Machine learning and AI have catapulted into enterprise awareness, with use cases spanning far more than simple chatbot engagements. Azure Machine Learning (ML) services, for instance, have contributed to such wide-ranging functions as detecting business anomalies, completing marketing sentiment analyses, and conducting predictive analysis for customer relationship management (CRM) tools. Yet the banking industry has begun taking Azure ML services in an exciting new direction, combining AI with biometric analysis to automate check processing.
The finance sector is well-positioned to leverage this innovative combination. The global market for 3D facial-recognition systems is expected to grow at a CAGR more than 36% from 2018-2022. Couple this trend with consumers’ increasing comfort with biometric systems, particularly for banking, and the industry has remarkable potential to leapfrog others in its technology adoption.
AI and biometric identification have enabled digital identity for some time. However, check processing using biometric analyses such as face detection and signature detection (via pattern matching) is a trending use case in the FinTech sector’s AI/ML implementation.
As the number of users/customers for bank increases, potentially increasing the number of operations and transactions, manual check validation and processing can bog down bank IT systems, reduce efficiency, and introduce human errors. Conversely, Azure-based solutions enable large volumes of data to be processed seamlessly and accurately by automating various tasks via both real-time processing and batch processing. The three core tenets of this next-gen check processing are facial recognition, OCR API, and Azure Form Recognizer.
Two-way verification is the current check-processing approach, with the system validating signatures using a digital image stored in the user’s account by two authorizing parties of the bank. Similarly, in an Azure AI-based system, the facial-recognition program compares the users’ current appearance – taken via smartphone camera – with an image stored in the cloud and associated with the account holder.
Users are trained to upload and prepare their pictures using a mobile app, with Azure image-processing techniques accounting for distortion correction, resizing, and skew correction. After uploading five photos with different backgrounds and lighting situations, the user will have trained the model to verify his or her face with better accuracy using the Azure ML model. When performing identifications, the Face API can compute the similarity of a test face among all the faces within a group, returning the most comparable person(s) for that testing face.
Integrating ML Studio-based OCR APIs is a supervised approach in which training and manual correction on recognition accuracy is required to improve the recognition rate of extracted text. By comparison, Forms Recognizer is an unsupervised model that improves the overall efficiency of the processing rate. Combining AI, Azure ML and biometric analysis can be a powerful force next-gen check processing. Bank branches currently average 300 manually processed checks per day. This automated, AI-driven approach can process more than 4,000 per day (~8.33 checks/minute) with an average accuracy of 98%, presuming the training model has at least 40,000 models built in. That level of speed, scale, and efficiency holds great potential for the banking industry and its customers alike.
The authors would like to thank Arvind Rao (Microsoft) and Senthil Gurusamy (Microsoft) for reviewing this article.
