Lucas A. Meyer, Credit Analysis Manager explains, “The volume of questions increases at the end of fiscal periods, especially at quarter and fiscal-year ends, where the demand for ad-hoc status requests and FAQs becomes a significant part of the workload of some participants in this process, like collectors, operations managers, collections managers and credit managers.”
Although each question asked is simple, it’s usually time-consuming to answer, requiring opening and filtering multiple reports with different taxonomies.
Microsoft developed a chatbot. Worldwide Credit Services now allows sales and operations users to automatically check the credit status of a customer’s account before placing orders, and to track the status of orders until they are collected. The chatbot can also recognise more than 30 FAQs and return data from multiple systems.
This was done internally within Microsoft using Microsoft Azure extensively, as well as LUIS, the Language Understanding Intelligent Service, a service of Cortana Intelligence Suite. LUIS translates questions from natural language into a common language that a computer can understand.
Best practice and innovation
Chatbots are particularly useful to connect data from multiple different systems: users now have to connect with only one modern and simple interface that requires little training.
Chatbots are especially useful in processes that span multiple areas. Everybody can talk to the same chatbot and have access to one set of information.
Chatbots can gather and keep context. Although they don’t do this as well as humans, they do much better than reporting systems.
Meyer recalls, “Chatbots are easier to implement than we thought. Our relatively complex chatbot was done in less than two months by one resource. The chatbot is incredibly popular. We didn’t expect as much. Whenever we show the chatbot, we get more and more users, and several requests to demonstrate it to other groups and customers.”
Meyer concludes, “There’s a lot more that will happen with chatbots in finance and treasury and we will be at the forefront of that wave.”
The chatbot is expected to answer 12,000 questions per month in approximately 3,000 interactions, saving the equivalent of 1,800 hours of work per month. In its first two months, without being programmed to answer the full set of questions expected, the chatbot answered more than 2,000 questions.
The automation of this work allows efficiency gains of 3.8% in credit services.
Key learning points
It’s not hard to create a chatbot.
Chatbots excel in answering simple questions or starting simple processes.
Chatbots excel in being a single user interface for multiple systems, and they require little user training.