Problem:

Leading distribution company in North America, with an Accounts Payable operation of  >4.5M invoices* a year with a straight through processing percentage of 35%, not including 10% rejected invoices. Extraction accuracy at an invoice level was less than 30% implying that 70% of invoices needed manual intervention at the extraction level. Downstream, the cost of manually processing invoice was upwards of $2.5 per invoice, leading to high cost of operation, process inefficiencies and compromised vendor relationships.

Solution:

Machine learning extraction and matching models were trained on 3 months of accounts payable data. The system was implemented in shadow mode so as to not disrupt the existing pipeline and to provide a comparative view of the efficiency of the heuristic system vs the machine learning system.

Outcomes: 

Against an existing straight through processing efficiency of 35%, Sapper APaaS was able to deliver straight through processing of 70% in the first 30 days of operation. Extraction accuracy was ramped up from 35% to 80% in the same timeframe. Below outcomes were delivered in the first 30 days, with further improvement due to continuous machine learning feedback.