AI has taken center stage, redefining efficiency and accuracy in processes that were traditionally labour-intensive and prone to human error. An example of this innovation is our newest tool, IRIS, designed to streamline the way that businesses handle invoice processing. Through a Q&A session with its two creators, Chris Develin and Tim Buric, we uncover the inspiration, challenges, and technological milestones behind this pioneering solution.
We were looking to build upon our AIM platform to provide a number of standard re-usable integrations that could be self-deployed by customers with minimal external assistance. Our experience in the ERP/Finance sector and having worked with many document scanning solutions provided us with insights into some of the challenges customers faced with processing invoices. The scanning market is abundant, and we were looking to build a low-cost solution that would allow budget constrained customers to also use this technology.
Gathering of a large dataset of invoices for training AI is crucial, however obtaining this data can be challenging. As our focus was for an affordable solution that we could bring to market relatively quickly, we opted for Microsoft Azure AI Document Intelligence services using pre-built models. During our testing, we found the accuracy of the prebuilt model on most image elements to be high, >99%. This allowed us to focus on invoice elements where we can add additional rules-based validation to increase the overall accuracy of the final extract.
We ensured that we used common interfacing methods for exchanging data between systems, such as email, Rest API and Secure File Transfer for legacy systems. Validation of data is key to being able to successful process a data file produced from an image. We leveraged our expertise of Unit4 ERP's payment processes to incorporate added validation routines to increase the accuracy of the data import files producing a high probability that the file would be imported and matched successfully.
There have been studies indicating that data entry errors can range from 1-5%. For companies processing large volumes or high value payments, errors can be costly. By automating the process of matching invoices to payments, it removes the risk of human error in manual data entry, such as typos or misplaced decimal points which can lead to significant financial discrepancies. Data entry is repetitive, and an AI process can perform the same task consistently while ensuring a high level of accuracy in data extraction and processing. Other benefits include rules-based data validation, detailed logging for audit purposes and over time, an improvement in it performance, particularly with AI models that can learn from their mistakes.
The Microsoft ISV Success program provided us all the resources necessary to assist with the publication of IRIS on Microsoft's AppSource. Microsoft provided numerous support resources to both our technical and marketing teams to ensure the publication process was smooth.
Expanding on the number of ERP and Finance systems is an obvious one, as is extending the image formats currently supported. Market and customer feedback is an important input to our product development lifecycle so we're eagerly awaiting to hear how customers have benefited from using IRIS and what challenges they still face.
Interested in learning more about IRIS and what it can do for your business? Check out our website!