The SEC found that Moody's did not establish effective controls on models outsourced from an affiliate.  The SEC also found that ratings were inconsistent across similar securities.  Even companies that are in the business of building and selling models have trouble with controls and consistent application.

A&M built SCRE to create a machine learning based model to remove bias and consistently apply a better model to predict credit ratings. https://www.alvarezandmarsal.com/insights/introducing-scre-credit-rating-estimation-service-built-ams-machine-learning-platform

SCRE using only 2 years of basic financials was more than 92% accurate within 2 notches at predicting S&P credit ratings.  One of the interesting outcomes when analyzing our misses, we noted that occasionally we showed quarterly changes in rating for a company when S&P didn't.  What if it wasn't a miss, but simply the rating agency deviating from a model implied rating?