It has all the tools to structure the machine learning problem efficiently and effectively. It has all kind of algorithms - supervised: linear regression, logistic regression, decision trees, random forest, gbm etc , unsupervised: kmeans, dbscans, spectral clustering, optics etc, and dimensionality reduction algorithms . An exhaustive list of clustering algorithms is implemented. It is possible to automate end-to-end model building workflow such as model building, comparison, selection using cross-validation or other approaches, storing the object for scoring or returning the prediction on unseen datasets.
Documentations is very well written - it not only explains the function definition but gives a good background of underlying mathematics used in algorithms.
Their deep learning framework is not as exhaustive as the other open source available software specific for it, but we are not missing out on these features as other open source projects are good alternate options. So one might have to experiment outside scikit if they want to explore more advanced neural network algorithms
It is a very solid tool for machine learning, if you are looking for unsupervised algorithms - it has an exhaustive list of algorithms to support your analysis and model workflows.
We are building predictive propensity models for customers to buy particular services using scikit-learn, and also use it for data preprocessing for deep learning applications.