Three million invoices a year, classified with natural language processing
Deep learning learned an E&P company's invoice filing logic, corrected hundreds of thousands of historical errors and now flags new ones.
Challenge
An asset-intensive E&P operation generated around three million invoices a year, with frequent human error and inconsistency in filing, and a business-analysis team unable to draw granular insight from the unstructured mass.
Solution
Arundo applied natural language processing and deep learning to learn the current invoice-filing logic, used the trained model to correct existing errors and raise alarms on future ones, and used neural networks to find which parts of the categorisation structure generated the most confusion, optimising it with the subject-matter experts.
Impact
Hundreds of thousands of errors corrected in the historical data and a structured invoice system that generates business insight.