In a previous blog, I discussed how to make self-service BI more appealing to the users, given that it allows an organisation to access and work with corporate information without the IT department’s involvement. So following on, I now examine the aspects that need to be considered in order to implement self-service BI effectively. Getting this right can in fact improve the overall decision making process for the company.
To recap, self-service BI is a paradigm shift in the way businesses use and implement their data, by enabling the users themselves to join and analyse large amounts of data. In essence – self-service BI provides a pathway to better decision making across the entire organisation.
Although no-one questions the notion, it takes considerable discipline, tactics and software to implement self-service BI correctly. There’s a lot more to it than acquiring and installing a user-friendly reporting tool. The following factors should be considered when setting up the self-service BI environment.
Complexity
The data environment is complex. There are normally overlapping data sources that hold the underlying data for most of the organisation’s analytic data. It’s important to provide access to all this data, without any limitations. To get all this data into the self-service environment, you may need to incorporate data integration, data virtualisation, master data management and many other aspects of data management. Fortunately, in most organisations, you don’t have to provide access to all the data at once. If you can provide access to some high priority information relatively quickly, you do have a little lead time in which to plan for and address the more complex data. But you don’t have forever…
Physical Independence
You need to configure your self-service environment so that it provides physical independence from the complexity of the underlying data. Business users should not be affected by how the data is physically stored, or even have to know why it is stored that way. Even dimensional models have complexities (like surrogate keys, bridging tables, time-stamps, active record indicators, etc.) that must be masked from the business users. Through technologies like data virtualisation, the self-service environment may even have to provide access to data outside the typical BI environment, and sometimes even to external data.However, you need to set up the self-service environment so that those physical implementation details are hidden from the users.
Variety
You may need to deal with a lot of data complexity in the form of variety as the big data people call it. The self-service environment may need to cater for non-relational and non-dimensional data in the near future. The self-service environment may have to include data from in-memory-based data appliances, operational stores, columnar databases, key-value NoSQL databases and Hadoop. A crucial implementation decision is whether to force structure on such unstructured big data items first, or whether to provide access to the data in its unstructured format (if the toolset in use can be used to do that.) Again, some data virtualisation may be required if you cannot get everything into the data warehouse quick enough.
Data Governance
Behind the scenes, there needs to be data governance – rules that dictate how data should be sourced, cleansed, managed, distributed and used. In most organisations, it’s hard to get data governance initiatives accepted, funded and kicked off, but once the business value of data governance is well understood, the data governance function can mediate on data quality, security, metadata, and more. If effective governance practices are in place, and are being followed, the trust in the data will increase, which in turn will increase the adoption and utilisation of self-service analytics. For one, if you can prove the badness of the data quality before you release it into the self-service environment, you just may have enough ammunition to make the business case for a data governance program, but keep mind –that in itself is a lot of work.
Collaboration
You’re not going to get away from this one in this day and age. Collaboration has become an important requirement for all reporting, dashboarding and visualisation tools, and with self-service reporting it is equally important. You will have to set up and enable capabilities such as embedded discussion forums and workflows through which the user community can easily share results, cooperate and partner to reach business conclusions. Other functionality you may have to provide include star ratings, commentary, interactivity, interfaces to internal social networks and bookmarks.
Training
Self service BI is more about the mentality and the approach, than the actual toolset. Sure, some toolsets support self-service BI better than others. Sure, some are more intuitive and easy to understand and use. But regardless of the toolset, there is training, coaching and mentoring required. You can’t expect to let a bunch of business users loose on a complex enterprise-wide dataset using a new tool, and expect accurate results, well, not very quickly. You would more than likely get disgruntled and dissatisfied business users. So training is necessary and guiding and mentoring is advisable. Think about running demo workshops on your own data once the users have been trained on the toolsets. The functionality of the tool will make more sense to them when they see how the toolset is applied to their own data.
Semantic Layer
For me, the most important aspect in the implementation of self-service BI, is putting a semantic layer in place. A semantic layer displays an intuitive model of the data in terms of commonly understood and standardised business names, thereby hiding the physical table and column names. The semantic layer is used to define a set of business objects such as metrics, dimensions and attributes that represent the enterprise data in a form that is easily understood and consumed by the business users. The semantic model shields the business users from the complexity of how schemas, tables, and columns are physically stored in one or more databases. It insulates the users from the complexity of SQL and the distinctions as to which of the elements in the semantic layer are atomic and which are derived.
However, all the different perspectives required by the business must be represented in the semantic model. The semantic model must capture the nuance of each perspective and represent the data accurately through a naming convention or syntax that provides clarity. When the semantic layer is constructed correctly, users can drag and drop semantic elements at will into reporting templates and have confidence that the results will be correct. The semantic layer ensures that the business users are accessing the correct data sources and that they are using consistent terminology.
Concluding Remarks
As organisations need to make effective use of more and more data originating from increasingly more sources, the business users must find a way to efficiently investigate, analyse and report on all that data. With the ever-increasing demands for information, you often get the situation where the BI team cannot always cope in satisfying those requirements. If we think about it, reporting is the easiest part of the whole BI ecosystem to open up to business users to do themselves. If you pay attention to the points discussed in this post when you implement self-service BI, the business users will be well positioned to meet their own needs and leave the BI team to properly provide the underlying data with a much higher degree of quality.
