Enterprise Oil used a probabilistic well-cost estimating system to plan offshore developments in Brazil's Campos Basin.

Engineers traditionally have used a deterministic approach when estimating times and costs for well construction operations. Peak developed P1 probabilistic software to provide the more realistic time and cost models the industry demands.
The model input is split into individual operation phases, such as "drill 121/4-in. hole." These phases are subdivided into activities that are assigned ranges of times or rates, such as "drill 121/4-in. hole from 4,167 ft to 5,807 ft (1,270 m to 1,770 m) at a rate of penetration of (ROP) of 98 ft/hour to 180 ft/hour (30 m/hour to 55 m/hour)."
The distribution of minimum, most likely and maximum rates reflects the normal variation in drill rates through a section. Potential problems can be assigned to each planned activity. For example, a 50% probability of "hole stability issues" can be assigned. Remedial actions for this, such as raising the mud weight, require an extra 6 to 24 hours. However, the remedial actions may not work. Figure 1 shows there is a 20% chance of the original problem escalating into "stuck pipe" with more serious consequences.
Operational costs are entered as a simple list. Day rates, lump sums and unit costs can be entered into the system, which automatically produces a cost model based on the calculations.
When the activities and potential problems have been entered, the system carries out a Monte Carlo simulation of the well using the @Risk software package. The system runs thousands of iterations (default value = 5,000) of the model to build a picture of possible outcomes and related probabilities.
The example in Figure 2 shows that the P0 well cost is US $14.6 million. Values close to that end of the graph, P1 or P5, can be taken as the stretch or limit target. The P100 value of $30 million is the maximum cost exposure for the company. The P50 value of $18 million is a more likely well cost.
Outputs include charts of the time and cost estimates for each phase. Any three probabilities can be chosen to illustrate the spread of results.
Figure 3 shows that the greatest opportunity for reducing time is in "Drill the 81/2-in. hole," with 6 days between the P10 and P50 values. The area of greatest risk is "Set 7-in. liner," with a spread of 23 days between the P50 and P90 values. This output can be used to concentrate engineering effort on the areas of high risk and high opportunity.
The system's developers emphasize it does not identify risk or assign probabilities or consequences; the well design team's skill and experience is required to build a realistic model.
P1 case study
The Bijupirá and Salema fields are in Brazil's Campos Basin, in 1,805-ft to 2,789-ft (550-m to 850-m) waters. Enterprise Oil acquired operatorship of both fields in April 2000.
The Enterprise project team had to engineer the most economic field development plan and generate an accurate project cost estimate and operations schedule. This task required input from subsurface, well engineering and facilities engineering teams. The well engineering team's role was to engineer the most cost-effective plan for drilling and completing wells to access nine horizontal production well targets and six injection well targets.
Enterprise's well engineering team demonstrated at an early stage that reducing the number of top-hole locations required to access 15 targets would generate significant savings by reducing the rig moves required. The team realized an additional benefit in the implementation of a batch approach to operations, reducing the number of blowout preventer (BOP) trips and exploiting the learning curve effect. The team generated a plan to drill and complete the 15 wells from only two drilling clusters, one in each field.
This drilling and completions plan carried several significant risks. For example, several long-reach and 3-D wells would be required with 121/4-in. hole sections through a notoriously problematic conglomerate formation. The 2,625-ft (800-m), 81/2-in. hole sections through the reservoir would require sand control measures, and the risk of differential sticking would be high due to depletion from pilot production. Operations scheduling would be critical due to concurrent manifold installation operations. Given the inherent risks in the drilling and completions plan and the importance of scheduling to the overall project, the team decided a true risked cost and time estimate was required. Enterprise approached Peak for assistance.
The first step in building the models was to collect offset data and extract significant values for all key parameters, such as ROP, expected tripping speeds and BOP running times. To assess the risk and opportunity involved in the project accurately, the team facilitated a series of meetings. Enterprise's standard well planning process includes risk assessment in a session involving the main contractors as well as the planning team. To build a P1 model, however, the emphasis must be shifted slightly. In a standard risk-assessment meeting, the emphasis is on identifying the potential downfalls and assigning responsibility for mitigation. To build a model, one also must consider the potential upside. Could this phase be drilled with fewer bit runs? Will new technology improve ROP compared to offset average? The likelihood of events must be quantified, based on offset data and expert opinion, and the possible remedial action considered. Meetings involving the Enterprise team and the drilling, directional drilling, drilling fluids, gravel packing and completions contractors took place.
Seven models were built, covering four well types. With each model, other wells of the same type were modeled simply by changing the directional trajectory.
Modeling campaign costs
With traditional methods, the risk element of the cost of individual wells is not considered, and field development budgets are produced by adding the deterministic budget estimates. This does not produce a realistic estimate for senior management. Therefore, Peak built on its experience of modeling individual wells to develop C1, a logical extension from considering probabilistic distributions for individual wells to considering campaigns of wells. C1 uses the probabilistic outputs from the P1 models as input to a model of the complete development, including learning curves and correlation effects.
C1 case history
Having created the building blocks of the Bijupirá and Salema development, the team began to focus on modeling the whole campaign. In order to maximize operational efficiency, Enterprise decided on a batch setting approach. Fortunately, P1 not only produces probabilistic distributions of duration and cost for a well, but also for each of the constituent phases. The campaign model was built using the phases output from P1 as building blocks.
The base case values range from a P0 of 460 days to a P100 of 1,120 days with a P50 of 601 days (Figure 4). Applying learning curves to the model reduces the P50 days to 574. Similar curves were produced for the cost estimates.
This innovative methodology allows true probabilistic forecasting of the development duration and cost. The Bijupirá and Salema output provided accurate and realistic estimates for Enterprise management to use in project approval and planning decisions.

Comments