Total cost of ownership

One of the things that always bothers me is that I see and hear about far too many users over-worrying about license costs as opposed to total cost of ownership. Of course I understand that it is easier to get budget for operational expenses as opposed to capital expense but I think this is a short-sighted attitude that results in false economies.

There are multiple examples. One is the body shopping that systems integrators tend to opt for when it comes to development and, especially, testing. When there are tools available for automating large parts of these processes it is going to cost far more in the long term to employ contractors rather than reusable tools. A similar argument applies to hand coding ETL (extract, transform and load) jobs and to manual data cleansing methods (according to Experian, 29% of companies still do this).

Another example is Hadoop. It looks cheap because the hardware is inexpensive and the software is open source (free). But in practice the administration and implementation costs of a DIY approach to Hadoop far outweigh the costs of actually licencing properly configured software and, for that matter, it may be better to actually pay more for your hardware, on the basis that it won’t break down so often and will cost less to administer. In fact, there is a good argument for adopting an appliance-based approach to Hadoop. I looked at this in detail in a paper I wrote last year but would also apply to, for example, Cray, which also has a Hadoop appliance.

A third example, potentially, is the idea that everybody should move everything to the cloud. Well, maybe they should but not without looking at the total cost of ownership involved, as opposed to simply looking at subscription costs: moving to the cloud is a bit like taking out higher purchase, or a mortgage. It may be cost effective but then again it may not be. This was brought home to me while doing a piece of research into the TCO (total cost of ownership) of implementing a business intelligence (BI) solution, which has recently been published.

In this paper I wanted to get a feel for how much it really costs to implement a BI solution: the software itself, the data warehouse, the ETL tool needed to transform and load the data, the data quality tools needed to cleanse the data, the training, the maintenance costs, the administration and so on. There are several things that are interesting within these results: firstly, referring back to the previous paragraph, not all the cloud-based offerings were less expensive than the on-premises solutions. Secondly, there was a huge disparity in terms of licensing costs across the various vendors I spoke to. Of course, discounting can make a huge difference here but some of the suppliers actually gave me average discounted prices to work with rather than list prices. Thirdly, though I did not comment on this specifically in the paper, I found that the license fees for data quality software was disproportionally high: perhaps this is why almost a third of companies still do manual data cleansing? Finally, of course, the basic message of the paper is that you have to think about everything that goes into a solution and look at it holistically, rather than just focus on up-front costs.