Friday, 7 October 2011

Why self-service BI is falling short

Having spent some time over the last couple of weeks introducing companies to Sabisu, it's clear that 'self-service' is seen as a big win - though the precise definition of what that means differs.

Every enterprise appears to have the same problems; complex business processes, a wide variety of often proprietary data sources, heavy use of IT expertise in integrating systems so that end users have the data where they want it. These problems result in duplication of data and a dependency on IT that destroys agility. How can you respond to incipient situations when you have to wait on an IT release schedule?

Everyone we've spoken to sees the answer in shifting capability out to the masses; empowering the end users by providing pre-configured reports or cubes where an end user can build report on-demand with recent data. Limited menu, often reheated data, but hot all the same - like fast food. This works to a degree, particularly in a slow moving environment, because you can schedule report generation or cube maintenance and the data will be 'recent enough'.

But it's not quite self-service. There's a long tail of requirements that the pre-built cubes aren't going to satisfy; all the IT department can do is invest more time, money and effort in building ever larger cubes as each new requirement is uncovered. Before you know it the costs associated with BI are spiralling, so the likelihood of tackling non-relational data or proprietary format manufacturing data is slight.

The end-user experience is often not great. User queries get invalidated as cubes are rebuilt (usually for IT reasons). Reports generated by different users don't tie up because different fields from different systems are confused - and implementing a data dictionary is often not viable, even if you can get cross-department agreement on a single version of the 'truth'. End users have to become proficient in what is, in effect, a development environment for building reports.

All this points to partial adoption at best on the grounds that the service just isn't great. It's got to be better.

I'm looking for:

  • End user driven platforms, so no IT involvement - particularly none that could invalidate a trusted, end-user designed report
  • Genuine end-user driven data access without needing to train everyone first - it's got to be built around modern UX principles
  • Real-time, or as near as makes no difference
  • Direct access to source data - if we're going to have a debate about the data, let's at least be clear on what we're looking at
  • Some way to action BI; curate it for a community, make it actionable, collaborate on it
  • Controllable expense - there's no way the enterprise should be penalised with increased expense or complexity for a user wanting to extract or share data

No comments: