The document discusses developing a unified approach for fitting loss models to insurance data. It proposes using a limited set of standard distributions, including the exponential, gamma, lognormal and Pareto distributions. These distributions could be combined in mixtures to better fit different parts of the data. The document presents an example of fitting a mixture of a lognormal and exponential distribution to insurance payment data. Graphs and statistical tests show the model provides an acceptable fit to the data. Automating the process of fitting mixtures and comparing models is an area for further development.