfilled it. Celonis documentation shows that organizations using this approach can automatically block repeat offenders, reject future maverick purchases from the same requester, and flag vendors that systematically fulfill off-contract orders. This is enforcement at transaction granularity, not annual review granularity.

Bottlenecks with dollar values attached. Stalled purchase orders, goods receipts created earlier than planned, freight settlement delays, invoice exceptions — process mining tools like Celonis and Signavio attach cycle time and cost data to each bottleneck. Fresenius Kabi used Celonis to identify that its cash discount realization rate was stuck at 61%. After targeting the payment-term and invoice-processing bottlenecks that process mining exposed, the rate rose to 90%, generating $550K in capital cost savings through improved Days Payable Outstanding.

Rework loops that inflate touch time. How many purchase orders get changed after creation? How many invoices require manual correction before posting? Process mining exposes these loops. Johnson & Johnson reported a 30% reduction in touch time and a 40% reduction in price changes after using Celonis to redesign delivery processes. The rework was invisible before because each individual change was small — but aggregated across thousands of transactions, it consumed hours of category manager time that should have been spent on strategic work.

The automation trap: making bad processes faster

This is the most significant finding from procurement process mining implementations. Celonis explicitly warns that one of its most valuable outputs is identifying processes you should NOT automate. When a procurement process has high deviation rates, automating it without redesign simply accelerates the wrong behavior. The industry term is "making bad processes faster" — a phrase that emerged during SAP's acquisition of Signavio and has become a standard caution in process mining literature.

Independent research on procurement automation reinforces this. AI Multiple's analysis of process mining use cases recommends combining process mining with RPA — not as a one-time assessment but as a continuous monitoring loop. Post-automation, process mining measures whether the RPA actually reduced cycle time or simply shifted the bottleneck downstream. This feedback loop is what transforms automation from a capital expense into a managed improvement program.

The four-step process mining cycle for procurement

• 1 Extract Pull event logs from SAP, Oracle, or your ERP — requisitions, POs, goods receipts, invoices, payments. Celonis and Signavio provide prebuilt connectors.
• 2 Discover Reconstruct the actual process map. Measure cycle times, identify variants, quantify rework loops. Compare to your designed workflow.
• 3 Analyze Run conformance checks. Flag maverick spend sources. Map bottlenecks to dollar impact. Identify automation-safe vs. redesign-required zones.
• 4 Act Trigger automated remediations, block maverick requesters, update planning parameters. Monitor post-change metrics. Repeat the cycle.

What enterprise procurement should do next

Three actions that separate organizations using process mining as a one-off analytics project from those building it into their procurement operations:

1. Run a baseline before any automation initiative. If your organization is planning an RPA deployment, a new S2P module, or an ERP migration in the next 12 months, run a process mining baseline first. SAP Signavio's implementation research shows that organizations that mine before migrating avoid the most common ERP failure mode: replicating broken processes in the new system. The baseline takes 60-90 days for a full-cycle P2P analysis and costs a fraction of what an automation project with the wrong scope would waste.

2. Audit your top three bottleneck categories by dollar impact. Do not start with all categories. Pick the three procurement categories with the highest transaction volume, the most PO changes, or the longest approval cycles. Run process mining on those categories for a 6-month window. The State of Oklahoma did this for its entire procurement operation and cut audit cycle times by 64 days by targeting the specific bottlenecks process mining revealed — not by applying a generic performance improvement program.

3. Build a continuous conformance monitoring function. Process mining is not a project. The platforms (Celonis, Signavio, Microsoft) all support continuous monitoring — new data feeds in, dashboards update, alerts trigger. Assign one procurement analyst to own the monitoring function. The job is not to produce reports. It is to identify deviations as they emerge, determine whether they are process failures or new valid variants, and trigger corrective action before the deviation compounds. The organizations that do this — Accenture's internal procurement, Fresenius Kabi, Arkema — report that the continuous monitoring catches 2-3x more value leakage than periodic project-based analysis.

---