Monitoring Process Change with Bayesian Methods

NOTE: Our original speaker is unavailable on this date so we will move the Web Analytics talk to later in the month.

Speaker: Mick Cooney

A common business problem is to evaluate the effect of a change of process. A bottleneck has been identified, a process has been improved, or some needless work has been eliminated and you now expect operations will improve, in some sense.

This talk will dicuss a general case for this situation, and discuss an approach for measuring this change. If at all possible it's good to be able to quantify the improvement - ideally in real-time - so that the feedback loop can be accelerated.

Bayesian statistics provides a very natural approach for this situation, allowing the use of simple conjugate priors.

Finally, we naturally come to the discussion of measuring differences between continuous probability distributions, including the Hellinger distance and the Kullback-Liebler divergence.