Cognitive Computing in Action to Enhance Invoice Processing with Customized Language Translation
Ying Li, Muthu Muthiah, Arindam Routh, Chitra Dorai · 2017
This paper describes our recent effort to develop a cognitive application to transform invoice processing in P2P (Procure to Pay) process in Finance operations. The context of this problem is with service providers who offer invoice processing services to their global clients. Due to the language barrier, the service provider usually maintain both offshore and near-shore teams, where the offshore team handles the main workload and is usually located in English-speaking countries, while the near-shore team helps in resolving language dependent issues which the offshore team is unable to handle. To reduce the dependency on near-shore resources and improve the efficiency of the offshore team, we have developed a web-based translation application which can provide secure and domain-specific language translation service between multiple languages in scope. The application also contain modules for language trainers to rate translations and provide feedback, which is used to train the language models in an incremental fashion. The translation service can also be integrated into the overall ERP (Enterprise Resource Planning) workflow to facilitate process automation. This solution has been deployed in production in an IBM delivery center and is currently used by the offshore team on a daily basis. Our three-month application usage tracking shows that we can achieve up to 85% translation accuracy at sentence-level between English and French/Spanish, and the offshore team is able to handle 40% more workload.