Improving Cost Estimating for Maintenance Turnarounds

We must do better, says Gordon Lawrence

THIS article is the fourth in a series of six, discussing aspects of maintenance turnaround preparation. It discusses the inaccuracy of turnaround estimates and the general over-optimism regarding accuracy range (precision) in the estimates. It then looks at a few simple ideas for improvement, including the definition of estimate classes, methods for assessing uncertainty and risk, documenting the estimate, and the gathering of data on actual hours and costs at the end of a turnaround to form a database for the future.

The Importance of Cost Estimating for Turnarounds

Maintenance turnarounds (sometimes called shutdowns or outages) are significant events in the long-range plan of any refinery, petrochemical plant, offshore production asset or other facility that uses large, continuous production process plant. They consume a lot of time, money and resources and they represent a considerable lost production opportunity while the facility is shut down for this essential maintenance, inspection and cleaning work. 

To try and get to the root causes of why turnaround estimates are neither accurate, nor precise, we need to look at several elements

The leadership team on a facility needs estimates of costs for maintenance turnarounds that are both accurate and precise. Accurate estimates are needed for long-term planning of cash-flow for the facility. Accurate and precise estimates are also needed during execution of the turnaround, to help manage and control expenditure.

Defining Accuracy & Precision

Before we discuss turnaround cost estimates in any detail, let us first define what we mean, in the context of this article, by accuracy and precision.

By accuracy, we are talking about whether, on average, the cost estimates match the actual cost outcome. By precision, we are talking about whether the accuracy range claimed for the estimates is validated by the range of actual outcomes in a sample set.

To illustrate the point, let us take a hypothetical sample of 100 turnarounds, each of which has an estimate of €50m for the P50 point value (50/50 chance of overrun or underrun). Let’s also say that the accuracy range assigned to each estimate is ±10% at the 80% confidence level.

If we look at Figure 1, this shows some hypothetical “actual” cost outcomes for our example set.

• Column 1: The estimates were both accurate and precise. We can say this because the average outcome of the set equals €50m and the spread of cost outcomes has 80 of the 100 turnarounds falling within 10% of the average.
• Column 2: The estimates were accurate but were overly precise. We can say this because the average outcome of the set equals €50m but the spread of cost outcomes has 80 of the 100 turnarounds falling in a spread that is broader than 10% of the average, ie the estimates turned out not to be ±10%.
• Column 3: The estimates were not accurate, but they were precise. We can say this because the average outcome of the set is higher than €50m but the spread of cost outcomes still has 80 of the 100 turnarounds falling within 10% of the average outcome.
• Column 4: The estimates were not accurate, and they were overly precise. We can say this because the average outcome of the set is higher than €50m and the spread of outcomes has 80 of the 100 turnarounds falling in a spread that is broader than 10% of the average.