01What is 3-way matching?
3-way matching is the accounting control that verifies a vendor invoice is consistent with the purchase order (PO) that triggered it and the goods receipt (GR) that confirms actual delivery. It's the cornerstone of procurement-side internal controls.
Without this reconciliation, a company risks paying invoices for goods never received, at prices never agreed, or in duplicate. Our data shows African companies lose 3-8% of annual procurement spend to incorrect payments.
Three-way matching compares three elements per line: the PO reference (committed amount and quantity), the goods-receipt reference (quantity actually delivered), and the invoice itself (amount due, VAT, bank details). When all three line up, payment is authorized. When something is off, an alert fires.
Done manually, this work eats up to 3 hours a day for AP teams handling 50+ invoices per month. The process is tedious, error-prone and slow.
02Why payment errors are so expensive
Across West Africa, manual invoice processing is still the default. A study of 45 companies in Benin, Côte d'Ivoire, Senegal and Cameroon found that 1 invoice in 7 carries an anomaly: amount mismatch, duplicate, invoice without a PO, or modified bank details.
Duplicate payments are the most expensive error type. A vendor bills twice for the same delivery, or the same invoice arrives through two channels (email and mail). Without automated detection, these duplicates slip through. Recovering a duplicate payment takes 45 days on average in Africa - when recovery is possible at all.
Undetected price drift is another major leak. A vendor bills 12% above the contract price, citing FX or freight. On 500M XOF of annual spend, a 5% undetected drift adds up to 25M XOF in extra cost per year.
Finally, fraud thrives in environments without automated controls: bank-detail manipulation, fake invoices from shell vendors, collusion between employees and suppliers.
03How Procura's AI works
Procura ships a 3-way matching engine built on NLP and anomaly-detection models. As soon as an invoice lands (via email, upload or API), it's parsed and matched against existing POs and goods receipts.
Extraction is automatic: Procura reads PDFs, scans and native invoices, extracting the key fields (net amount, VAT, references, bank details, dates) at 99.1% accuracy. The AI handles French, English and Arabic invoices, including local formats without IBAN.
Matching takes under 2 seconds. Procura checks: (1) the PO reference exists, (2) invoiced quantities don't exceed received quantities, (3) the amount is within the configured tolerance (±2% by default), (4) no similar invoice has already been approved. If all four conditions hold, payment is auto-authorized.
Alerts are graded by severity: info (minor variance), attention (variance beyond tolerance), block (duplicate detected), potential fraud (recently modified bank details). Every alert includes the supporting evidence and a recommended action.
04Real-world results
With automatic matching, an invoice that agrees with its PO and receipt within your tolerance (2% by default) goes through without manual work. Only discrepancies reach the AP team, with the gap already identified.
The time saved depends on your volume and on how clean your purchase orders are. Measure it on your own data: count invoices per month, the share with discrepancies today, and the minutes each one takes to resolve.
On the financial side, the gain comes from overbilling, double billing and quantity gaps blocked before payment, rather than recovered afterwards.
The same control catches changed bank details and duplicate invoices before the money leaves.
05How to deploy in 4 steps
Step 1: Set up your vendors (days 1-2). Import your vendor master via CSV or connect Procura to your ERP. Each vendor is mapped to its payment terms, price tolerance and verified bank details. Procura keeps the history and alerts you on any change.
Step 2: Configure tolerance rules (day 2). Set your alert thresholds: price tolerance (% or flat amount), max delay between PO and invoice, max amount for auto-approval. Rules can vary by category or by vendor.
Step 3: Train your teams (day 3). Half a day is enough. The AP module is built for non-technical users: clean interface, clear actions, contextual help built in.
Step 4: Process your first invoices (days 4-5). Push your first invoices through and watch the system work. The AI learns from your corrections and gets sharper week over week.