Machine learning techniques are generally applied for solutions covering prediction, classification, fraud detection and anomaly detection use cases.
On prediction, if you look for possible use cases specific to CRM, then you could consider predicting the likelihood of conversion of each incoming lead or predicting when each progressing deal will be won. Typical algorithms include logistic and linear regression.
If you look for CRM use cases on classification, then you could consider utilizing machine learning to auto segment/profile customers into various categories or classify customer emails on the basis of the sentiment or intent it relates to. Typical algorithms include binary and multi-class classifications.
On anomaly detection, you could utilize AI to continuously keep an eye on current pipeline or sales team performance and if any abnormal deviation on performance numbers identified compared to the average expected parameters, then those pipeline deals can be surfaced out automatically for immediate attention.