Email triage

A growing customer base always goes hand in hand with an increasing number of processes. At Bayerische Landesbank (BayernLB), for example, more than 2,000 emails are received every day for processing credit transactions. And the number is growing! Until recently, employees in the credit department had to process these emails manually.

Is it a request for a repricing? Is it a request for a payment adjustment or a credit extension? Or is it a request for information about a payment to a customer or a payment notification? Employees read all the requests manually and assessed their content. On this basis, they created orders in the bank's CRM system and assigned them to the relevant experts in the bank.

The manual analysis, classification and processing of the ever-increasing number of emails was becoming increasingly time-consuming. In addition, the classification results were increasingly susceptible to human error.

As part of its initiative to introduce Generative Artificial Intelligence (GenAI) and the introduction of the AI Strategy Framework, BayernLB, together with its digitalisation partner adesso, identified mailbox management in credit management as the optimal starting point for the introduction of GenAI in the department. The aim was to develop a GenAI solution that would automate the analysis, classification and recording of credit management processes from emails.

The project team built the application on top of BayernLB's existing Microsoft Azure cloud infrastructure. This enabled adesso and BayernLB to ensure that regulatory requirements for processing customer data were met at all times. Together, they developed an application based on Microsoft Azure OpenAI and the Microsoft Power Platform.

The application automatically creates an entry in BayernLB's CRM when an email is received. This triggers a Microsoft PowerAutomate Cloud Flow, which in turn triggers natural language processing to evaluate the content and extract and process the attachments. The extracted information is added to the previously created ticket in the CRM and assigned to the appropriate contact person.

The extracted data includes:

• The order type: rate setting, repricing, rollover, fee, customer payout, payment advice, syndicate notification, APL, Aval, etc.
• Customer information
• The client ID
• The order related settlement date

As a result, the GenAI-based application has fully automated the entire process of classifying incoming mail. Not only has the processing speed increased, but the consistency and conformity of the evaluation of correspondence and the recording of orders in the CRM system has also been improved. The higher quality of classification and content evaluation has greatly increased the efficiency of the department, freeing up staff time for higher-value tasks.