Predict the probability that a Deal is won given a set of factors. Progress is being maintained on the wiki.
My goal is to assign a probability of a Deal closing as won, no matter how far along it is in the negotiation process, given a set of known features (or attributes).
The following data elements can be derived from internal systems, and are likely good predictors.
My plan is to capture the features listed above with periodic snapshots of our open deals every 60 days. These periodic snapshots will be kept in separate files with each row representing a single deal. As the deals in these snapshots are closed over time, I will mark as won (1) or lost (0) and add them to an accumulating file. The end result is a dataset we can use to train a deal probability prediction model.
The statistical methods used for generating this model/algorithm are yet to be determined. However, it should be a method that favors producing outputs between 0 and 1.
To make sure the model continues to improve, the periodic snapshots and posting to the accumulating file will continue to happen after the original algorithm is built. Re-training the model against the growing dataset will effectively increase its accuracy.