A process for modeling multivariate Electric Submersible Pump data in a central host system is proposed in to support managing fields by exception by using artificial intelligence models to identify failure modes and operating conditions. The AI model enables operators to immediately identify failure modes and operational conditions, as it is continuously analyzing, facilitating quicker decision making. It also increases the number of wells an operator can effectively manage, and can be used as an educational tool, empowering users to interpret complex ESP trends. Methods, Procedures, Process: The approach to Electric Submersible Pump trend analytics is based on field data observed from over 1400 wells across the United States. Standard trend data for ESPs such as Motor Frequency, Surface Motor Current, Downhole Motor Temperature, and Pump Intake Pressure are considered in the model. A process is described for cleaning and standardizing raw sensor data, detecting anomalous operating conditions, and classifying the anomalies using multivariate statistical analysis. The model recommended is extensible to consider arbitrarily many sensor signals in classifying the anomalies. Results, Observations, Conclusions: Upon sequential iterations the accuracy of operational conditions classifications improved to about 80%, and eventually achieved 90% accuracy after multiple validation cycles on the 130 test ESP wells. We determined the algorithms we are using to classify operating conditions limits the accuracy but increases the meaning to the end user by the way it is presented. There are more advanced algorithms available with the potential of achieving higher accuracy but at a cost of understanding and explaining the results to the user. The broken shaft and gas slugging cases studies presented in this paper showcase the value driven by the model’s ability to identify failures and operational conditions that allow expedited planning of resolution procedures. Thereby reducing the downtime of high production ESP wells and the impact of lost production. The model presented in this paper will continue to expand into more classifications over time. Further work is required to build out recommendations for ‘next-steps’ based on the classifications presented. This will enhance the understanding of why the anomaly is occurring and the steps to take to resolve the problem. Continuing on this path will inevitably lead us to the beginning stages of autonomous control. Novel/Additive Information: The ESP community is adapting to new ways of analyzing trends over time. Circular ammeter charts were the only piece of downhole information available for many years. As downhole sensors became standard in the industry, new variables became available but that also meant the learning curve became exponentially steep overnight. This method for analyzing Electric Submersible Pump trend data is novel in the diversity of its data sources (over 1400 wells representing diverse reservoir conditions and well designs) and in its ability to generalize wells with diverse sensor configurations and levels of data quality/availability.