Rightsizing Service Cost Models, Part II

In my previous article, I used the most complex service I could imagine—a distributed Web-based application running in a virtualized environment—to illustrate that you cannot take a bottom-up approach to achieve fair cost apportionment in a cost-effective manner.

Instead, I recommend using a top-down approach; using the simplest method possible and then progressively drilling down until fair service costing can be achieved with a manageable level of effort. In this article, I will elaborate on that theme and provide some recommendations for drilling down. I will use a familiar example to illustrate how one industry has achieved optimization. Then I will apply the techniques to our use case.

Non-IT Use Case

The fast food industry provides us with an illustrative use-case: determining appropriate costing methods for a sandwich. First, note that they do not measure the exact time it takes to prepare and sell your sandwich (e.g., 35 seconds of cook time, 27 seconds at the register). Nor do they measure the exact amount of onions, pickles, and other condiments that comprise your sandwich.

Measuring onions and pickles for a single sandwich seems ridiculous, but this approach mirrors the naïve, bottom-up approach that many IT departments have pursued in costing their services. Instead, the fast food industry has used a number of techniques to “rightsize” their costing model. Those include determining which cost elements are best dealt with as being direct (e.g., number of sandwich patties, cheese) and which are best dealt with as being indirect (e.g., onions, pickle chips).

Next, they determine what cost elements are best dealt with as being unabsorbed. Many times a statistical proxy is used rather than actual measurements for apportioning absorbed and unabsorbed costs. For example, one might know that, on average, a 10-pound bag of onions with an estimated cost of $X will yield Y sandwiches. Thus, the cost of each sandwich includes an estimated unit cost of $X/Y for onions, whether the consumer want onions or not. The fast food industry has determined that providing a discount for not using onions adds cost and complexity while adding little benefit to customers.

Next, they take unabsorbed overhead (e.g., rent, salaries, electricity) and distributed those costs across all cost items as a percentage uplift. Finally, they apply business logic to modify user behavior and perform demand management. For example, knowing that the profit on French fries is huge, they formulate special meal packages which induce their customers to buy fries by providing an embedded discount on sandwiches. Let’s now apply similar techniques to our IT use case.

Direct Cost Elements

In the case of our IT scenario, we probably have services that utilize dedicated resources. These cost elements are simply assigned to the services they support. This might include dedicated physical servers, software, hardware and personnel.

Easily Measurable Indirect Cost Elements

This might include resources such as the mainframe. Since the mainframe was created to manage a shared resource, usage accounting mechanisms were built in and have bee used by most organizations for over 40 years. Although UNIX and other distributed environments are similar, applications have to be instrumented correctly so that usage by service is captured.

For example, if your applications all run under the super user ID, you will not have the metric data required for accurate apportionment. Network environments are similar, where DHCP is used and not configured in a manner to allow mapping of usage data to specific services. Cost elements that are not easily measurable can be left unabsorbed or absorbed using a proxy metric.