Why generic AI approaches really struggle with investment data

For many financial institutions, the first phase of enterprise AI adoption has followed an all too familiar script. Connect a large language model to internal data, layer on retrieval tools, and expect the kind of fluid answers the average Joe has become accustomed to through applications like ChatGPT.

In practice, results inside investment management firms have often fallen short of expectations. That is not necessarily because the models themselves are weak. More often, the issue lies in a