Techniques for community-based spend analysis in a software as a service computing environment are disclosed. The techniques leverage information available about community spend to provide new insights into a particular customer's spend. The new insights result from receiving a request for an optimization recommendation, importing spend data comprising a plurality of spend data lines, and using a trained deep learning classifier to classify the spend data lines in a set of standard spend categories to compute a first and second aggregate of monetary amounts. An optimization recommendation is sent to the customer based on a comparison of the first and second aggregate of monetary amounts is sent to a particular customer.