One of the frequent statements vendors make is “Agile Analytics”. In pitches towards business units, they often claim that it would only take them some weeks to do agile analytics. However, this isn’t necessarily true, since they can easily abstract the hardest part of “agile” analytics: data access, retrieval and preparation. On the one hand side, this creates “bad blood” within a company: business units might ask why it takes their internal department so long (and there most likely has been some history to get the emotions going). But on the other side, it is necessary to solve this problem, as agile analytics is still possible – if done right.
In my opinion, there are several aspects necessary to go for agile analytics. First, it is about culture. Second, it is about organization and third is it about technology. Let’s start with culture first.
The company must be silo-free. Sounds easy, in fact it is very difficult. Different business units use data as a “weapon” which could easily be thermo-nuclear. If you own the data, you can easily create your own truth. This means that marketing could create their view of the market in terms of reach, sales could tweak the numbers (until the overall performance is measured by controlling), … So, business units might fight giving away data and will try to keep it in their ownership. However, data should be a company-wide good that is available to all units (of course, on the need to know basis and with adhering to legal and regulatory standards). This can only be achieved if the data unit is close to the CEO or any other powerful board member. Once this is achieved, it is easier to go for self-service analytics.
Similar like culture, it is necessary to organize yourself for agile analytics. This is now more focused on the internal structure of an organization (e.g. the data unit). There is now silver bullet for this available, it very much depends on the overall culture of a company. However, certain aspects have to be fulfilled:
- BizDevOps: I outlined it in one of my previous posts and I insist on this approach being necessary for many things around data. One of them is agile analytics, since handover of tasks is always complicated. End-to-end responsibility is really crucial for agile analytics
- Data Governance: There is no way around it; either do it or forget about anything close to agile analytics. It is necessary to have security and privacy at control and to allow users to access data easy but secure. Also, it is very important to log what is going on (SOX!)
- Self-Service Tools: Have tools available that enable you to access data without complex processes. I will write about this in “Technology”.
Last but not least, agile analytics is done via technology. Technology is just an enabler, so if you don’t get the previous 2 right, you will most likely fail here – even though you invest millions into it. You will need different tools that handle security and privacy, but also a clear and easy to use Metadata repository (let’s face it – a data catalog!). Also, you need tools that allow easy access of data via a data science workbench, a fully functional data lake and a data abstraction layer. That sounds quite a lot – and it is. The good news though is, that most of that comes for free – as all of them are mainly open source tools. At some point, you might need an enterprise license but cost-wise it is still manageable. And remember one thing: technology comes last. If you don’t fix culture and organization, you won’t be capable to deliver.