# What are the steps of the cost estimation process?

To estimate costs, it is useful to follow a well-defined sequence of steps. You must first decide which costs you want to model over which period ($$TC_p$$). Depending on managers’ needs, this can range from all operating costs over a year to specific cost items over short periods of time, for instance daily maintenance costs. As we will discuss later, the chosen time window (from a starting date to an ending date) and period (hour, day, week, month, quarter or year) have a huge impact on the quality of data as well as the meaning and usages of cost functions. Therefore it is important to define them at the very beginning.

Once the costs, time window and period have been defined, you should identify its underlying cost driver(s) within that time frame ($$Q_p^d$$). Indeed, cost drivers depend not only on the type of cost but also on the time window and period. As we will see later, some costs which do not change over short periods of time can change over longer periods of time, for legal, contractual and technological reasons. The search for the underlying cost drivers is both qualitative, listing all the potential causes affecting the level of the selected cost, and quantitative, assessing correlations between cost drivers and cost levels.

Another choice which is related to, but different from, the choice of the time window and period is the range of activity, i.e. the volumes of the cost driver(s), for which we want to build a cost function ($$[Q_{min}^d,Q_{max}^d]$$). This range of activity is called the relevant range. Estimates are likely to differ between low, moderate and high levels of activity and thus different cost functions should be built for each range. A linear cost function is only valid for a specific range which has to be systematically specified.

Data (i.e. the historical values taken by $$TC_p$$, $$Q_p^d$$, $$V_c^d$$ or $$FC_p$$) must also be collected. Data collection is a crucial and time consuming phase of the cost estimation process which determines heavily the cost estimation technique which can or should be applied and the quality of the resulting estimates. It is therefore important to also check the quality of raw data and to clean them before even starting applying estimation techniques.

Once the cost, time window, period, cost driver, data, and relevant range have been specified, you can apply one or several cost estimation techniques to estimate the missing parameters of the cost function. Depending on the data at hand, specific methods may have to be applied and sometimes several methods have to be combined. Moreover, since different methods have different strengths and weaknesses, they are often used simultaneously to check the reliability of their respective estimates.

This brings to the final steps of use and validation of cost functions. These two are so closely intertwined that it is difficult to really separate them. Indeed, you need to validate your cost functions to use them, and you need to use them to validate them. And when you observe discrepancies between expected and observed costs, this may suggest a problem with the forecast, the performance, or both.

In the subsequent sections, we will address all these issues successively. But before that, we need to understand cost behavior.