Improving the accuracy

Startup providing mobile application to help users to organize their finances. We worked together developing the solution which enables users to gather their invoices in one place and pay them through the application as well as quickly switch to the different service provider based on price and quality comparison.

Problem

One of the application’s functionalities allowed users to scan a paper invoice and recognize its data to pay it through the application. The Invoice scanner had low accuracy - less than 50% of the elements were scanned and recognized properly and some of them were not recognized at all.

Solution

We have analyzed this issue and prepared a solution plan according to the findings. We decided to use Python, Flask, Google Vision and Google Storage. Our developers rewrote the invoice scanner from scratch.

Result

Result

Invoice scanner accuracy was increased to >90% in 8 weeks. From that time all the elements are recognized properly and some of them were reported to be recognized in 100% of cases.