Ownerless Data - Published in www.businessintelligence.com February 2004

Recently, I was working with a client to develop a data stewardship policy. The intention of this policy was to define what the role of stewardship entailed, to clearly demarcate the roles and responsibilities, and to elaborate on the authority associated with data stewardship.

What is data stewardship? Although there are different definitions and wordings, basically, data stewardship is the (frequently voluntary) assignment and acceptance of responsibility for the oversight of management aspects of information, including data quality, security, access control, etc. The stewardship approach is to clearly define roles associated with these functions, the senior management authority over the data sets in question, and the organizational policies for compliance (or rather, noncompliance) with the stewardship roles.

A data stewardship program is essential to any organization that wants to institute an enterprise-wide data quality program. One open question, though, is how the enterprise-wide responsibility is organized. This clearly depends on a few factors - organization structure, historical applications, vertical lines of business. For example, there may be a large pharmaceutical company with many subsidiaries, such as companies that manufacture devices, drugs, or cosmetics. Within each subsidiary there may be different lines of business, and within each line of business there may be different applications. The stewardship roles may be allocated geographically by region, vertically by line of business, or even horizontally by application (e.g., accounting).

Data stewards are then selected from within the groups based on the hierarchical breakdown. Consequently, the senior authority is also selected from within the same hierarchy, whether it is regional, vertical, or horizontal.
The common theme associated with this allocation of responsibility is data ownership. Data stewardship is assigned based on the ownership model associated with the data used within the hierarchy. In this approach, the senior authority for a collection of data sets is the senior manager of the group that "owns" the data, and stewards are appointed from within that administrative collective. This is a good model, with some excellent data quality results. One of the reasons for this, especially with respect to information quality, is that along with the responsibility of oversight comes the authority to correct problems in the processes surrounding the data that introduce flaws or problems. What this means is that when a data quality problem occurs, one can pinpoint an individual who can address the problem and fix it at the source instead of correcting the problem's symptoms.
There is one potential area where this model is flawed, and this area is relevant to the BI manager: ownerless data. By that I am referring to information that does not have a clear association with any specific individual or group within an enterprise, but is used by one or more individuals or groups across the enterprise.
Here are some examples:

• Cross-reference data: Enterprise-wide reference data that has been centralized as an enterprise-wide asset and is replicated to all information clients that reference that data.
• Enterprise data standards: Data definitions and metadata imposed on the entire organization from the enterprise level.
• External data standards: Suppose that some information exchange is governed or affected by data definitions contained within a data standard defined by an external organization.
• Transient aggregation data: When data sets are extracted and are collected at a landing pad for integration or preparation for loading into a data warehouse.
• Input to management reporting information: when reports are generated based on the extraction or collection of information from multiple sources.

These are just a few examples - I am confident that I will hear from readers with more examples. The issue that needs to be resolved, then, is how to associate a stewardship role with data that possibly lies outside administrative ownership. Solving this issue requires some creative thinking: it either means assigning ownership to ownerless data, or dissolving the ownership model when it comes to stewardship.

The tack I would take is the former: assigning an ownership relationship to this ownerless data. The first reason is that despite the potential difficulty of trying to catalog sets of information that exist outside of an organization, or that exist for only short periods of time, the result of the process provides a wider catalog of information that needs management, contributes to corporate knowledge, and also assigns stewardship responsibility at the same time. The second reason is that the data ownership models that already exist within a company are so ingrained in the corporate psyche that it would be a Sisyphean task to attempt to dissolve them.

Business intelligence managers: consider these issues with respect to your projects, since your success may depend on how stewardship is assigned to the kinds of "transient data" that ultimately populates your warehouses and marts.