All of us are aware that Data Marts are a part of natural evolution towards a bigger data warehouse initiative. It does not mean that one cannot have big bang Data Warehouse initiatives, but they are generally rare. It also does not mean that Data warehouse and data marts cannot co-exist (Refer Data Warehouse vs Data Marts ). When you work on a data mart, it should be more around a business process (like Leads Management, order processing, Customer Service ) even if it touches multiple departments. For example customer service themes could include Quality Management, Product Management and Distribution Management etc. This ensures that you have a single dimensional model and a source of truth for a given theme.

In real life, in the initial stages of data marting, it becomes a challenge for the departments to work as a team to develop a common dimensional model. However, if this challenge is not over-come early, you will end-up with many data marts providing the same set of information.

The question is how to make individual departments to work to gather to make business process based Data-Marts. This challenge exists, when an organization is not thinking of an enterprise level data warehouse. The best way to handle it is to assign the business process champions and encourage them to drive the Data-mart initiatives. For example, while ops, sales and marketing all work on leads management, you may assign someone in marketing as the champion of overall success of leads management. This role will be like a practice head for leads management, who creates guidelines, best practices, tracking mechanisms etc for leads management. This role will be better equipped to drive data mart around leads management process.

This trick is lesser solution to 'head of information management' kind of enterprise level role, which will be driving data warehouse or data-mart initiatives. If you have an enterprise level role, that role can drive the choice of business themes and also the road-map of all DW efforts in the enterprise. In that case the challenge of process-based vs. functional data-mart will not be of much bearing.