30-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s oil and gas sector
WEEK 2 OF 6 · GOVERNANCE, PEOPLE AND THE SUBSURFACE · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Part One · Leadership and the Operating FoundationSESSION 6
Governing AI and Earning Autonomy
Why This Matters
In Session 5 you separated information from control. Today you decide how much permission any system behind your decision may hold, and how you would take it back. The chapter begins with an uncomfortable fact: an application may be accurate enough for one task and unsuitable for another, useful on one asset and unreliable on a different configuration, dependable in stable operation and wrong when conditions change. Governance begins by defining the intended use precisely enough to recognise those boundaries. By the end you will have registered the actual system or proposal behind your decision, set how it will be judged, and written down what would withdraw its permission.
25-Minute Module
0:00 – 4:00
Govern the use, not the product
- A gauge can read well on one service and badly on another, so you approve it for a duty, not in general. An AI application is the same: its fitness belongs to a defined use, asset and set of conditions.
- THE DOCTRINE
Governance begins by defining the intended use precisely enough to recognise its boundaries. Permission belongs to the approved use, not to the model's general reputation.
- The NIST AI Risk Management Framework organises its core around Govern, Map, Measure and Manage, with governance cross-cutting. It is voluntary rather than a certification of an operator's systems; it offers a useful vocabulary for purpose, context, performance and risk (NIST, Artificial Intelligence Risk Management Framework AI RMF 1.0, 2023, section 5).
When someone says the tool is approved, ask: approved for what, on which asset, by whom?
4:00 – 9:00
Register the actual system
- Do not register only a product name. The same product can support uses with different risk: retrieving a supplier certificate is not assessing a well intervention, and an agent processing a routine document request is not an agent initiating a payment.
Register field What it must answer Business and technical owners Who answers for the outcome and for the system, by role Intended users Who may act on the output Data sources Which records feed it, under your Session 4 data contract Approved purpose The specific use, not the capability list Version and permissions Which release is live and what it may touch Consequence of error What happens when it is wrong Failure detection and response How failure is noticed and who responds Review date When the approval is next examined - The register should follow the use, not merely the software. Where the use changes materially, revisit the approval.
A register with one line per vendor is an asset list. A register with one line per use is governance.
9:00 – 14:00
Evaluate decisions, and bound the sources
- A high average accuracy can be unhelpful when the events that matter are rare and costly. A model right on every quiet day and silent on the bad one has a fine average and no operating value.
- For early warning, examine missed events, false alarms, warning time and the action available. An alert just before failure may be technically correct and operationally useless; an earlier one is still useless if the intervention cannot be prepared.
- THE DOCTRINE
The signal can direct attention; competent personnel establish whether action is justified. Record that relationship rather than attributing every successful intervention to the algorithm.
- Compare current practice. If an existing engineering threshold gives equally useful warning at lower cost, the complex model needs further justification.
- Generative AI needs a source boundary. A language model can write a fluent explanation your approved records do not support. Give it authorised documents, require source identification, show document status, and have it say so when evidence is absent. Users check the cited passage, its version and its relevance to the present equipment. On legal, engineering or financial matters, a polished draft is still a draft until the relevant authority reviews it.
Ask for the missed events before you ask for the accuracy figure.
14:00 – 21:00
Permission is earned separately, and watched after
- The Autonomy Ladder from Session 1 describes increasing permission, not inevitable progress toward unrestricted execution. A useful system may remain advisory indefinitely. An agent's identity should hold only the permissions its task needs, and approval checkpoints should be enforced by the surrounding system, not merely described in a prompt.
- Administrative execution needs limits: a routine classification error can misroute an invoice or expose a record without touching the plant. Plant interventions need separate engineering and operating controls and formal change arrangements where applicable, and the business must be able to refuse execution.
- AT THE ASSET · THE WORKED CASE
- Asset. The book's teaching case: an application recommends maintenance inspections and performs well in a controlled trial. The sponsor proposes purchase-order access so it can obtain parts itself. The model is unchanged, but this is a new use.
IN THE OPERATING REVIEW · THE PARALLEL- Review. Procurement and finance join the reliability team to examine spend authority, item identity, supplier eligibility, duplicate orders, maximum commitment, exception handling and cancellation.
- An intermediate design lets the application draft a requisition with linked evidence for an authorised person to review and submit. Technical performance has earned a right to be considered, not financial authority.
- Approval is a point in time. Monitor performance, data quality, user behaviour, overrides, incidents and material changes, and investigate a rise in rejections before assuming users are resisting innovation.
- THE DOCTRINE
Retain the ability to suspend, reduce permissions and restore the previous method. A system that cannot be stopped safely has been granted more authority than the organisation can govern.
If wider permission is later justified, grant a bounded scope and watch the actual actions, not the demonstration.
21:00 – 25:00
The negative case, and your wrap
- Every review needs a negative case: what would justify withdrawal? A sponsor should answer that as clearly as the question about expansion, naming who may pause access and on what evidence.
- The objective is dependable assistance within a known boundary: you can explain what the system does, why it is allowed to, and who remains responsible when it is wrong.
- Before you close: write the register entry for the system or proposal behind your Session 1 decision, its evaluation, the rung it may hold and the trigger that would take it down a rung.
- That is Section 6 of your 90-Day Intelligent Operating Plan. Write your assignment now.
I would rather approve a modest permission I can withdraw than a generous one I cannot.
Key Terms
System register
The record of each approved use of an application, with its owners, users, sources, purpose, version, permissions, consequence of error and review date.
Source boundary
The limit that keeps a generative AI assistant to authorised documents, with sources identified and absent evidence admitted.
Approval checkpoint
A point where an authorised person must approve before a workflow proceeds, enforced by the surrounding system, not by a prompt.
Advisory application
A system that provides information or a recommendation for competent human review within its approved scope.
Self-Check — Before You Submit
1. A fictional operator's inspection model did well in a controlled trial. The sponsor now wants it to raise purchase orders itself, since the model is unchanged. What is the right response?
- A. Approve it, because trial performance is proven
- B. Treat it as a new use, reviewed with procurement and finance, and consider a reviewed draft requisition first
- C. Refuse the application any procurement role permanently
- D. Let the reliability team decide alone, since it owns the model
B. Treat it as a new use, reviewed with procurement and finance, and consider a reviewed draft requisition first — Permission belongs to the use, and purchase-order access is a new use with financial consequences. Approving on trial performance is the tempting error: technical performance earns consideration, not financial authority.
2. Which evidence matters most when evaluating an early-warning application for a rare, costly failure?
- A. The highest average accuracy across the test set
- B. The number of staff trained on the dashboard
- C. The vendor's list of other clients
- D. Missed events, false alarms, warning time and the action available
D. Missed events, false alarms, warning time and the action available — When the events that matter are rare, a high average can hide the misses. A correct alert too late to act on is operationally useless, which is why warning time and available action are tested.
3. Six months after approval, users are rejecting far more of a system's recommendations. What should the owner do first?
- A. Investigate whether the cause is the model, a changed asset condition or poor usability
- B. Remind users the system is approved and require compliance
- C. Retrain the model immediately on the latest data
- D. Remove the users' ability to override
A. Investigate whether the cause is the model, a changed asset condition or poor usability — Each possible cause has a different remedy. Treating rejections as resistance to innovation is the tempting mistake, and removing overrides would destroy the evidence monitoring depends on.
Today’s Assignment
Write Section 6 of your plan: the system register entry for your decision, how it will be evaluated, and the permission it has earned.
- Name the system or proposal behind your Session 1 decision and its approved purpose in one sentence. If your company has no AI in use yet, that is normal: register the proposal in front of you and the method it would replace.
- Complete its register entry, with both owners named by role and the consequence of error stated.
- Set the evaluation on decisions, not averages, and name the current practice or simpler method it must beat. If it retrieves or drafts text, state its source boundary.
- State its rung on the Autonomy Ladder, what it may not do, and where the approval checkpoint is enforced by the system.
- Write the negative case: who may suspend it, reduce its permissions and restore the previous method, and on what evidence.
- The honest part: where your company approves a product name rather than a use, or holds a permission it could not quickly withdraw. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 6 — The system register entry, evaluation and the permission it has earned.
How this is marked — the rubric
| Registers a specific use, not a product, with business and technical owners named by role and the consequence of error stated | 25 |
| Sets evaluation measures that fit the decision and names the current practice or simpler method it must beat | 25 |
| States the permitted rung, explicit exclusions and a system-enforced approval checkpoint | 25 |
| Names the withdrawal trigger, who may suspend, and how the previous method is restored | 25 |
| Total · pass mark 70 | 100 |
Submitting session 6
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 6, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
System registrationDecision-based evaluationSource boundariesBounded permissionPost-approval monitoring
Part One · Leadership and the Operating FoundationSESSION 7
Building the Workforce That Can Use Intelligence
Why This Matters
Session 6 decided what a system may do. Today is about the people who must use it, check it and, when the evidence says so, refuse it. The chapter's opening point is plain: a new application changes work only when people understand how to use it in the decisions they already make. A field supervisor, a finance reviewer and a technical authority have different learning needs, and one generic AI awareness session serves none of them well. You will leave with the roles your decision depends on, the knowledge that must outlive the people who hold it, and an honest view of the incentives in your company that help or hinder.
25-Minute Module
0:00 – 5:00
Begin with the role
- The learning need follows the decision. For a field supervisor it may be checking the source of an alert, recognising its limits and escalating the condition. For a finance reviewer, testing benefit attribution and the timing of cash. For a technical authority, judging whether a model has been validated for a particular configuration.
- Map the tasks that will change. For each, name the judgement required, the evidence the user must check and the action the user may take. That gives you a competence standard you can observe.
- THE DOCTRINE
Train the decision, not the feature. A supervisor who can explain why an alert warrants investigation shows more than any course attendance record.
- Equally observable: an analyst who can distinguish a measured quantity from an estimate; a manager who can identify a decision outside delegated authority. Train on realistic, labelled cases that test uncertainty and disagreement, not the answer the software prefers.
If you cannot describe the behaviour you want, you cannot train it or test it.
5:00 – 10:00
Knowledge that survives a change of personnel
- Experienced staff hold knowledge that may never have reached a controlled procedure. It is valuable, and it still needs technical validation before it becomes an instruction.
- The book's illustrative case: a retiring supervisor knows that a recurring start-up difficulty follows a particular sequence of conditions. A knowledge assistant can help capture the account and connect it to records. That is where the assistant's part ends.
Stage What happens What it is not Capture The account is recorded and linked to the relevant records An approval Validation Engineering examines whether the explanation is correct, the response safe and the configuration current A formality Control The accepted lesson enters the controlled knowledge system with an owner and a review date A permanent truth - This also protects the expert. A recollection can be incomplete or tied to conditions that have changed, and treating it as permanent procedure creates risk for the company and the person.
The recording of a conversation is supporting evidence, not an automatic approval.
10:00 – 15:00
Supervisors make adoption practical
- People cannot learn, correct data and review recommendations while their workload stays unchanged. Supervisors should show where the application saves work and where it creates work: an alert can create an inspection requirement, a quality check a correction queue. Those tasks need capacity and priority.
- THE DOCTRINE
A culture that treats disagreement as disloyalty will encourage superficial acceptance and hidden workarounds. Allow users to challenge the system, and track the reasons for overrides.
- Override reasons are diagnostic. Some reveal a data problem. Some show the model does not understand a local constraint. Some show the approval process is too slow. Each has a different remedy, and lumping them together hides all three.
- AT THE ASSET · THE WORKED CASE
- Asset. On a fictional onshore asset, a new alert raises inspection requests for a crew whose week was already full. The supervisor logs each override with a reason; most cite a reading the crew knows to be stale.
IN THE OPERATING REVIEW · THE PARALLEL- Review. The monthly review sees a rising override count. The weak response is to call it resistance. The useful response reads the reasons, sends the stale-reading cases back to the data owner, and asks who released crew time for the new inspections.
A rejection supported by evidence is a gift to the application owner.
15:00 – 20:00
Incentives, consequences and Nigerian capability
- Performance measures shape behaviour. A department rewarded only for local throughput may resist a change that improves group margin but reduces its reported volume. A manager rewarded for closing work orders may find administrative closure more attractive than effective repair.
- THE DOCTRINE
Review incentives alongside the application. A model cannot compensate indefinitely for a performance system that rewards the wrong result.
- Employment data needs particular care. Nigeria's Data Protection Act 2023 provides the statutory framework for personal-data protection (Federal Republic of Nigeria, Nigeria Data Protection Act 2023). It is a reason to involve your privacy and legal owners in access, profiling or employee-data uses, not a substitute for checking the provisions relevant to the activity. An opaque score is never a complete assessment of a person; keep evidence, a responsible reviewer and a route for correction.
- Assess Nigerian operating capability by work the team can do independently: maintain the data contract, review a recommendation, diagnose a failure, recover the system, challenge a vendor. Permanent dependence on one external specialist may solve today's problem and leave you unable to sustain it.
Plan knowledge transfer and succession from the start, not at the end of the vendor contract.
20:00 – 25:00
Train the decision, then test the person
- An attendance sheet proves presence, not the ability to recognise an unreliable recommendation during a busy shift. The book's fictional exercise: give the user three recommendations, one supported by current records, one based on a stale signal and one outside the approved operating scope. Ask them to identify the source, state the permitted use and choose the authorised next step, with a route to obtain help when uncertain.
- Test the actual role. A field user must recognise limitations and escalate. An application owner must investigate model behaviour and manage withdrawal. A sponsor must judge the evidence for continued funding. If the exercise succeeds only when an instructor interprets every warning, revise the interface or process before wider deployment.
- Before you close: for your Session 1 decision, list the roles that touch it, the one piece of knowledge you would lose if a single person left, and the incentive most likely to bend the record. That is Section 7 of your 90-Day Intelligent Operating Plan. Write your assignment now.
Competence is part of the operating arrangement, not an attachment to the software purchase.
Key Terms
Competence standard
An observable statement, per role, of the judgement required, the evidence to check and the action permitted when using an application.
Controlled knowledge system
The repository of validated operating knowledge, where each accepted lesson carries an owner and a review date.
Override reason
The recorded explanation for rejecting a recommendation, used to separate data problems, local-constraint gaps and slow approvals.
Benefit attribution
The method used to determine how much of an observed change belongs to a particular intervention.
Self-Check — Before You Submit
1. A retiring supervisor has recorded how to handle a recurring start-up difficulty. The operations manager wants to issue the transcript as a procedure next week. What should happen first?
- A. Issue it, since the supervisor is the most experienced person on the asset
- B. Leave it out of the procedures entirely to avoid liability
- C. Engineering checks it is correct, safe and current, then it enters the controlled system with an owner and review date
- D. Ask the knowledge assistant to rewrite it as a formal procedure
C. Engineering checks it is correct, safe and current, then it enters the controlled system with an owner and review date — Captured knowledge needs technical validation before it becomes an instruction. Issuing it on reputation is the tempting error, and it exposes both the company and the expert if the recollection is incomplete or out of date.
2. Which evidence best shows that a field user is competent with a new alerting application?
- A. Given current, stale and out-of-scope recommendations, they identify the source, permitted use and authorised next step
- B. They attended the vendor's full training course
- C. They open the dashboard every day
- D. They accept nearly every recommendation the system makes
A. Given current, stale and out-of-scope recommendations, they identify the source, permitted use and authorised next step — The book tests the decision, not attendance or usage. Accepting every recommendation is the tempting wrong answer; a competent user recognises the stale signal and the out-of-scope recommendation.
3. A maintenance manager is rewarded mainly for the number of work orders closed. What risk does the book identify?
- A. Work orders will be raised too slowly
- B. Too much will be spent on spares
- C. Reliability data will become too detailed
- D. Administrative closure may become more attractive than effective repair
D. Administrative closure may become more attractive than effective repair — Performance measures shape behaviour, and a closure count rewards closing rather than fixing. A model fed by those records learns from the wrong result.
Today’s Assignment
Write Section 7 of your plan: the roles, knowledge, supervision and incentives your company's decision depends on.
- List the roles in your company that will use, check or act on the intelligence behind your decision, each with its judgement, evidence to check and permitted action.
- Name the knowledge held by one or two people that must survive a change of personnel, and how it will be validated, owned and reviewed.
- Describe the supervisor's part: where work is saved or created, whose capacity absorbs it, and how override reasons are reviewed.
- Name the incentives that help or hinder honest records. If employee data is involved, say how your privacy and legal owners take part.
- Describe a role-based test of the decision and which tasks still need an external specialist.
- The honest part: where your company currently measures training by attendance, or rewards a result that bends the record. If no AI is in use yet, work from the decision or proposal in front of you. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 7 — The roles, knowledge, supervision and incentives our decision depends on.
How this is marked — the rubric
| States a competence standard for each named role: judgement, evidence to check and permitted action | 25 |
| Names specific knowledge at risk and a validation route with owner and review date | 20 |
| Shows where work is created, who absorbs it, and how override reasons are used | 20 |
| Identifies at least one incentive that bends the record or decision, honestly | 20 |
| Describes a role-specific test and the current dependence on external specialists | 15 |
| Total · pass mark 70 | 100 |
Submitting session 7
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 7, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Role-based competenceKnowledge captureSupervisor-led adoptionIncentive reviewCapability building
Part Two · Subsurface, Wells and ProductionSESSION 8
Exploration, Appraisal and Reserves Stewardship
Why This Matters
A prospect can look more attractive as its presentation improves: the colour maps become clearer, the volumetric range narrows, a ranking score appears. None of that necessarily resolves the uncertainty that controls the investment. This session gives you one executive question to carry into any technical review: which uncertainty is worth paying to reduce before the company commits capital? If your company does not explore or appraise, do not skip it. Apply it to the operator decision you support, the partner case you must approve, or your nearest equivalent decision where a technical interpretation drives spending.
25-Minute Module
0:00 – 5:00
Separate the questions
- In plain language, four different questions hide behind one map. Exploration asks whether a potentially valuable accumulation exists. Appraisal examines its extent and behaviour. Development planning asks how it could be produced and monetised. Reserves stewardship concerns the classification and assurance of the quantities the company reports.
- They are connected, but they should not be collapsed into one predicted volume. A model can assist interpretation while leaving substantial uncertainty about deliverability, cost or commercial access.
- The Petroleum Resources Management System provides recognised principles for evaluating and classifying petroleum resources; SPE's official reserves and resources page identifies the 2018 PRMS and its updated release (SPE and partner societies, Petroleum Resources Management System, 2018).
- THE DOCTRINE
A machine-learning output does not independently establish a reserves classification. Distinguish a prospect's potential from the quantities and project status the applicable classification basis can support.
- This is a reporting discipline as well as a technical one. The operator applies the relevant definitions through competent evaluation and assurance.
When one number answers four questions, three of them have not been asked.
5:00 – 10:00
Use AI to challenge an interpretation
- Pattern recognition can assist the examination of seismic attributes, logs and historical interpretations, and help organise a large body of records to find inconsistent assumptions. The useful result is a better question for the geoscientist, not a verdict.
- Training data must fit the intended setting. An application trained on one geological environment may not behave reliably in another. Similar-looking attributes do not establish geological equivalence.
- Watch the labels. Where a model learned from historical interpretations, it may reproduce earlier assumptions rather than provide independent evidence. The reviewer should examine what the model learned and which conditions were absent from its evaluation.
- Compare several hypotheses. A single preferred interpretation may be necessary for planning, but the alternatives should remain visible, and the executive paper should explain what observation would support or weaken each material hypothesis.
Ask what the model was trained on before you ask what it found.
10:00 – 16:00
Pay for information that changes a choice
- Before you buy a test, ask: would any plausible result make us do something different? That is the idea behind value of information. It compares the best decision under present evidence with the expected decisions after more evidence, includes the cost and delay of acquiring it, and retains the option of declining the investment.
- Perfect information is a conceptual upper bound, not an achievable promise. A test can be imperfect or ambiguous, and its value depends on how the team will act on the result.
- AT THE ASSET · THE WORKED CASE
- Asset. The book's invented prospect has two principal uncertainties. The geological team is unsure whether the connected volume supports a large development. A new seismic interpretation could improve that question.
IN THE OPERATING REVIEW · THE PARALLEL- Review. Commercial is unsure whether a usable delivery route will exist within the proposed investment period. If route access is necessary under every development concept and cannot presently be secured, more subsurface detail may have little immediate decision value.
- The next useful expenditure might instead establish the route's feasibility and terms. Prepare a decision tree with the feasible choices, the possible evidence and how management would act on each result, including waiting or declining.
- THE DOCTRINE
Fund information that can change a choice, and keep the rest in a deliberate surveillance or later-appraisal plan.
- The book is careful here: this does not make commercial uncertainty more important in every prospect. It makes the controlling uncertainty specific to the decision.
The geological work may remain worthwhile later. The question is whether it is the next thing to buy.
16:00 – 21:00
The prospect review, and what to measure
Who presents What they bring to the committee Subsurface director The evidence, the alternatives and the uncertainty ranges Asset team The development concept and its constraints Commercial The route to market Finance A test of cost, timing and downside exposure - The committee should see which assumptions are measured, interpreted or adopted for screening. A narrow range is not automatically a better estimate; it can reflect omitted uncertainty.
- Keep technical and commercial probabilities separate where they describe different events. Do not multiply overlapping estimates as if they were independent, and do not treat technical success as proof that the commercial project will proceed.
- The decision record states the next evidence required and the conditions for proceeding, revising or stopping. That gives appraisal a purpose and makes later review possible.
- Interpretation speed can be measured, but speed alone does not prove better investment decisions. Examine forecast calibration, the relevance of additional information, the quality of independent review and how uncertainty changes as evidence arrives. Where an interpretation was wrong, preserve the reason rather than rewriting the earlier case.
When the range narrows between two papers, ask which evidence narrowed it.
21:00 – 25:00
Your wrap
- Before you close: take one resource or technical decision linked to your Session 1 decision. Write the separate questions behind it, the hypotheses still in play, and the one uncertainty that controls the choice.
- Operators will often choose a prospect, appraisal or reserves decision. A service company, a non-operated partner or a company with no subsurface activity should take the operator decision it supports, the partner case it must approve, or its own nearest decision where an interpretation drives spending.
- That is Section 8 of your 90-Day Intelligent Operating Plan.
- Pop Quiz 2 opens now — five questions, closed book, on Sessions 6 to 8. Sit it before Session 9, then write your assignment.
AI helps the team investigate at scale. Leadership keeps the uncertainty visible.
Key Terms
Value of information
The expected improvement in a decision from obtaining additional evidence, compared with its cost, delay and limitations.
Petroleum Resources Management System (PRMS)
The recognised principles for evaluating and classifying petroleum resources, applied through competent evaluation and assurance, not by a model's output.
Controlling uncertainty
The uncertainty whose resolution could change the investment choice; it is specific to the decision and not always the geological one.
Reserves stewardship
The classification and assurance of the quantities a company reports, kept distinct from a prospect's potential.
Self-Check — Before You Submit
1. A fictional operator's geoscientists want funds for further seismic interpretation. Commercial reports that an evacuation route is needed under every development concept and cannot presently be secured. What is the most defensible next expenditure?
- A. The seismic interpretation, because subsurface uncertainty always comes first
- B. Both studies at once, to remove all uncertainty
- C. Nothing until the reserves are classified
- D. Work on the route's feasibility and terms, with the geological work kept in a later-appraisal plan
D. Work on the route's feasibility and terms, with the geological work kept in a later-appraisal plan — When route access controls every concept, more subsurface detail may have little immediate decision value. Funding the seismic work by habit is the tempting error; fund the information that can change the choice.
2. A machine-learning model ranks a prospect highly and estimates its volume. What can that output establish on its own?
- A. No reserves classification; the operator applies the PRMS definitions through competent evaluation and assurance
- B. A reserves classification, if the model's accuracy is high
- C. A provisional classification pending review
- D. The commercial probability of the project proceeding
A. No reserves classification; the operator applies the PRMS definitions through competent evaluation and assurance — A machine-learning output does not independently establish a reserves classification. The provisional option is tempting because it sounds cautious, but it still lets the model classify.
3. This quarter's prospect paper shows a much narrower volumetric range than last quarter's. What is the right first reading?
- A. The estimate has improved and the prospect is less risky
- B. A better model has been adopted, so the range can be trusted
- C. Ask which evidence narrowed it, since a narrow range can reflect omitted uncertainty
- D. The appraisal programme can now be cancelled
C. Ask which evidence narrowed it, since a narrow range can reflect omitted uncertainty — A narrow range is not automatically a better estimate. Assuming improvement is the tempting reading; the committee should see which assumptions are measured, interpreted or adopted for screening.
Today’s Assignment
Write Section 8 of your plan: where AI could challenge your company's technical interpretation, and the information worth paying for.
- Choose one resource or technical decision linked to your Session 1 decision. Operators: a prospect, appraisal or reserves decision. A service company, non-operated partner or company without subsurface activity: the operator decision you support, the partner case you must approve, or your nearest decision where an interpretation drives spending.
- Separate the questions behind it and say which your company's papers collapse into one number. State the competing hypotheses and the observation that would support or weaken each.
- Describe how AI could challenge the interpretation, and where its training data or historical labels may not fit your setting.
- Name the controlling uncertainty and the information worth paying for: how management would act on a favourable and an unfavourable result, its cost and delay, and the option to wait or decline.
- Name what you would deliberately not buy now, and the conditions for proceeding, revising or stopping. The honest part: where your company has funded information that could not change the choice. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 8 — Where AI challenges our technical interpretation, and the information worth paying for.
How this is marked — the rubric
| Separates the questions behind the decision and identifies where they are currently collapsed | 20 |
| States competing hypotheses with the observation that would distinguish them, and the data-fit limits of any AI challenge | 25 |
| Identifies the controlling uncertainty and shows how a favourable and an unfavourable result would change the choice, with cost, delay and the option to decline | 35 |
| Names what is deliberately deferred, the proceed-revise-stop conditions, and an honest past example | 20 |
| Total · pass mark 70 | 100 |
Submitting session 8
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 8, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Pop Quiz 2 — Governance, People and Appraisal opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Pop Quiz 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Separating technical questionsChallenging interpretationsValue of informationProspect review disciplineCapital discipline
Part Two · Subsurface, Wells and ProductionSESSION 9
Reservoir Management and Recovery Decisions
Why This Matters
A reservoir recommendation can be technically persuasive and commercially incomplete. An infill well may add fluid that the surface system cannot handle. More injection may support pressure while tightening a disposal or integrity constraint. This is the binding-constraint discipline of Session 3 carried beneath the ground. Today you learn to keep physics in the discussion, to design surveillance around decisions rather than routine, and to tell additional recovery from earlier production, a difference that can move a capital decision. If your company does not manage a reservoir, apply the session to the operator decision you support or your nearest decision where a forecast drives action.
25-Minute Module
0:00 – 5:00
Keep physics in the discussion
- The reservoir team needs a defensible representation of the asset and the observations that constrain it. Pressure, production, injection, tests and intervention history should be examined together.
- A learned model can help recognise trends or give a faster approximation for repeated scenarios. That faster stand-in is a surrogate model. The team should know the conditions it was trained on and how its outputs compare with the engineering model it approximates. Good for screening; possibly unsuitable for a final decision outside its validated range.
- History matching, in plain terms, is adjusting a model until it reproduces what the reservoir has already done. It deserves careful language: several explanations can fit the same history.
- THE DOCTRINE
Agreement with past observations does not prove that a model represents the future correctly. Retain materially different cases where the evidence does not distinguish them.
- A learned model should never be used to conceal an uncertain physical relationship.
When someone says the model matches history, ask how many other models would too.
5:00 – 10:00
Design surveillance around decisions
- Surveillance is the programme of tests and observations used to watch the reservoir and the wells. It should answer questions the team intends to act upon. Which wells need a test? Where might an unexpected pressure response change the plan? Which observation would distinguish a mechanical problem from reservoir behaviour?
- A regular collection programme can continue faithfully while missing the data the current decision needs. Do not judge surveillance solely by activities completed.
- THE DOCTRINE
Rank missing information by consequence. A critical uncertainty affecting a large investment may justify a targeted test; a minor one may be handled through a bounded assumption.
- AI can help identify departures from expected behaviour and prioritise review. It should display the evidence and uncertainty, not simply label a well as underperforming.
The objective is adequate evidence, not unlimited data collection.
10:00 – 15:00
Rank recovery opportunities across constraints
- A recovery opportunity should state the technical action, expected response, timing, cost and dependencies. The comparison must include the surface system and the route to market, which Session 3 taught you to see as one system.
- The book's teaching scenario: three opportunities compete for a limited budget, an injection change, a workover and a new well. Each has a different time to first benefit and a different failure consequence. A ranking on incremental volume alone would miss both.
- AT THE ASSET · THE WORKED CASE
- Asset. In a fictional application of that scenario, the new well and the injection change both rely on the same limited surface capacity. Each looks attractive on its own; together they may not be feasible.
IN THE OPERATING REVIEW · THE PARALLEL- Review. The asset manager asks for the relevant cash flows and uncertainty for each, with the shared capacity modelled as an interaction. The ranking that reaches the committee shows time to first benefit and failure consequence beside volume.
- Two individually attractive proposals may not be simultaneously feasible. Where opportunities share equipment or capacity, model the interaction rather than adding the benefits.
A ranked list that ignores the constraint register is a wish list.
15:00 – 21:00
Recovery, acceleration and the evidence for action
- The analysis must show whether a proposal improves recovery, accelerates production or merely shifts production within the chosen horizon. Those have different economic meanings.
What the proposal does What the paper must show Changes expected ultimate recovery The supporting technical basis and its uncertainty, with alternative cases where later performance is poorly understood Leaves cumulative recovery unchanged but brings it earlier Call it acceleration; it may still be valuable because cash arrives earlier; include the costs and constraints that make it possible Looks additional only inside a short window Extend the comparison across an appropriate asset horizon so the later consequence is not omitted - The book's teaching scenario: an intervention raises forecast production for two years and lowers it later, and the presentation shows only the early period. Finance and subsurface should agree the reporting basis before the committee sees the ranking; AI can speed scenario comparison but should not choose the horizon to maximise apparent benefit.
- The evidence for action is shared work: the reservoir authority sets the technical basis, production and process engineering confirm the feasible response, integrity examines limits, and the asset manager resolves the combined decision and its resources.
- THE DOCTRINE
An optimisation output is supporting evidence, not permission. Where a proposal changes operating conditions, the approval record should identify the applicable procedure and authority.
- Set a review interval. Do not declare success after one favourable observation, nor continue indefinitely because the response is hard to read. Name the observation that would make the team change course.
Ask for the cumulative line, not just the first two years.
21:00 – 25:00
Learn without rewriting the baseline, and your wrap
- After action, compare observed behaviour with the approved expectation and the credible alternative. Account for changes in other wells, operating availability and surface constraints. The outcome cannot automatically be attributed to the reservoir application.
- Update the model when new evidence warrants it, but retain the earlier version and decision basis, the same source-to-decision path you protected in Session 4. The company needs to learn how its judgement changed, and a later review must be able to tell a reservoir error from a timing or execution error.
- Before you close: write down the decision your surveillance must serve, the one opportunity you would rank, and whether its benefit is recovery or acceleration. That is Section 9 of your 90-Day Intelligent Operating Plan. Write your assignment now.
The useful executive report explains the remaining uncertainty and the next choice, not every simulation.
Key Terms
History matching
Adjusting a reservoir model until it reproduces past observations; agreement with history does not prove the model represents the future.
Surrogate model
An approximation used to represent another model or relationship within a validated range, suitable for screening but not automatically for final decisions.
Acceleration
Producing the same cumulative volume earlier; it may be valuable because cash arrives sooner, but it is not additional recovery.
Surveillance
The tests and observations used to watch wells and reservoir behaviour, designed around the decisions the team intends to make.
Self-Check — Before You Submit
1. A fictional recovery paper shows an intervention raising production for two years and calls every extra barrel incremental recovery. The later years are not shown. What should the executive ask for?
- A. Approval on the two-year uplift, since early cash is valuable
- B. A comparison over an appropriate asset horizon that separates cumulative recovery from timing
- C. A longer AI forecast that maximises the apparent benefit
- D. A new history match before any discussion
B. A comparison over an appropriate asset horizon that separates cumulative recovery from timing — A short window can make acceleration look like additional recovery. Approving on the uplift is tempting because earlier cash can be valuable, but the paper must name it correctly and show the later consequence.
2. The reservoir team reports that its model now closely matches the production history. What does that establish?
- A. That the model will forecast the future correctly
- B. That the model can be used outside its validated range
- C. That surveillance can be reduced
- D. Agreement with the past only; several explanations may fit, so retain materially different cases
D. Agreement with the past only; several explanations may fit, so retain materially different cases — History matching shows agreement with past observations, not correctness about the future. Treating a close match as a forecast guarantee is the tempting error.
3. A surrogate model trained on a narrow range of conditions is proposed as the basis for a final investment decision well outside that range. What is the right position?
- A. Useful for screening, but possibly unsuitable for a final decision outside its validated range
- B. Acceptable, because surrogates are faster than the engineering model
- C. Acceptable if its average error is low
- D. It must be discarded entirely
A. Useful for screening, but possibly unsuitable for a final decision outside its validated range — Speed and a low average error are tempting arguments, but they say nothing about behaviour outside the trained conditions. Discarding it ignores its legitimate screening use.
Today’s Assignment
Write Section 9 of your plan: surveillance designed around your company's decision, one opportunity ranked across constraints, and the evidence for action.
- State the decision your surveillance must serve. Operators: a recovery or production decision. A service company, non-operated partner or company without a reservoir: the operator decision you support, or your nearest decision where a forecast drives action.
- Design surveillance around it: the questions the team will act on, the observations that answer them, and missing information ranked by consequence.
- Rank one opportunity across constraints, including the surface system, route to market and failure consequence, and say whether it improves recovery, accelerates production or merely shifts it.
- Agree the evidence for action: who confirms what by role, the procedure and authority, the review interval, and the observation that would change course. Say how the baseline and earlier model version are preserved.
- The honest part: where your company currently judges surveillance by activities completed, or presents earlier production as additional recovery. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 9 — Surveillance designed around decisions, and the evidence for action.
How this is marked — the rubric
| Links surveillance to a named decision and ranks missing information by consequence | 25 |
| Ranks the opportunity across constraints and correctly distinguishes recovery, acceleration and shift | 25 |
| Names the evidence for action by role, with authority, review interval and the change-course observation | 30 |
| Preserves the baseline and earlier version, and gives an honest current weakness | 20 |
| Total · pass mark 70 | 100 |
Submitting session 9
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 9, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Physics-based challengeDecision-led surveillanceConstraint-aware rankingRecovery versus accelerationBaseline discipline
Part Two · Subsurface, Wells and ProductionSESSION 10
Drilling, Completions and Well Delivery
Why This Matters
A well programme becomes an operating commitment long before the rig arrives. Design assumptions, equipment choices, contractor capacity and logistics fix what the team can do when conditions change. The book's position is short: AI can assist that preparation; it cannot take responsibility for the well. This session closes Week 2 by testing whether you can say where prediction stops and authority begins, and whether your contractor and logistics interfaces were designed or merely assumed. If your company does not drill, apply it to the well programme you support or your own nearest delivery commitment, where the same discipline applies.
25-Minute Module
0:00 – 5:00
Start with the well objective
- State why the well is being drilled or completed: the technical objective, the evidence required to establish success, and the conditions that would change the plan. Build the schedule around that objective; the most optimistic sequence is not necessarily the most useful forecast, so include what could delay mobilisation, equipment delivery, approvals or the operating window.
- Offset-well analytics, in plain terms, compares a planned well with previous wells to learn from their difficulties and test candidate designs. The reviewer should establish which wells are relevant and why; a model should not treat every historical record as equally comparable.
- Completion decisions connect reservoir requirements, sand or fluid behaviour, measurement, integrity and later intervention access. An apparently cheaper design may create a more expensive operating obligation.
- THE DOCTRINE
AI can assist the preparation of a well. It cannot take responsibility for the well, from the investment paper through execution, handover and review.
If the objective is not written, the schedule will become the objective.
5:00 – 10:00
Prediction and well-control authority
- In plain language, well control is about keeping fluids where they belong, inside the barriers designed to hold them in the well or the reservoir. A drilling application may spot a departure in a signal or help interpret an emerging pattern. Its output belongs inside an authorised review process.
- IOGP's Well Control Incident Definitions describes a well-control incident through failure or degradation of the defined barrier envelope intended to retain fluids in the well or reservoir, and exists to support consistent classification and learning (IOGP, Well Control Incident Definitions, Report 660).
- That supports a barrier-focused discussion; it is not an operating procedure. The response remains governed by the approved programme, competent personnel and the actual conditions. Evaluate the application on unusual operations and degraded data too: a signal that works in routine drilling may be unreliable during a transition.
- THE DOCTRINE
This book does not prescribe well-control actions. The leadership requirement is competent authority, tested escalation and sufficient resources for the approved response.
Ask what the application has been tested on that was not routine.
10:00 – 15:00
Contractor interfaces are part of the design
- The operator, the drilling contractor and the service providers need a shared understanding of who supplies, verifies and acts on information. A responsibility that is unclear before mobilisation is unlikely to become clearer under pressure.
Interface What must be agreed before mobilisation Data Access and ownership The record How observations are communicated and which record forms the official operational account Recommendations How an analytical recommendation enters the existing supervision and decision process Disagreement A technical reconciliation when the contractor's model and the operator's model use different evidence, not competing confidence scores Failure and hours Support outside normal hours, and the procedure when the analytical service fails - The operation must not depend on an unavailable specialist to interpret every exception. This is the vendor-exit and recoverability thinking of Session 5, written into the well plan.
Settle which record is official while everyone is still calm.
15:00 – 20:00
Logistics, handover and the performance review
- Remote, swamp and offshore locations bring mobilisation, access and contingency requirements. A part that looks inexpensive in the bill of materials can control the entire delivery date; AI-assisted forecasting earns its place when it shows that dependency early enough to act.
- Do not apply a generic contingency and assume every risk is covered; some events hit several costs and the time to first production at once. If two campaigns need the same vessel or specialist team, leadership sets an explicit priority before it becomes a daily execution problem.
- A delivered well needs a usable operating record: accepted configuration, integrity evidence, measurement basis, procedures, limitations and outstanding actions. An assistant can check completeness; technical authorities verify the content.
- THE DOCTRINE
A complete folder is not proof of a correctly completed well.
- Review technical objectives, barrier assurance, non-productive time, cost and schedule variance, handover quality and the initial operating response, and preserve lessons with the conditions to which they apply.
Find the cheap part with the long lead time before it finds you.
20:00 – 25:00
The morning drilling review, and your wrap
- AT THE ASSET · THE WORKED CASE
- Asset. The book's fictional well is behind schedule. Analytics has ranked the events associated with delay. One delay category combines several causes; another includes time deliberately spent obtaining necessary assurance. A model that learns every long period as undesirable can recommend removing work whose purpose it has not understood.
IN THE OPERATING REVIEW · THE PARALLEL- Review. Before using the ranking, the drilling leader examines the event definitions and labels. The review separates avoidable operational delay, planned activity, required verification and unresolved interpretation; the drilling team owns that distinction.
- The output is a bounded action, such as readying a required service or correcting an incomplete work package, with an authorised owner and the evidence expected at the next review. No change to the approved well programme comes from the ranking.
- Before you close: write the delivery objective behind your decision, the point where prediction stops and authority begins, and the interface you have assumed rather than agreed. That is Section 10 of your 90-Day Intelligent Operating Plan, and it closes Week 2.
- Practice Exam 2 opens now — fifteen questions across Sessions 1 to 10. Sit it before Week 3, then write your assignment.
A schedule target must never displace the well's technical requirements.
Key Terms
Barrier envelope
The defined barriers intended to retain fluids in the well or reservoir; IOGP describes a well-control incident through their failure or degradation.
Offset-well analytics
Comparing a planned well with relevant previous wells to examine past difficulties and candidate designs, after establishing which wells are genuinely comparable.
Long-lead item
Equipment or a service whose delivery time can control the schedule, regardless of its cost in the bill of materials.
Handover record
The operating record of a delivered well: accepted configuration, integrity evidence, measurement basis, procedures, limitations and outstanding actions.
Self-Check — Before You Submit
1. At a fictional morning drilling review, an analytics ranking recommends cutting a recurring period labelled as delay. Some of that time was spent obtaining required verification. What should the drilling leader do?
- A. Cut the period, since the model was trained on many wells
- B. Ask for the model to be retrained before the next review
- C. Examine the event definitions and labels, separate the kinds of time, and set a bounded action with an owner
- D. Change the approved well programme to remove the period
C. Examine the event definitions and labels, separate the kinds of time, and set a bounded action with an owner — A model that treats every long period as undesirable can recommend removing work whose purpose it does not understand. Cutting on the model's authority is the tempting error; the drilling team owns the distinction.
2. What does the book say IOGP's Well Control Incident Definitions (Report 660) is for?
- A. Consistent classification and learning, describing incidents through failure or degradation of the barrier envelope
- B. Prescribing the well-control response for each type of event
- C. Certifying drilling analytics for use in well control
- D. Setting Nigerian regulatory reporting deadlines
A. Consistent classification and learning, describing incidents through failure or degradation of the barrier envelope — The book uses it for a barrier-focused discussion and says it is not an operating procedure. Reading it as a response procedure is the tempting error; the response stays with the approved programme and competent personnel.
3. During a fictional campaign, the contractor's model and the operator's model disagree about an emerging pattern. What should happen?
- A. Follow whichever model reports the higher confidence score
- B. Follow the operator's model, since the operator holds the licence
- C. Pause all analytics until the vendors agree
- D. Reconcile the two evidence bases technically, within the authorised review process
D. Reconcile the two evidence bases technically, within the authorised review process — Different models may use different evidence, so the book calls for technical reconciliation, not competing confidence scores. Choosing by confidence or ownership is quick, but neither examines the evidence.
Today’s Assignment
Write Section 10 of your plan: the delivery objective behind your company's decision, the limits of prediction, and your contractor and logistics interfaces.
- State the delivery objective behind your decision. Operators: the well or completion, with its technical objective, the evidence of success and the conditions that would change the plan. A service company, non-operated partner or company that does not drill: the well programme you support, or your own nearest delivery commitment.
- Say where prediction stops and authority begins: what analytics may flag, who reviews it, and which authority acts. For barrier-related signals, state the escalation and resources leadership must provide, without prescribing well-control actions.
- Write the contractor interfaces your company must agree before mobilisation, using the five in the session table, and the logistics dependencies: long-lead items, shared vessels or teams, and who sets priority.
- Describe what your morning review needs: the delay categories it must separate, the form of a bounded action with an owner, and what a usable handover record must contain.
- The honest part: which interface your company currently assumes rather than agrees, and where a schedule target has pushed against a technical requirement. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 10 — The delivery objective, the limits of prediction and our contractor interfaces.
How this is marked — the rubric
| States a delivery objective with the evidence of success and the conditions that would change the plan | 20 |
| Draws a clear line between prediction and authority, with escalation and resources named by role | 30 |
| Specifies contractor and logistics interfaces concretely, including the official record and service failure | 30 |
| Sets out what the morning review and handover need, and names an honest current weakness | 20 |
| Total · pass mark 70 | 100 |
Submitting session 10
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 10, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Practice Exam 2 — Foundations and the Subsurface opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Practice Exam 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Objective-led well planningPrediction versus authorityContractor interface designLogistics dependenciesHandover assurance