Layer 02 · Automation
74% of PO processing automated. Same-day orders, finally standard.
Enterprise Software · Finance Operations
The Client
Thousands of purchase orders a year, every one a PDF
An enterprise software company sells to organizations whose procurement runs on purchase orders: a daily queue that adds up to thousands a year across new business, amendments, and renewals. Every PO arrives as a PDF, formatted however the customer's procurement system chose to format it, and each one had to be validated against its Salesforce CPQ quote before billing could move.
The Challenge
The team was good at this work. The problem was it was human work being done by humans.
The validation rules were real and important: a 10% amount tolerance, date alignment requirements, specific product line matching logic. But those rules weren't written down anywhere. They were passed person to person through training and institutional memory. New team members took months to ramp. Errors crept in under volume pressure. There was no audit trail.
Our Approach
Codify the rules before writing a single recipe
Before building anything, we mapped exactly how PO validation worked: sitting in on the process, tracing every decision, documenting every edge case. The finding was that most POs fell into one of three categories: clean matches, minor discrepancies correctable within tolerance, and genuine exceptions requiring human judgment. That breakdown shaped the entire architecture. We didn't build a system to automate everything. We built a system to route correctly — the same codify-first discipline behind all of our automation services.
The Architecture
One pipeline in. Three lanes out.
Vision AI extracts each PDF, the engine validates it against the CPQ quote, and every PO routes down one of three lanes — with a field-level audit log underneath all of it.
Design Principles
Three paths, not two
Most automation makes a binary decision: approve or reject. We built three lanes: clean, auto-correct within tolerance, and escalate to human.
Audit trail from day one
Every validation event, pass or flag, gets logged with field-level detail. This is what makes the system trustworthy and defensible in a finance context.
Codify rules before automating
The team's validation logic had never been written down. We documented every rule before building. That document is now also the team's training resource.
Vision AI for variability
Different customers format their POs differently. Vision AI handles that variability without requiring manual data entry.
The Results
A third auto-approving by day four. 74% at steady state.
The routing paid off in the first week. A third of PO invoices were auto-approving within four days of go-live. At steady state, 74% of the manual effort was eliminated, with errors down 90%.
Same-day processing is now the standard. Orders that used to queue behind manual validation clear the day they arrive, and every one carries a field-level audit record — pass or flag.
The client called it out at launch. The same senior finance leader quoted below described the go-live as "a key milestone in our process automation and AI adoption."
"OneSolve cut our PO processing time by ~75% and removed the majority of manual errors."
Senior Finance Leader
What's Next
Rules the business can edit
The validation tolerances that live inside the routing engine today are being externalized to Workato Decision Models, so business users will be able to adjust tolerances and matching rules themselves — no redeploy required. The rules started in people's heads, moved into documentation, then into code. The last step puts them back in the business's hands, this time with an audit trail.
Technology
Got a process living in someone's head?
Bring us the manual, rules-heavy workflow that breaks under volume. We'll show you what a documented, audit-trailed, three-path automation looks like in a 30-minute working session.