DIGITALISATION Q2 2021 1
Review > Advisory meeting > > > > > Swedish Rules
In-built feedback loops create valuable data and documentation for dialogue, and the refinement and development of supplementary AI solutions. Advisory services with AI in the cloud A concrete example of how we have explored and used AI to create value for the business, with the intention of spending more time with our customers instead of on administrative tasks, was when we began looking at NLP. The work began several years ago, in close collaboration with the operations. We believe that we create the most business value when innovation permeates the entire organisation, compared with restricting innovation to a separate lab. In this way, we have successfully taken trends within new technology to production and scaled up to all branches in Sweden. Considering our generally low risk appetite, but also our desire to experiment and investigate new technology, we chose initially to apply the technology in an internal process, with an AI technology we deemed mature enough to provide practical benefits. In the advisory review that takes place after our customers have received investment advice, we saw opportunities to investigate the technology and to create business value for the Bank in a relatively short time. The documentation drawn up is quality assured according to multiple criteria in a subsequent process. Through close co-operation with the branches and access to suitable new environments, the first version was put into production after approximately eight months. The AI solution particularly supports employees by assessing which documents require additional review by a human. That is, someone with the competency both to assess the quality of the advice and documentation, and to provide feedback on the assessment made by the AI solution. By systematically generating data points where the AI’s assessment and the human advice reviewer’s assessment are compared, the conditions are created to improve future assessments made via both AI and human. Since then we have used data from the solution to add supplementary AI models, and today a cross-functional team of AI and humans works together to review advisory documentations. One concrete consequence is that employess at branches no longer need to devote as much time to reviewing these documents. The technology and lessons learned have also been implemented in other initiatives and processes. The most recent step was incorporating NLP for the advisory review into a cloud-based environment. This has resulted in a steep drop in the administrative costs for the solution, and means that ongoing adjustments to code are much simpler. HANDELSBANKEN 15