oilfield review winter 2000-2001 - making decisions in the...
TRANSCRIPT
2 Oilfield Review
For help in preparation of this article, thanks to Joe Fay,Austin, Texas, USA; Kent Burkholder and Alexander Lythell,London, England; Paige McCown, Houston, Texas; PatParry, Centrica, Slough, Berkshire, England; KennethRicard, Aker Maritime, Houston, Texas; and LaurenceWickens, AEA Technology, Didcot, Oxfordshire, England.Decision Tree and Peep are marks of Schlumberger. DPS-2000 is a mark of Aker. Excel is a mark of MicrosoftCorporation.
Decisions in the oil and gas industry determinethe direction and course of billions of dollarsevery year. The complexity of a decision canrange from simple and Shakespearean—to drillor not to drill—to elaborate. Some of the moremonumental decisions determine the maximumbid for a lease, the best development process fora given asset, the drilling priority of a company’sexploration opportunities, the timing of increas-ing facility capacity, or whether to sign a long- orshort-term supply contract.
While simpler problems may be analyzedwith a few quick calculations, reaching morecomplicated decisions can take a companymonths or years of preparation. For example, oneof the dilemmas challenging E&P companiestoday is how to develop deepwater prospects.Sometimes subsea development is best; some-times a tethered floating structure is the solution.Typically, oil companies spend 12 to 18 months inthe decision cycle—gathering information, ana-lyzing data and modeling risk and uncertainty—before selecting a production system. Streamliningthis process may increase profit by shortening thetime to first production.
Ellen CoopersmithDecision FrameworksHouston, Texas, USA
Graham DeanCentricaSlough, Berkshire, England
Jason McVeanCalgary, Alberta, Canada
Erling StorauneAker Maritime, Inc.Houston, Texas
D
RIL
L
WIL
DC
AT
RU
NS
EIS
MIC
DR
YH
OLE D
RIL
L
DR
OP
AC
REA
GE
DR
ILL
LAR
GE
FIE
LD
BU
YA
C
REA
GE
$3
00
,000
$50,0
00
$50
,00
0
$5
0,0
00
The difference between a good decision and a bad one can be the difference
between success and disaster, profit and loss, or even life and death. Decision-
analysis software can help decision-makers identify factors that influence the
decision at hand and choose a path to desirable outcomes.
Making Decisions in theOil and Gas Industry
1. Newendorp PD: Decision Analysis for PetroleumExploration. Aurora, Colorado, USA: Planning Press,1996.
2. Bailey W, Couët B, Lamb F, Simpson G and Rose P:“Taking a Calculated Risk,” Oilfield Review 12, no. 3(Autumn 2000): 20-35.
3. Net present value is one possible value measure, butmany others can be used, including rate of return andprofit-to-investment ratio.
4. Newendorp, reference 1, chapter 4.
Winter 2000/2001 3
Several methods are available to help decision-makers evaluate uncertainty, reduce risk andchoose workable solutions.1 These methodsinclude net present value (NPV) calculations, dis-counted cash-flow analysis, Monte Carlo simula-tion, portfolio theory, decision-tree analysis andpreference theory—all of which were reviewedin a recent Oilfield Review article.2 Elementarysituations can be solved with basic expected-value computations, but more involved casesrequire a decision-analysis process that com-bines information from multiple disciplines,allows for uncertainty and evaluates the impactof different decisions. This article focuses ondecision-tree analysis and how it works, asdemonstrated through two case studies.
Decision-Tree AnalysisDecision-tree analysis is one way to frame andsolve complex situations that require a decision.The key to obtaining a useful solution is to clearlydefine the problem at the start and determinewhat decisions need to be made. The problem-definition stage includes identifying all knowninformation and listing any factors that may influence the final outcome. To expedite the process, decisions that can be deferred are post-poned so that future information can aid the deci-sion process.
Capturing the essence of a problem by deter-mining which are the influential factors helpsdecision-makers concentrate on only thoseissues that play a major role in the outcome.Such sensitivity analysis prioritizes the impor-tance of factors to be considered in a decision.For example, one decision may depend on six fac-tors: oil price, oil volume, gas price, gas volume,capital expenditures and operating costs—but the relative importance of these is unknown.For given uncertainties, or a range of possiblevalues, for each factor, sensitivity analysis cal-culates the net present values (sometimes writ-ten as after-tax cash) represented by those
uncertainties and ranks each factor (left).3 Thefactors that most affect project outcome are theones with the highest NPV range. The shape of thegraph, with large values on the top and small val-ues on the bottom, gives it the name “tornadoplot.” In this example, the two most importantfactors are oil price and oil volume. Operating-cost uncertainty does not significantly affect theoutcome, and so can be treated as a certaintywithout markedly affecting the results.
Once the problem is framed, decision treeshelp find a route to an advantageous solution.Decision trees are diagrams that portray the flowof a decision-making process as a sequence ofevents and possible outcomes (below). Eventsare represented as points, or nodes, and out-comes are depicted as branches issuing fromeach node. Nodes are either decision nodes—atwhich the decision-maker determines whatbranch to take—or uncertainty nodes, wherechance rules the outcome.4 Associated with eachbranch is the expected monetary value of the out-come. Additionally, branches emanating fromuncertainty nodes are weighted by the probabilityof that outcome occurring. In common notation,decision nodes plot as squares and uncertaintynodes as circles.
Sensitivity Analysis
Influence, %
Oil price
Oil volume
Capital expenditures
Gas volume
Gas price
Operating costs
51
31
09
05
03
01
30,000
31,538
36,068
42,083
41,776
42,862
43,643
Factors45,010 60,000
59,937
57,873
54,004
50,956
50,157
46,181
Net present value dollars, thousands
> Tornado plot showing factors that most influence a decision. Of the six factors selected for analysis,oil price and oil volume have the highest range in net present value (NPV), making the outcome mostsensitive to those factors.
Dry hole_ $50,000
2 Bcf+ $100,000
5 Bcf+ $250,000
$0
+ $100,000
Drill
Don’t drill
0.7
0.1
0.2A
B
> Simple decision tree showing one decisionnode (square) and possible outcomes. Outcomesare labeled with their expected value multipliedby the probability of that outcome occurring. The path with the highest expected value is high-lighted in yellow. (Adapted from Newendorp, reference 1.)
In this simple example, the decision nodemarks where the decision-maker elects to drill ornot to drill. The expected value associated with adecision not to drill is $0; that is, no money isspent or gained. The value expected from thedecision to drill depends on what is encountereddownhole: there is a 10% chance of 5 Bcf of gas,a 20% chance of 2 Bcf, and a 70% chance of adry hole. The expected reservoir size is in fact acontinuous distribution rather than three-valued,but for the purpose of this example, three sizesare examined (above). Ideally, branches of theuncertainty node try to capture the most impor-tant aspects of this continuous distribution.
The expected value of an uncertainty node isthe sum of all probability-weighted expected val-ues of the outcomes stemming from that node. Inthis way, working back from the end, or right-hand side of the tree, expected values can be cal-culated for every outcome. Once all the expectedvalues have been calculated, the optimal deci-sion route—the one along maximum expectedvalue—can be taken.
The same method works for more compli-cated decisions (next page). In this example, thedecision to buy acreage or not depends on under-standing the possible outcomes of a succession
of decisions, including to run a seismic survey ornot, to drill or not, and to drill a second well ornot. The possible final outcomes—large field,marginal field and dry hole—are the sameregardless of the decision route. However, theyhave different probabilities of occurrence at dif-ferent stages of the decision tree because as thetree grows, more information becomes available.For this decision tree, the solution that deliversthe highest expected monetary value traces thefollowing branches: Buy acreage, run seismicsurvey, seismic survey confirms lead, drill—andif the first hole is dry, drill a second wildcat.
Assigning probabilities to the tree branchesrequires technical expertise and, in this case,relies on prior knowledge of the region. The like-lihood and value of the various outcomes alsocan be based on the result of more detailedMonte Carlo simulations. For example, the cutoffvalue for what constitutes a large field could bethe high side of a probability distribution that isthe result of Monte Carlo simulation of the reser-voir volume parameter.
Depending on the type of decision to bemade, specialists from many oilfield disciplinesmay be called upon to supply information for
decision-tree analysis. In addition to the unknownreservoir size and content, outcomes that need tobe predicted include the following:
• price of oil and gas• quality and reliability of seismic imaging or
logging data• cost and risk versus value of additional
information• likelihood of drillpipe or logging tools get-
ting stuck, and other nonproductive rig time• reservoir compartmentalization, or number
of wells• reservoir fluid properties and performance• completion complexity• cost of transport to market• improvement gain from stimulation,
workover or enhanced-recovery methods.Less obvious to many oilfield professionals,
perhaps, but also important to estimate in casesthat lend themselves to decision-tree analyses,are eventualities such as government legislationand stability, company mergers, court cases andhealth, safety and environment (HSE) issues.
Numerous software products are availablethat facilitate decision-tree analysis for oil andgas E&P and other industries. These includePrecision Tree by Palisade, Decision Program-ming Language (DPL) by ADA (Applied DecisionAnalysis) and the Decision Tree package devel-oped by Merak, a Schlumberger company. Thesesystems link to calculation engines that computenet present values for each branch of the tree.Broadly speaking, decision-tree software pack-ages link to Excel as the calculation engine. Onlythe Merak Decision Tree software also linksdirectly to the Peep economic analysis program,which is a standard asset-management packagein the petroleum industry.
Decision trees can be helpful for analyzingmany types of oilfield decisions. Examplesinclude deciding whether to replace wireline logswith logs acquired while drilling, evaluatingwaterflood programs, optimizing workovers, andchoosing the best offshore platform topsidesconfiguration.5 The next section describes howdecision trees can help evaluate a deepwaterproduction system.
4 Oilfield Review
Maximum possiblereserves
Mean reserves
Minimum possiblereserves (dry hole )
> Continuous distribution of expected reservoir size. Although theexpected value of the reservoir size can fall anywhere in the contin-uous distribution, the most likely values should be selected for thedecision-tree branches. (Adapted from Newendorp, reference 1.)
Winter 2000/2001 5
Choosing a Production SystemAker Maritime, Inc., a maker of offshore platformsand spars, was approached by an operator prepar-ing to select a deepwater production system for adevelopment offshore West Africa.6 The client hadto decide whether to act early and buy a produc-tion system that could be adapted in case thereservoir turned out to be larger than expected, orwait for more information and optimize the size ofthe system. An early decision could mean quickerproduction of first oil. And an adaptive system hasthe flexibility to allow for future additions of fluid-processing modules or wells. However, such adecision would be based on minimal information.The alternative was to drill more wells, acquire
more information and buy a production systemoptimized for the reservoir size, but at additionalexpense and production delay.
Aker Maritime worked with DecisionFrameworks, a decision analysis and facilitationconsulting firm, to structure the decision andmodel development alternatives. The DecisionFrameworks approach relies on technical andcommercial petroleum-industry expertisepaired with Merak software, specifically theDecision Tree product and Peep economicdatabase application.
The first steps in the decision analysis wereto structure the problem, understand issues associated with the deepwater discovery andreview alternative development solutions.
> More complicated decision tree for buying acreage. In this example, the decision depends on a succession of decisions (highlighted in yellow) including running a seismic survey and drilling one or two wells. (Adapted from Newendorp, reference 1.)
5. Beck GF: “Examination of MWD (Measuring WhileDrilling) Wireline Replacement by Decision AnalysisMethods: Two Case Studies,” Transactions of theSPWLA 37th Annual Logging Symposium, New Orleans,Louisiana, USA, June 16-19, 1996, paper U.Martinsen R, Kjelstadli RM, Ross C and Rostad H: “TheValhall Waterflood Evaluation: A Decision Analysis CaseStudy,” paper SPE 38926, presented at the SPE TechnicalConference and Exhibition, San Antonio, Texas, USA,October 5-8, 1997.Macary S and El-Haddad A: “Decision Trees OptimizeWorkover Program,” Oil & Gas Journal 96, no. 51(December 21, 1998): 93-97, 100.MacDonald JJ and Smith RS: “Decision Trees ClarifyNovel Technology Applications,” Oil & Gas Journal 95,no. 8 (February 24, 1997): 69-71, 74-76.
6. A spar, sometimes called a dry-tree unit, is a floating vertical cylinder that is moored to the seafloor. Sparsallow production from deepwater fields to “dry” surfacefacilities as opposed to subsea facilities.
B
C
A
IH
GF
J
DE
Large field+$36 MM
Marginal field+$11 MM
Dry hole
Drill second
wildcat
Large field+$35 MM
Marginal field+$10 MM
Drill
Don’t buy acreage$0
Buy acreageRun seismicsurvey
Seismic survey
confirms lead
Large field+$33 MM
Marginal field+$8 MM
Dry hole,drop acreage
–$7 MMDrop acreage
–$6 MM
Drill secondwildcat
Large field+$34 MM
Marginal field+$9 MM
Dry hole
Drill
0.90
0.05 0.05
0.075
0.075
0.85
0.50
0.50
0.15
0.15
0.700.20
0.200.60
Dry hole,drop acreage
–$5 MM
Seismic shows nostructure, drop acreage
–$5 MM
Drop acreage–$4 MM
Drop acreage–$5 MM
Decision Frameworks worked with Aker and theiroil company client to define parameters of thediscovery, such as reservoir size, productionrates, number of wells and drilling schedule.Then, high-level decision trees were constructedfor the four development concepts under consid-eration. Two concepts were adaptive structures,the Aker adaptive spar and DPS-2000 (above andnext page, top). The other two were designs that
could be optimized to suit the reservoir size: afloating production, storage and offloading(FPSO) system and an optimized spar. All fourconcepts allowed oil storage.
The Decision Tree analysis for the adaptivedesigns called for selection of surface structureand topsides based on information from only twowells. This was followed by drilling of two wells,installation of the structure, development drilling
and production, then installation of additional pro-duction modules or wells as necessary (next page,bottom). The case of optimized designs starts withinformation from four wells before selecting andinstalling the production system, followed bydrilling of additional development wells and finallysome limited adjustment of the well count,depending on what production information indi-cated the size of the actual reservoir to be.
6 Oilfield Review
> Aker Maritime adaptive spar.
Winter 2000/2001 7
> Aker Maritime DPS-2000.
Limitedadjustment
of well count
Add productionmodules and
adjust well count
Development drillingand productionTwo penetrations
Four penetrationsDevelopment drilling
and production
Adaptive Designs
Optimized Designs
Decision Tree Structure
Selectstructure
and topsideequipment
Predrilltwo wells
Indicatedreservoir size
(capacity/wells)
Installstructureand drill
Installstructureand drill
Furtherindication ofreservoir size
Start withmore
information
Indicatedreservoir size
(capacity/wells)
Selectstructure
and topsideequipment
Furtherindication ofreservoir size
> Decision Tree structure for adaptive design compared to optimized design. The adaptive design (top) starts with less information and drills fewer wells. The optimized design (bottom) uses information from four wells to size thedevelopment concept.
The key uncertainty was the reservoir size,which governs the facility capacity and the num-ber of wells required to develop the reserves. Theresults of the Decision Tree analysis are the eco-nomic impacts of multiple scenarios that occur ifthe reservoir is:
• believed to be large and actually is large,medium or small;
• believed to be medium and actually islarge, medium or small;
• believed to be small and actually is large,medium or small.
A sample decision tree demonstrates the netpresent values calculated for one of the four
development concepts: the DPS-2000 adaptivesystem (above). The total NPV for this concept is$412 million. Comparing this figure to thoseobtained for the other three concepts shows theDPS-2000 to have the greatest NPV (below left).
The timing of the steps in the developmentplays a key role in investment payback. A largepart of the value in selecting an adaptive systemis in the reduced time to first oil. The DecisionTree software followed a schedule from January2001 to June 2005 that included front-end engi-neering and design (FEED), construction, deliveryand commissioning (below right). Both adaptiveconcepts could deliver first oil in 2003, compared
with the 2005 deliveries possible with the optimized systems. However, the added value ofthe adaptive systems was accompanied by added risk.
The Decision Tree software helped demon-strate the added value achievable with the earlyadaptive production systems and allowed AkerMaritime to present a full range of decisionoptions to the oil company client. It also under-scored the fact that uncertainty often exists evenafter more information is gathered. Recognizingthis during the selection of development con-cepts is important, and can add value.
8 Oilfield Review
Adaptive Concepts Optimized Concepts
Value Captured
DPS-2000
Adaptive spar
$412 MM
$350 MM
Optimized spar
FPSO
$313 MM
$182 MM
> Comparison of final NPV calculations for the four develop-ment concepts under consideration. The adaptive conceptsoffer up to $230 million higher NPV than the optimized concepts.
Jan.2001
Jun.2001
Jan.2002
Jun.2002
Jan.2003
Jun.2003
Jan.2004
Jun.2004
Jan.2005
Jun.2005
Jan.2006
Decision Tree Schedule Inputs
Aker DPS-2000
Adaptive spar
FPSO
Optimized spar
First oilDec. 2003
First oilAug. 2003
First oilFeb. 2005
First oilApr. 2005
Includes front-end engineering and design (FEED), construction, delivery and commissioning
> Schedule of Decision Tree events. The adaptive concepts start first and producefirst, while production from the optimized projects lags by about 18 months.
0.48 large
0.36 medium
0.16 small
882
NPV $MM
359
9
Actual reservoirprobability, size
0.23 large
0.46 medium
0.31 small
882
361
12
0.23 large
0.36 medium
0.41 small
882
361
14
Large reservoirindicated
Medium reservoirindicated
Small reservoirindicated
Predrill2 wells
Predrill2 wells
Predrill2 wells
InstallDPS-2000
InstallDPS-2000
InstallDPS-2000
p =0.2
8p = 0.39
p = 0.33
Two penetrations
DPS-2000
$412 MM
Decision Tree Results for Aker DPS-2000
$554 MM
$373 MM
$339 MM
26 wells, $100 M modification
14 wells, 1 dry hole
6 wells, 2 dry holes
14 wells
6 wells, 1 dry hole
14 wells
6 wells
26 wells, $100 M modification
26 wells, $100 M modification
> Decision Tree output showing NPV calculated for the DPS-2000 adaptive deepwater system.
Winter 2000/2001 9
Making a CaseThe decision-tree methodology also can beapplied to other types of E&P problems. As part ofCentrica’s strategy to acquire additional UK conti-nental shelf assets, the company can—when italready has contracts to buy the gas from theassets—often be required to consider potentiallyconflicting buyer-seller interests. In one example,Centrica needed to consider the impact of a pastdispute on the future value of an asset under con-sideration. The dispute related to the previousfailure of the sellers to meet contractual obliga-tions to provide gas and to Centrica’s applicationof the contractual remedies. The sellers objectedto this action and litigated to limit it. Centrica hadto consider the possible outcomes from a litigatedversus negotiated settlement on the future valueof the asset—the Hewett field (above).
Several conditions complicated the decisionprocess. Acquiring additional equity in the fieldor taking on operatorship could increase reservesand value, and allow Centrica to provide moregas, but the field was old and nearing expensiveabandonment. However, there also was potentialfor workover or developing neighboring fields.
So many elements were involved in the decisionthat the problem appeared quite difficult to solve.
Centrica consulted with AEA Technology plc tohelp them frame the problem. The resulting deci-sion tree was large, requiring evaluation of 7000alternatives with hours of computer runs per out-come that totalled a man-year of effort. An auto-mated solution was needed to generate and inputnumbers for the decision tree. Centrica analystsused the Merak Decision Tree product, and, byparing down some of the constraints, were ableto achieve a solution with 500 outcomes and com-puter run times of 7 minutes (above right).
The benefits of a Decision Tree solution weretwofold: first, the decision-analysis process pro-vided clear insight into the problem. In spite of thecomplexity of the situation, the Decision Treeresults clarified what was driving the decision aswell as the direction to be taken. For the first time,everyone involved was in agreement with therationale for the set of decisions. Second, theMerak tools made it easy to solve the problem andcompleted the calculations and analysis quickly.
Simplifying Decision-MakingFor large organizations like those in thepetroleum industry, it is still people, not pro-cesses, that make complex, expensive decisions.The decision-analysis technique is usually cus-tomized from one organization to the next, butthe most successful system is one of framing theproblem, understanding uncertainties, develop-ing stronger, often hybrid solutions, and balanc-ing risk against expected value.
As the petroleum E&P industry continues topursue prospects in more remote and potentiallysensitive regions, decision-making tools thatincorporate information from all sources ofexpertise will make important contributions toproject success.
Although decisions are ultimately made bypeople, computer and software solutions makethe job easier. Decision-analysis products canhelp identify how sensitive a decision is to all thefactors involved, determine the value of movingahead or gathering data, point decision-makersin the most valuable direction, and produce moreconsistent decisions.
The benefits of a consistent decision-analysisprocess are felt by decision-makers in all parts ofthe company, allowing technical staff and plan-ning organizations to increase the efficiency andvalue of their work. —LS
> Schematic decision tree created to help Centrica analysts reach a decisionon the Hewett field case. The tree uses a compact notation whereby a nodenext to the outcomes of a previous node means that node is repeated foreach input branch. In this way, the first decision node, “Acquire additionalequity,” applies to all three outcomes of the previous node, “Calculatedchange in gas initially in place.” Similarly, the decision node to settle bynegotiation applies to all Yes or No outcomes of the previous decision, andso on. This notation conveys the same information as a whole tree butkeeps the tree compact and manageable.
Calculatedchange ingas initiallyin place
Low
Middle
High
Acquireadditionalequity
Yes
No
Yes
No
Low
Middle
High
Low
Middle
High
Identityof expert
A
Settlelitigationby negotiation
Negotiatedoutcome
Expertdecision
BU K
N O R WAY
HewettField
N O R T H S E A
> Hewett field, UK North Sea, where low yields prevented Centrica and partners from meeting contracts to provide gas. Decision Tree analysishelped Centrica decide whether to proceed with the ensuing settlement.