1) Using web site log file data, build up groups of "visit chains/sessions/paths" arranging them by most common vistor's behavior (model needs to be smart enough to ignore some pages in order to put visitors into groups. Useful data may be path and time on page.
2) Arrange visitors into groups based on statistics of hitting certain goals on the site (registration, file download, view a certain page, make payment, etc.). This might require scripting code to log all click-stream/events.
3) Detect most critical attributes in order to tell site admin which are most important (referrer, specific pages viewed, time on certain pages, etc.)
4) Create a model that can predict a visitor's industry based on data about previous visitors (simplistic example: if previous visitor is identified as belonging to "insurance" industry and that visitor performed actions a, b, c, on the site, then a new visitor who performs actions a, b, c is also likely to belong to that industry).