Finance automation · FLO Group · 2025 – 2026
Nobody should have to read an invoice
Expense forms and supplier invoices across four companies, read and keyed by hand.
In short
Situation: Expense and invoice handling across four group companies ran on forms and manual SAP entry. Tax exceptions and company-specific rules were applied by hand. Decision: Read with OCR and AI, decide with a rule engine, approve in six defined steps, post to SAP automatically. Put every exception into a rule rather than a person. For energy invoices, replace a brittle screen-scraping robot with AI reading and cost-centre matching. Result: One flow for four companies, with the approval chain and the exceptions (such as the 70/30 non-deductible split) encoded and traceable. My role: Product owner: process analysis, rule definition with finance, prototype iterations, SAP integration design.
- My role
- Product owner · IT
- Period
- 2025 – 2026
- Scale
- 4 companies · expense and energy invoice flows
- Key figure
- 6 approval steps · no manual re-keying into SAP
Situation and complexity
Two flows carried most of the pain. Expense and supplier invoices were photographed, described, approved by email and typed into SAP by finance. Energy invoices for hundreds of stores arrived from many utilities and were handled by a screen-scraping robot that needed fixing whenever an invoice format changed.
Most of the complexity sat in the rules: four companies with different cost-centre structures, tax treatments such as the 70/30 non-deductible split, and an approval chain that depended on amount, type and company.
Options on the table
- Keep the robot and patch it after every layout change
Rejected. Maintenance had become the product.
- Outsource invoice processing
Rejected. The rules are the company's; handing them to a vendor would have meant re-explaining them forever.
- OCR and AI reading, a rule engine, defined approval, automatic postingChosen
The expensive part, the rules, is written once and reused for every invoice.
The decision and how it was organised
The expense module reads the document, extracts the fields, applies the rules (company, cost centre, tax treatment, limits), routes it through a six-step approval chain and posts the result to SAP without a person retyping it. What the finance team knows now sits in the rule engine. When a rule changes, the change is made in configuration and its history is kept.
In the energy-invoice flow, AI reading and cost-centre matching replaced the robot, and the final and reconciliation reports came from the same data. The screen finance uses went through dozens of prototype versions before it was built.
Edge-case document types that would have delayed the main flow stayed out of the first release. Each went into a later phase with a named owner.
Result
Finance tracked processing time and error rate per company.
What went wrong, what I learned
Intent: Cover every document type in the first release.
What went wrong: The exception list grew during analysis. Covering all of it would have delayed the main flow by months.
What changed afterwards: The first release took the high-volume document types, and the rest went into a named phase.