Integrating Learning Curves in Probabilistic Well-Construction Estimates

Abstract

Summary For multiple-well drilling and completion campaigns, cost and schedule performance tend to improve over time. This trend in improvement is commonly referred to as a "learning curve." When a learning curve is anticipated, the campaign cost and schedule estimates may be reduced dramatically relative to an assumption of constant performance. That is, ignoring the learning curve will lead to overly pessimistic estimates. While learning curves can be observed in campaigns of various lengths and complexity, they are typically most important in large campaigns where the majority of wells are drilled after a significant portion of the learning has occurred. Conversely, they may not be appropriate in short campaigns where there is a limited time to implement learnings, or in campaigns with highly idiosyncratic wells where learning does not necessarily translate across projects. Many operators consider the use of learning curves a best practice and provide procedures for estimation and implementation in their cost-estimating guidelines. In cases where comparison projects exist, estimating a learning curve for a prospective project can be achieved with some certainty. This form of deterministic learning is a well-established topic in the drilling-engineering literature and in practice. However, in cases where the sample of comparison projects is small, there may be significant uncertainty in the rate and magnitude of learning over time, and some form of probabilistic learning is more appropriate. This form of learning is not well established in the literature or in practice. This paper investigates methods for systematic integration of learning curves in probabilistic estimates. Brief reviews of probabilistic estimating methods and learning curves are provided. A general method and specific procedures for integrating learning curves in probabilistic estimates are then provided. For each method, the key assumptions are itemized and discussed and a demonstration is provided. While no single procedure will fit every situation, it is concluded that the general method is straightforward, transparent, and can be implemented using off-the-shelf spreadsheet software. The proposed procedures generate results that provide engineers and decision makers with a refined representation of uncertainty and can improve capital-investment valuation and decision making.