Objectives

Our objective was to automate the AP process as much as possible to reduce invoice processing time and costs. After the process discovery through multiple calls with stakeholders we had the following milestone to achieve:

Collect invoices from multiple emails.
Extract invoice data with AI
Apply some pre-defined business rules on the captured information
Use some of the captured data points to search for PO into their existing system and do an automatic validation within their finance system.
Send email for exceptions for human checking.

Challenges

The main challenge was to capture invoice data with high accuracy and take those data to the accounting system for further processing.
Multiple stakeholders receive invoices in their email and It was a challenge to create a bot to read each email. There was a probability of receiving invoices in new emails in the future.
Invoices were in different formats and patterns and traditional OCR was not a solution at all. We had to achieve high accuracy for minimum intervention.
The client has many business rules and the existing accounting system had a limited number of APIs and did not have APIs for all the operations.

Solutions

We first created an AS-IS process flow diagram and verified it from the stakeholders. Then we designed the TO-BE process and included RPA bot and pre-trained Invoice processing AI model in the TO-BE process. We automated many manual steps in the TO-BE process with RPA.
Since multiple stakeholders used to receive the invoices, we created a new inbox where all invoices can be forwarded. We created an RPA bot that would monitor the inbox (every 10 minutes) and pick up the attached invoices from the email and send them to the AI model.
Initially the AI model did not give high accuracy but as we got more and more variety of invoices, we kept on fine tuning the model and did some post-processing to get the desired accuracy.
Since the existing software system did not have adequate APIs, we created RPA bots for integration and repetitive tasks automation.

Outcomes

Stakeholders from the client side and technical team worked together with intense communication and that is one of the main reasons for successful automation. The implementation took 3 weeks and then we kept on observing and improving the automation for the next 10 weeks.