THE INTELLIGENT PIA LEADERSHIP PROGRAMME
The Petroleum Industry Act belongs in the operating model.
15-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s petroleum industry
WEEK 3 OF 3 · TRUST, CAPITAL AND THE NINETY DAYS · SESSIONS 11–15 · WORK AT YOUR OWN PACE
Trust and Control · Chapter 12SESSION 11
AI Governance for PIA Decisions
Why This Matters
Week 3 begins with trust, because every tool you have specified so far will be used for something it was never approved to do. Nobody plans that. A tool approved for document support is convenient, so people start using it to interpret a licence or prepare a submission, and nobody records the change. This session gives you the governance record that stops an application becoming a moving operating assumption: approved purpose, owners, data boundary, authority, and who may accept a change. It also covers the two things governance most often forgets — whether human review is real, and whether personal data has a lawful basis.
25-Minute Module
0:00 – 5:00
How governance drifts
- A company approves an assistant for document support. Six months later it is interpreting licences, assessing suppliers and preparing submissions. No one decided that. Users discovered convenience and extended the use beyond what was ever evaluated.
- I have sat in the meeting where this is discovered. Without a record of what the tool was allowed to do, you cannot even say the use has drifted.
- THE DOCTRINE
Governance should follow the decisions the system can influence: approved purpose, owner, users, data boundary and authority, and who can accept a change. Otherwise the application becomes a moving operating assumption.
- Notice what is not on that list: a policy document. A large policy cannot compensate for uncontrolled use at the point of work.
Ask your team which AI tools they actually used last week. Then ask who approved each use. The gap between the two lists is your first finding.
5:00 – 11:00
The governance record
- For each application, keep one record that can be used, not filed.
Field What it records Intended use The approved purpose, and the authorities it is expressly excluded from Owners Process owner, model owner, data owner and accountable executive, each named Evaluation evidence What was tested, the results and the limits accepted Monitoring and rollback How performance is watched and how the use is reversed Change triggers A new source, a new model or a new connection, each reassessed against consequence - The change triggers matter most. A new source introduces a new population of permissions. A new model may alter interpretation. A new connection may permit transactions. Each is a different application wearing the old name.
- NIST’s AI Risk Management Framework offers a voluntary structure around Govern, Map, Measure and Manage. It can inform your design. It is not Nigerian petroleum law and it is not evidence of regulatory approval.
- Governance should be proportionate. A read-only search service and a transaction agent both need ownership, evaluation and a boundary; the level of consequence determines the additional controls.
If the record cannot tell you who may accept a change, it is a description, not a control.
11:00 – 16:00
Real review, and personal data
- Human review is meaningful only if the reviewer has competence, time and access to evidence. A polished document arriving minutes before a deadline leaves no practical opportunity to challenge it. A signature under those conditions is a formality.
- Define the review for each kind of output. A licence response needs the current instrument, facts, attachments and authority. A fiscal calculation needs the period, rule, inputs and treatment. An engineering recommendation needs its technical basis and limits.
- Record material corrections and the reasons. A system that keeps only accepted outputs cannot show how much repair it needed.
- THE DOCTRINE
Keep reviewers able to disagree. A process that rewards speed while penalising requests for evidence turns human approval into a formality. Make the quality of review part of acceptance.
- The Nigeria Data Protection Act 2023 is a separate legal framework. Community, employee and supplier records can hold personal information: establish lawful basis, purpose, access, retention and any transfer arrangements before processing. Significant automated decisions involving personal data need specific assessment.
- Consent is not the only lawful basis — and “compliance requires it” does not authorise unlimited processing. Minimise: a project summary may need a category and a case identifier, not names, addresses or medical details.
Ask your most senior reviewer how long they had with the last pack.
16:00 – 21:00
A fictional drift, and the first governance pilot
- An assistant is approved to retrieve training records. A supervisor starts using it to rank employees for a safety-critical assignment. The original evaluation never tested ranking accuracy, fairness or competence for that decision.
- Stop the unapproved use; let retrieval continue within its accepted boundary; investigate whether rankings influenced assignments and correct them through proper authority. It is a governance failure even if one ranking happens to be right.
- UPSTREAM · THE WORKED CASE
- Upstream. Illustrative: at a fictional operator’s flow station, the training-records assistant ranks technicians for a safety-critical task, and the ranking enters the crew plan unreviewed. The fix is the record, not a better ranking.
ACROSS THE CHAIN · THE PARALLEL- Midstream and downstream. Illustrative: at a fictional depot, an assistant approved to retrieve documents is drafting readings of permission conditions. Retrieval continues; interpretation returns to the authorised specialist.
- The first governance pilot takes an existing application and maps actual use against approved use. Observe how outputs enter official records, ask reviewers whether evidence arrives in time, and test access restrictions and change approval.
- Define escalation for material errors, privacy events and unauthorised actions, and name who can suspend the service and how work continues manually. For the board: approved uses, material corrections, overdue reviews, failed access tests and incidents.
Governance is working when it changes operating behaviour and makes responsibility visible. A thicker policy file proves neither.
21:00 – 25:00
Before you close the laptop
- Write down three things: one AI tool in use or proposed in your company, the use nobody approved, and the person who could suspend it tomorrow.
- This session produces Section 11 of your 90-Day PIA Game Plan — the governance record and authority boundary.
- Govern the tool closest to your Session 1 obligation.
If you cannot find an AI tool in use, look harder. It is usually in a personal account.
Key Terms
Governance record
The usable record for one AI application: intended use and excluded authorities, named owners, evaluation evidence, accepted limits, monitoring and rollback.
Excluded authority
A decision or action the application is expressly not approved to take or influence, such as interpreting a licence or ranking staff.
Meaningful review
Review by a person with the competence, time and access to evidence needed to challenge an output, and freedom to disagree.
Lawful basis
Under the Nigeria Data Protection Act 2023, the legal ground for processing personal data. Consent is one basis, not the only one.
Self-Check — Before You Submit
1. A depot manager discovers that an assistant approved only for document retrieval has been used for three months to rank drivers for hazardous loads. One ranking was later checked and found sound. What is the right response?
- A. Approve the ranking use retrospectively
- B. Stop the ranking use, let retrieval continue within its boundary, and investigate and correct affected assignments through authority
- C. Suspend the whole assistant permanently
- D. Add a disclaimer to ranking outputs and continue
B. Stop the ranking use, let retrieval continue within its boundary, and investigate and correct affected assignments through authority — The ranking use was never evaluated, so it is a governance failure even if one result was right; retrieval was approved and can continue. Retrospective approval rewards drift, and a total suspension discards a use that was properly accepted.
2. Which statement about the NIST AI Risk Management Framework is accurate?
- A. It is part of the Nigerian petroleum regulatory framework
- B. Following it demonstrates regulatory approval of an AI tool
- C. It replaces the need for a company governance record
- D. It is voluntary guidance around Govern, Map, Measure and Manage that can inform governance
D. It is voluntary guidance around Govern, Map, Measure and Manage that can inform governance — The book is explicit: NIST’s framework is voluntary and can inform design, but it is neither Nigerian petroleum law nor evidence of approval. The tempting error is treating alignment with a respected framework as a regulator’s acceptance.
3. A tax manager receives an AI-prepared fiscal schedule forty minutes before submission and signs it. What has the company actually demonstrated?
- A. Very little: review needs competence, time and access to evidence, and forty minutes may leave no practical chance to challenge it
- B. Full human review, because a qualified person signed
- C. That the AI tool is accurate enough for unsupervised use
- D. Compliance with the Nigeria Data Protection Act 2023
A. Very little: review needs competence, time and access to evidence, and forty minutes may leave no practical chance to challenge it — A signature proves someone signed, not that the period, rule, inputs and treatment were examined. The trap is equating a qualified signatory with meaningful review when the process left no time to challenge the output.
Today’s Assignment
Write the governance record for one AI tool in use, or proposed, in your company, including how its actual use differs from its approved use.
- Name the tool, its approved purpose and the authorities it is excluded from.
- Name the process owner, model owner, data owner and accountable executive by role. Where one is missing, say so.
- Describe actual use at the point of work — ask the people using it — and set it beside approved use. Name each drift and what you will do about it.
- Define the review for its most consequential output: what the reviewer needs, how long they get, and how a material correction is recorded.
- State the personal-data position: whether personal information is processed, the lawful basis, and what you would minimise.
- Name who can suspend the tool and how work continues manually. If your company has no AI tool in use, govern the one most likely to be proposed for your Session 1 obligation.
YOUR GAME PLAN
Section 11 — The governance record and authority boundary.
How this is marked — the rubric
| Approved purpose and excluded authorities stated precisely, with the four owners named by role and gaps admitted | 30 |
| Actual use observed and compared with approved use, with at least one honest drift named and a response stated | 30 |
| Review defined for the most consequential output, including time and evidence, and a correction record | 20 |
| Personal-data treatment, suspension authority and manual continuation each stated | 20 |
| Total · pass mark 70 | 100 |
Submitting session 11
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: 11, 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
AI governance recordsDrift detectionMeaningful human reviewData protection by designSuspension authority
Trust and Control · Chapters 13 and 14SESSION 12
Audit Readiness and Controlled Autonomy
Why This Matters
Months after a return is filed, someone asks why a figure changed and who approved the adjustment. If the honest answer is a person’s memory, the company is not audit-ready, however good the final document looks. This session joins two disciplines that belong together: building a submission so it can be reconstructed by a stranger, and giving an AI agent enough authority to help without enough to act on a misunderstanding. Both come down to the same question — can you prove what actually happened, step by step? By the end you will specify your recurring return and draw the boundary any agent near it must respect.
25-Minute Module
0:00 – 4:00
The return nobody can reconstruct
- A return is assembled near its deadline. Functions reconcile differences under pressure, write a narrative and submit. The final document exists. Its decision history does not.
- Audit readiness begins before submission: a repeatable chain through sources, calculation, review, signature, submission and receipt.
- THE DOCTRINE
AI can assemble the chain and expose its gaps. It must not make an incomplete chain appear complete.
Pick last quarter’s return and ask who approved its largest adjustment. Time how long the answer takes.
4:00 – 10:00
The submission package and the assurance chain
- Give each recurring submission an identifier and an approved specification: regulator, entity, asset coverage, period, due basis, template, evidence, reviewer and signatory, linked to its parent obligations.
- Build evidence backwards from the final statement. A measured quantity needs its approved source and reconciliation; a completed action needs acceptance evidence; an estimate needs its method and authorisation.
- Freeze submitted packs as controlled records. A correction produces a new version with a reason and an approved resubmission route. Editing the stored final file disconnects your records from what the recipient received.
Event What it shows What it does not yet show Transmission initiated Sending began That sending finished Transmission completed Sending finished That the recipient has it Receipt acknowledged The recipient has it That it was accepted Acceptance confirmed The recipient accepted it Nothing further on that submission - Distinguish no evidence found from action not performed: one may be a filing weakness, the other an execution weakness, and the owner investigates before anyone concludes. A neatly populated form using the wrong period is still wrong.
- The assurance chain: the process owner performs the work; a separate reviewer checks material assertions; independent assurance samples completed packs — clean ones as well as flagged ones — and reconstructs them without relying on the preparer. Readiness must survive the preparer leaving.
If only the application chooses what assurance examines, nobody is testing what the application missed.
10:00 – 15:00
Agents: start with a bounded job
- An agent calls tools, retrieves records, creates tasks and prepares transactions. That makes it useful for repeated work, and places authority at the centre of the design. A model that misreads a document produces a wrong draft; an agent with excess permissions acts on the misreading.
- Start small: monitor an approved calendar, find missing preparation evidence and create owner review tasks; or assemble a draft pack in staging. Both run with limited rights.
- Specify permitted sources, tools, destinations and changes as an allowlist. Occasional access to one document does not justify general access to a system.
- THE DOCTRINE
Keep approval separate from execution. Approval refers to an exact version and scope, not a general instruction to proceed. The agent has its own identity — never a manager’s credentials.
- A separate service identity makes actions traceable and revocation simple. Permissions follow the agent’s job, not its sponsor’s seniority.
Ask which credentials your most automated workflow runs under. If the answer is a person’s name, you have found the first fix.
15:00 – 20:00
Untrusted instructions and reliable transactions
- A retrieved document may contain text telling the agent to ignore rules, send information elsewhere or use another account. Treat it as material to analyse, not authority. The boundary belongs in tool and permission design — restricted destinations, file locations and transaction types — not only in prompts.
- UPSTREAM · THE WORKED CASE
- Upstream. Illustrative: a fictional operator’s renewal agent finds an attachment instructing it to send the complete pack externally. It ignores the instruction as an action request, preserves the letter as evidence and stays in staging. Where inspection acceptance cannot be found, it raises an exception rather than substituting a related report.
ACROSS THE CHAIN · THE PARALLEL- Midstream and downstream. Illustrative: a fictional terminal’s agent gets no response after submitting a return. It does not resubmit. Receipt is investigated first, transaction identifiers and receipt evidence are stored, and a model’s sentence saying “submitted successfully” is not treated as proof of delivery.
- Every task operation needs an identifier, and after a failed response you check whether the transaction succeeded before retrying. One communications error can otherwise create several tasks or duplicate submissions.
- Provide and practise manual recovery before the first incident: an authorised operator suspends the agent, reaches the sources and continues the work.
- Pilot with representative documents and dummy destinations; test missing attachments, conflicting dates, wrong entity, duplicate task, denied access, hostile instructions and interruption. Examine the logs beside the final narrative, then run shadow use before any consequential execution.
A correct final answer can conceal an unauthorised intermediate action. Read the log, not just the output.
20:00 – 25:00
Before you close the laptop
- Write down your company’s most important recurring return, its signatory, and where its evidence chain would break if the preparer left tomorrow.
- This session produces Section 12 of your game plan — the submission package and the agent boundary.
- Pop Quiz 3 — Governance and Assurance opens now: five questions on Sessions 11 and 12, closed book. Sit it before Session 13.
For the board, count completed work after controlled-system confirmation. Agent activity is not reliable execution.
Key Terms
Submission package
A recurring return’s identifier and approved specification: regulator, entity, asset coverage, period, due basis, template, evidence, reviewer and signatory.
Frozen pack
A submitted pack held as a controlled record. Corrections create a new version with a reason, never an edit to the stored final file.
Allowlist
The explicit list of sources, tools, destinations and changes an agent may use. Anything not on it is refused.
Service identity
A separate account under which an agent acts, so its actions are traceable and revocable, and its permissions reflect its job rather than its sponsor’s authority.
Self-Check — Before You Submit
1. Your agent submitted a monthly return to the regulator’s portal and the connection timed out before any response. The deadline is today. What happens next?
- A. Resubmit immediately so the deadline is not missed
- B. Accept the agent’s log line saying the submission succeeded
- C. Investigate whether receipt occurred before any resubmission, using stored transaction identifiers and receipt evidence
- D. Edit the frozen pack and send a corrected version to be safe
C. Investigate whether receipt occurred before any resubmission, using stored transaction identifiers and receipt evidence — An uncertain response requires investigation of receipt before another submission, otherwise you risk duplicate or contradictory records. Resubmitting feels prudent under deadline pressure, and a model’s success sentence is not proof of delivery.
2. Which sequence correctly orders the receipt events for a submission?
- A. Transmission initiated, transmission completed, receipt acknowledged, acceptance confirmed
- B. Receipt acknowledged, transmission completed, acceptance confirmed, transmission initiated
- C. Transmission completed, acceptance confirmed, receipt acknowledged, transmission initiated
- D. Acceptance confirmed, receipt acknowledged, transmission initiated, transmission completed
A. Transmission initiated, transmission completed, receipt acknowledged, acceptance confirmed — Each later event proves something the earlier one does not; treating acknowledgement as acceptance is the common error. The distinction matters whenever a deadline or a payment depends on the outcome.
3. A sponsor proposes running a renewal agent under the regulatory manager’s own login “so it can reach everything it needs”. What is the best response?
- A. Agree, because the manager is accountable anyway
- B. Agree, provided the prompt forbids external sending
- C. Agree for the pilot only and revisit at scale
- D. Refuse: give the agent its own service identity with an allowlist matched to its job
D. Refuse: give the agent its own service identity with an allowlist matched to its job — A manager’s credentials make tracing and revocation harder and grant the sponsor’s broad authority to a bounded job. A prompt instruction alone is not a boundary; restrictions belong in tool and permission design.
Today’s Assignment
Specify the submission package for your company’s most important recurring return and, if an agent were introduced, its exact permitted actions, approval point and recovery route.
- Write the package specification: identifier, regulator, entity, asset coverage, period, due basis, template, evidence, reviewer and signatory, and the parent obligation.
- Trace the last submission’s largest figure backwards to its source. Say honestly where the chain breaks and whether the problem is filing or execution.
- State how your business freezes submitted packs and records the four receipt events today. If it does not, say so.
- Describe the assurance chain: who performs, who reviews, and who independently samples — including clean cases.
- For an agent near this return: list its allowlist, the exact version-and-scope approval point, its service identity, and the manual recovery route. If your company is not a filer, use the report or certificate your work supplies to a licence holder.
YOUR GAME PLAN
Section 12 — The submission package and the agent boundary.
How this is marked — the rubric
| Complete package specification naming the regulator, period, due basis, reviewer and signatory by role | 25 |
| One figure traced backwards to source with an honest break point identified as filing or execution weakness | 25 |
| Freezing, versioning and the four receipt events addressed, with the assurance chain including independent sampling of clean cases | 20 |
| Agent boundary specified as an allowlist, a separate service identity, a version-specific approval point and a practised recovery route | 30 |
| Total · pass mark 70 | 100 |
Submitting session 12
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: 12, 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 3 — Governance and Assurance 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 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Audit-ready submissionsEvidence built backwardsAssurance samplingBounded agentsTransaction reliabilityHostile-input defence
Capital and Capability · Chapters 15 and 16SESSION 13
Fiscal Intelligence, Retirement and Nigerian Capability
Why This Matters
Sooner or later your PIA work meets capital: an acquisition, an extension, a retirement liability, a budget request for the evidence programme itself. That is where weak distinctions become expensive. A sensitivity gets mistaken for a statutory computation, a provision for cash in a fund, a contractor’s promise for your own capability. This session teaches the fiscal and retirement distinctions, walks two worked numbers from the book, and asks you to write the honest capital case for your own initiative — including the uncomfortable result when the arithmetic does not pay. Then it asks which capability your company must own rather than rent.
25-Minute Module
0:00 – 5:00
The fiscal rule set is current law, not the original text
- Acquisition, extension and development need a defensible account of permissions, tax treatment, liabilities, retirement scope and funding. These are not specialist side-matters.
- The Nigeria Tax Act 2025 contains petroleum taxation provisions in Chapter Three and consequential PIA amendments in section 197. A fiscal engine should use operative law and transition treatment, not assume the original PIA fiscal text is unchanged.
- Maintain rules by entity, asset, regime and period. Specialists approve classification, deductions, allowances and filing assumptions; deterministic computation uses those parameters; the language layer explains differences and missing evidence. There is no universal rate schedule, and AI must not reduce different regimes to one tax percentage.
- THE DOCTRINE
An investment sensitivity may use an assumed tax rate — labelled. A statutory computation needs the applicable rule, approved inputs and evidence. An assumption does not become a filing basis because it appears in a board paper.
- Keep current and historical periods distinct. A later law may change future treatment without rewriting earlier decisions.
Ask your finance director which instrument your fiscal model cites. If it is only the PIA, you have found a review.
5:00 – 11:00
Retirement: liability, schedule and cash are different things
- The Nigerian Upstream Petroleum Decommissioning and Abandonment Regulations 2026 address plans, updates, asset records, funds and reporting. Regulation 25 revokes the 2023 regulations with savings, so an upstream process relying only on the former instrument needs review.
- Link physical inventory, engineering scope, estimate, approved plan, contribution and fund evidence. Financial recognition, an approved funding schedule and cash held are related but distinct.
Illustrative funding sensitivity Figure Assumed retirement estimate today $20 million Horizon and assumed cost escalation 10 years at 4% a year Projected nominal cost ($20m × 1.0410) ≈ $29.605 million Spread evenly over ten years ≈ $2.960 million a year Assumed fund balance $8 million Gap, before contributions, investments, exchange and scope changes ≈ $21.605 million - This is arithmetic illustrating escalation and timing. It is not a statutory contribution formula or a substitute for the approved funding basis. Test shorter life and higher cost: if retirement comes in six years, a ten-year assumption is potentially misleading.
- In diligence, a clean data room is not proof of a clean asset.
Ask for three numbers on one page: the recognised liability, the required contribution and the cash held. If you receive one, ask again.
11:00 – 16:00
The honest AI investment case
- Now the evidence programme itself. These are the book’s teaching inputs, illustrative only.
Illustrative investment case Figure Establishment cost ₦30 million Annual operating cost ₦12 million Demonstrated recurring improvement (reduced rework and preparation) ₦24 million a year Net annual benefit ₦12 million Simple payback (₦30m ÷ ₦12m) 2.5 years Present value of three ₦12m benefits at 15% ≈ ₦27.399 million NPV (₦27.399m − ₦30m) ≈ −₦2.601 million - On that basis alone the project is not financially attractive. Say so. Other justified reasons may still support proceeding — necessary control repair, for one.
- THE DOCTRINE
State the real reasons instead of inventing avoided fines. Leadership can approve essential work while remaining honest about its financial basis.
The case I trust is the one that shows a negative NPV and still argues for proceeding on stated grounds.
16:00 – 21:00
Capability you own, not rent
- Contractual flow-down can require suppliers to perform controls or supply records. It does not transfer the licence holder’s statutory accountability.
- The Nigerian Oil and Gas Industry Content Development Act 2010 is a separate statutory basis. Link its obligations into your programme without calling them PIA provisions.
- Make local capability measurable: who maintains sources, explains limitations, manages permissions, investigates incidents and continues work during an outage? A training promise is incomplete until people can do those jobs. Train by role; assess through tasks, not attendance.
- UPSTREAM · THE WORKED CASE
- Upstream. Illustrative: a fictional operator’s integrity contract flows record-keeping down to the contractor. When a record is missing, the operator still answers for the obligation. Flow-down moved the work, not the accountability.
ACROSS THE CHAIN · THE PARALLEL- Service company. Illustrative: a fictional contractor bids for integrity work and its assistant drafts a persuasive response from earlier jobs. The reviewer checks that team and equipment actually match the scope. If mobilisation depends on buying equipment, cost and timing go into the bid decision — the assistant must not erase dependencies by writing in the future tense.
- A missing attachment and a missing capability call for different responses. In procurement, require exportable records and a handover your own staff demonstrate.
Have your own staff recover from a simulated outage with the vendor out of the room. That is your capability.
21:00 – 25:00
Before you close the laptop
- Write down the establishment cost, annual cost and demonstrated benefit you could defend for your PIA initiative today — and mark which is a baseline and which is a hope.
- This session produces Section 13 of your game plan — the honest capital case and the capability you will own.
- If your arithmetic comes out negative, do not adjust the inputs. Write the non-financial reason, or reconsider the initiative.
Finance will trust the second case you bring them only if the first one was honest.
Key Terms
Statutory computation
A calculation using the applicable rule, approved inputs, adjustments and evidence. Distinct from an investment sensitivity that uses labelled assumptions.
Funding sensitivity
Illustrative arithmetic showing how escalation, timing and fund balance affect a retirement estimate. Not a statutory contribution formula.
Contractual flow-down
Contract terms requiring suppliers to perform controls or provide records. They do not transfer the licence holder’s statutory accountability.
Net present value (NPV)
The present value of future net benefits at a stated discount rate, minus the initial cost. Negative NPV means the case is not financially attractive on those inputs alone.
Self-Check — Before You Submit
1. Using the book’s illustrative inputs — ₦30m to establish, ₦12m a year to operate, ₦24m a year demonstrated improvement, 15% over three years — what is the honest conclusion?
- A. NPV is roughly −₦2.601m, so it is not financially attractive on that basis alone; any other justified reason must be stated
- B. Payback of 2.5 years makes it attractive, so NPV can be ignored
- C. Add an estimate of avoided regulatory fines until NPV is positive
- D. NPV is roughly +₦42m because ₦24m a year over three years exceeds ₦30m
A. NPV is roughly −₦2.601m, so it is not financially attractive on that basis alone; any other justified reason must be stated — Net benefit is ₦12m a year, worth about ₦27.399m in present value, against ₦30m to establish. Adding invented avoided fines is the tempting repair; the book says state real reasons such as control repair instead.
2. A fictional upstream operator’s board paper shows a $20m retirement estimate escalated at 4% for ten years and a fund of $8m. A director asks whether the ≈$2.960m a year is the statutory contribution. What do you say?
- A. Yes, because it is calculated under the 2026 regulations
- B. Yes, but only for the first ten years
- C. No: it illustrates escalation and timing, and the approved funding basis governs; we should also test a shorter life
- D. No, because the 2023 regulations still apply in full
C. No: it illustrates escalation and timing, and the approved funding basis governs; we should also test a shorter life — The book states the arithmetic is not a statutory contribution formula. The trap in the last option is the instrument: regulation 25 of the 2026 regulations revokes the 2023 regulations with savings, so relying only on 2023 needs review.
3. Your operator’s contract requires a contractor to keep inspection records. One is missing at audit. Where does statutory accountability sit?
- A. With the contractor, because the contract flowed the obligation down
- B. With the licence holder, because flow-down does not automatically transfer statutory accountability
- C. With the regulator, because it approved the contractor
- D. Nowhere, until the contract is renegotiated
B. With the licence holder, because flow-down does not automatically transfer statutory accountability — Flow-down can require suppliers to perform controls or provide records, but the licence holder remains answerable. Believing the contract moved the accountability is exactly the mistake that leaves an obligation unowned.
Today’s Assignment
Write the honest capital case for your company’s PIA initiative and name the capability your business must own rather than rent.
- State the initiative and the obligation it serves, carried from your earlier sections.
- Give establishment cost, annual operating cost and the benefit you can actually demonstrate, each with its source. Label every assumption as illustrative or approved.
- Calculate net annual benefit and simple payback; add NPV on a stated discount rate and period if finance can support one. Show the arithmetic.
- If the case is weak or negative, say so plainly, then state any non-financial justification such as control repair. Do not invent avoided fines.
- If a fiscal or retirement figure is involved, separate sensitivity from statutory computation and name the current instrument a specialist must confirm.
- Name the one capability your company must own, the role that will hold it, and how you will test it through a task rather than attendance.
YOUR GAME PLAN
Section 13 — The honest capital case and the capability we will own.
How this is marked — the rubric
| Costs and demonstrated benefit stated with sources, every assumption labelled, and arithmetic shown | 30 |
| The financial conclusion stated honestly, with any non-financial justification named and no invented avoided fines | 30 |
| Sensitivity kept distinct from statutory computation, with the current instrument identified where relevant | 15 |
| One owned capability named with a role and a task-based test, distinct from contractual flow-down | 25 |
| Total · pass mark 70 | 100 |
Submitting session 13
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: 13, 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
Fiscal rule disciplineRetirement funding literacyHonest investment casesNPV and paybackOwned capabilitySupplier accountability
The Board · Appendix BSESSION 14
The Board Scorecard and the Investment Paper
Why This Matters
Your board probably receives a great deal about compliance and very little it can decide on. Counts of documents, prompts and licences show activity; they do not show that obligations are executed. This session gives you a scorecard of thirteen measures, each with a denominator, evidence and owner; the status words that stop one colour hiding several different problems; a way to run the monthly executive meeting so it resolves rather than reads; and the one-page investment recommendation that asks for an exact decision. You leave with the five to seven measures you will report and the decision you will request.
25-Minute Module
0:00 – 4:00
Activity is not execution
- A dashboard that reports two thousand documents indexed and five hundred prompts answered tells the board the system is busy. It says nothing about whether the renewal will be granted or the return accepted.
- THE DOCTRINE
The board needs a small set of measures tied to decisions. Define each measure’s denominator, evidence and owner before using it for performance review.
- Report counts beside percentages. Four of five selected obligations passing is not the same as eighty of one hundred. Show the selection basis so nobody mistakes a pilot population for the company.
Take one measure from your current board pack and ask what its denominator is. Silence is an answer.
4:00 – 11:00
The obligation execution scorecard
- Appendix B of the book sets out thirteen measures. The first seven test execution.
Measure What it reveals Applicability approved Unfinished legal mapping: priority obligations without current source and signed applicability Ownership assigned Responsibility gaps Evidence complete Readiness against approved criteria, not file volume On-time accepted completion Outcomes with required acceptance; submission kept distinct from external approval Open high-consequence exceptions Decisions requiring attention, under your approved consequence method Decision age Time since a material exception was ready for a named authority to decide Resource decisions outstanding Corrective work awaiting funds, staff or inspection - The other six test whether the system can be trusted and whether value is real.
Measure What it tests Assurance reconstruction Whether sampled completed cases can be independently rebuilt from evidence Wrong-source incidents Corpus control: decisions using obsolete, unauthenticated or out-of-scope sources Reviewer correction effort The hidden cost: specialist time correcting output for comparable work Permission control failures The boundary: failed access tests or actions beyond authority Recovery demonstrated Resilience: manual continuation or rollback actually exercised Benefit verified Outcomes with approved attribution and financial treatment, kept apart from projection - The book sets no universal green percentage or exception-age limit. Thresholds follow consequence: a low-volume critical obligation may need immediate escalation while a routine evidence correction follows an agreed queue.
Decision age is the measure executives least like and most need. It is the one that measures us.
11:00 – 16:00
Status words and the monthly meeting
- Use statuses that describe the process condition: proposed, preparing, blocked, ready for review, approved, submitted, receipt confirmed, completed. Approved means an authorised person accepted a defined version and purpose; completed means the approved closure condition is met.
- Overdue is a time condition, not an outcome. An overdue item may be preparing, blocked or awaiting an external response, and each needs a different decision. Rejected evidence is not a rejected application.
- Open the monthly executive meeting with material changes since the last review, then the decisions that need the committee’s authority. For each: obligation, current evidence, operational consequence, options, cost, recommendation and latest useful decision date.
- THE DOCTRINE
Record the decision precisely — scope, funding, owner, conditions, next review — and review previous decisions against evidence, so the meeting does not treat its own instructions as completed execution.
“Resolve the issue” is not a decision. It is a hope with a minute number.
16:00 – 21:00
The one-page investment recommendation
- Twelve sections, one page, evidence and financial model attached: decision requested, operating problem, obligation connection, alternatives, evaluation, live evidence, economics, authority, dependencies, sustainability, recommendation and conditions.
- Write the decision before the solution narrative. “Approve a defined evaluation budget for a read-only completeness assistant at one facility, subject to source authentication and access testing” is reviewable. “Approve AI transformation across the group” cannot establish scope or acceptance.
- In the decision rights matrix, advisory contribution stays distinct from authority. The model’s confidence does not enlarge anyone’s delegation.
- UPSTREAM · THE WORKED CASE
- Upstream. Illustrative: a fictional operator’s retrieval pilot shows faster source search, material condition omissions still present and reviewer hours unchanged. The right decision is narrower retrieval use with specialist review — not expansion into automated completeness approval.
ACROSS THE CHAIN · THE PARALLEL- Midstream and downstream. Illustrative: a fictional refinery asks the board to fund a completeness checker for product release evidence. The decision requested names one site, a budget and a read-only boundary; product release stays with the authorised quality role, and the conditions say what evidence must exist before authority advances.
If you cannot write the decision requested in one sentence, you are not ready to ask for it.
21:00 – 25:00
Before you close the laptop
- Write down the three measures your board would miss most if you removed them, and one it receives today that shows only activity.
- This session produces Section 14 of your game plan — the board scorecard and the decision you will ask for.
- Practice Exam 3 — Full Syllabus opens now: fifteen questions across Sessions 1 to 14. Sit it before Session 15. Three of the Final Exam’s questions will be rebuilt from your own earlier mistakes, so read every returned explanation.
A board that can see decision age will start asking why decisions wait. That is the point.
Key Terms
Denominator
The defined population a measure is counted against. Without it, a percentage cannot be interpreted or compared.
Decision age
Time since a material exception became ready for a named authority to decide. It reveals authority or capacity bottlenecks.
Latest useful decision date
The last date on which a decision can still change the outcome. Shown with every decision put to the executive committee.
Decision requested
The first line of the one-page investment recommendation: exact amount, scope, authority and timing, written before any solution narrative.
Self-Check — Before You Submit
1. A pilot report says “80% evidence complete”. What should a director ask first?
- A. Whether 80% is above the industry green threshold
- B. Which AI model produced the figure
- C. Whether the figure can be rounded up for the annual report
- D. What the denominator is — for example, four of five selected obligations — and how they were selected
D. What the denominator is — for example, four of five selected obligations — and how they were selected — Counts belong beside percentages, and four of five is not eighty of one hundred. The tempting option assumes a universal green threshold, which the book expressly declines to set.
2. An item on the register has been marked overdue for six weeks. What does that status alone tell the committee?
- A. Only that a time condition has passed; whether it is preparing, blocked or awaiting an external response must be stated
- B. That the obligation has failed
- C. That the owner has not done the work
- D. That the regulator has rejected the application
A. Only that a time condition has passed; whether it is preparing, blocked or awaiting an external response must be stated — Overdue is a time condition, not an outcome. Treating it as failure or blame skips the real question, because a blocked item and one awaiting external response need entirely different decisions.
3. A fictional distributor’s pilot gave faster retrieval, but reviewers still correct material omissions and their hours have not fallen. The vendor proposes group-wide automated completeness approval. What should the board decide?
- A. Approve expansion, because retrieval is demonstrably faster
- B. Narrow to retrieval with specialist review, address the omissions, and set a fresh evaluation gate
- C. Stop and discard the pilot entirely
- D. Approve expansion if the model reports high confidence
B. Narrow to retrieval with specialist review, address the omissions, and set a fresh evaluation gate — The book’s illustrative review keeps the demonstrated capability while addressing the weakness. Expansion ignores unchanged reviewer hours, stopping wastes useful work, and model confidence never enlarges a delegation.
Today’s Assignment
Choose five to seven scorecard measures for your company’s PIA programme and draft the decision requested and conditions lines of your one-page investment recommendation.
- Select five to seven of the thirteen measures. For each, give the denominator, the evidence source, the owner by role, and why your board needs it.
- Include at least one measure that will make your business look worse before it looks better, and say which.
- Name one measure your board receives today that shows activity rather than execution, and what replaces it.
- Write the “decision requested” line: exact amount, scope, authority and timing, in one or two sentences.
- Write the “conditions” line: the evidence required before expenditure or authority advances.
YOUR GAME PLAN
Section 14 — The board scorecard and the decision we will ask for.
How this is marked — the rubric
| Five to seven measures drawn from the scorecard, each with a stated denominator, evidence source and owner by role | 35 |
| An honest uncomfortable measure included and one current activity measure identified for replacement | 20 |
| Decision requested states exact amount, scope, authority and timing, written before any solution narrative | 25 |
| Conditions name the specific evidence required before expenditure or authority advances | 20 |
| Total · pass mark 70 | 100 |
Submitting session 14
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: 14, 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 3 — Full Syllabus 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 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Board scorecard designMeasures with denominatorsStatus disciplineExecutive meeting designInvestment recommendation
The Leadership Mandate · Chapter 17 · CapstoneSESSION 15
The Ninety-Day PIA Game Plan
Why This Matters
Fourteen sessions ago I asked you to choose the one obligation that matters most to your business and answer four plain questions about it. Today you turn everything you have written since into a plan someone can sign and hold you to. This session has three parts: how to choose the first ninety days so they prove something; the twelve leadership commitments that turn aspiration into evidence; and the capstone, your signed 90-Day PIA Game Plan. I have tried to be direct with you throughout rather than encouraging. I will keep that up for one more session, because the plan is only as good as the honesty in it.
25-Minute Module
0:00 – 4:00
Start with a decision the company needs
- The first ninety days establish controlled capability around one consequential process, not every obligation at once. The pilot should improve the process even if the AI application is rejected.
- Select by consequence, evidence readiness and owner commitment. A renewal pack, a measurement reconciliation or a recurring return may work. Avoid a process with disputed authority and inaccessible data, unless resolving those foundations is itself the authorised objective.
- UPSTREAM · THE WORKED CASE
- Upstream. An upstream operator can begin with measurement reporting and later connect communities, environment and retirement.
ACROSS THE CHAIN · THE PARALLEL- Midstream and downstream. A gas business can begin with supply constraints and emissions methodology; a refinery with permission dependencies and product evidence; a terminal with receipt-to-inventory.
- Service company. Begin with qualification-to-delivery. Common governance, different first decisions.
The most visible application is rarely the one most likely to establish dependable capability.
4:00 – 10:00
The ninety days, and the gates
Window The work The gate Days 1–30 Sponsor and owner appointed. Entity, asset, sources and interpretation mapped. Decision, evidence chain and manual process defined. Baseline recorded: time, rework, unresolved exceptions, specialist effort. Reference cases include known failures. Map, evidence index, workflow, baseline and signed charter — or fix the foundation first. Days 31–60 The smallest useful application — retrieval, comparison or completeness checking — on realistic sources. Test wrong versions, missing conditions, conflicting dates, wrong assets, denied access. Practise manual continuation. A readiness decision on evidence. A demonstration is not the gate. Days 61–90 Live work inside the accepted boundary. Record output, review, correction and outcome; measure total effort; review exceptions weekly. Scale, narrow, repair or stop — each with a reason and evidence. - Bring finance in during the first thirty days. A benefit estimated after a successful demonstration is less dependable than a baseline agreed beforehand.
- Do not weaken criteria because the team wants success.
- THE DOCTRINE
At day ninety, decide to scale, narrow, repair or stop. Each can be sensible. Continuing an unsuccessful application because money has been spent is not disciplined implementation.
Write the day-ninety criteria into the charter now, while nobody yet has a stake in the answer.
10:00 – 15:00
The twelve commitments, and the day-ninety paper
- Twelve leadership commitments: approved applicability; named ownership; controlled sources; dependable identity and time basis; documented decision rights; reproducible calculations; proportionate human review; limited permissions; visible exceptions; tested recovery; measured outcomes; sustained ownership.
- They are recommended disciplines, not twelve additional legal duties.
- THE DOCTRINE
Each commitment needs a deliverable and an owner. Approved applicability needs a signed source-and-scope record; recovery needs a completed exercise; sustained ownership needs a team and a budget. Converting aspiration into evidence stops the programme relying on slogans.
- In the pilot charter, write the problem without naming a product, and define acceptance as an operating right: successful retrieval might permit use by trained reviewers at one facility, not general approval for regulatory automation.
- The day-ninety paper shows problem, baseline, scope, evaluated cases, live outcomes, failures, full cost and limitations, and compares expansion with a simpler process repair. Expansion must not silently increase autonomy.
A commitment without a deliverable is a sentiment.
15:00 – 20:00
What you have built
- Your game plan already exists in fourteen pieces.
Sections What they establish 1–2 The obligation, the four questions, the trace and its first fix 3–5 Obligation record, decision rights, evidence index and data gaps 6 The AI role: layers, rules versus model, and the approval boundary 7–10 Permissions, quantities, stakeholder or environmental evidence, the critical handoff 11–12 Governance record, submission package, agent boundary 13–14 Capital case, owned capability, scorecard and the decision you will ask for - A plan that has learned nothing in three weeks would worry me more than one with visible corrections.
Read the assembled plan as your most sceptical board member, and fix whatever would not survive the one question that matters: how do you know.
20:00 – 25:00
The capstone, the Final Exam, and a last word
- The capstone is Section 15: Sections 1–14 assembled into one signed 90-Day PIA Game Plan with a one-page executive summary.
- The Final Exam opens in this session: fifteen questions across all fifteen sessions, three of them rebuilt from your own earlier mistakes, closed book. The pass mark is 70. The certificate of completion requires both the capstone and the Final Exam at 70 or above.
- AI can help people see, compare and prepare. It cannot carry their responsibility.
- THE DOCTRINE
Fund the necessary controls. Assign responsibility. Make decisions while they can still change results. Require evidence before closure. Build capability deliberately, one consequential process at a time.
- My hope for you is practical. Take one process that matters, make its evidence dependable and bring decisions into the open. Improve it, measure the outcome, and carry the discipline into the next process.
Let the quality of the work establish the value of the technology.
Key Terms
90-Day PIA Game Plan
The signed document assembled from Sections 1–14: one obligation, its owners, evidence, AI role, controls, capital case, board measures and a ninety-day pilot with gates.
Readiness gate
The day-sixty decision, made on evidence from realistic tests, whether the application is ready for shadow use, limited advice or further repair. A demonstration does not pass it.
Day-ninety decision
The choice to scale, narrow, repair or stop, each with a reason and evidence. Money already spent is not a reason.
Leadership commitments
Twelve recommended disciplines, from approved applicability to sustained ownership, each converted into a deliverable and an owner. Not additional legal duties.
Self-Check — Before You Submit
1. On day fifty-five, a vendor gives an impressive live demonstration to the executive committee. The sponsor suggests treating it as the day-sixty readiness decision. What do you advise?
- A. Accept it, because the committee has seen the tool work
- B. Accept it, provided the vendor signs a performance warranty
- C. Decline: readiness is decided on evidence from realistic tests, including the failure cases
- D. Skip the gate and move straight to the day-ninety decision
C. Decline: readiness is decided on evidence from realistic tests, including the failure cases — A demonstration is not the readiness gate; the book requires a decision based on evidence from realistic testing. A warranty is a commercial remedy, not proof the application handles the failure cases.
2. At day ninety, the pilot has cost ₦18 million, reviewers still correct material errors, and the team argues that stopping would waste the investment. Which outcomes may the committee properly choose?
- A. Only scale, because stopping would waste the investment
- B. Scale, narrow, repair or stop, decided on evidence; sunk cost is not a reason to continue
- C. Only stop, because errors remain
- D. Defer the decision until the money has been recovered
B. Scale, narrow, repair or stop, decided on evidence; sunk cost is not a reason to continue — Each of the four outcomes can be sensible, and narrowing or repairing may preserve real value. Continuing because money has been spent is exactly what the book calls undisciplined implementation.
3. Which of these is a properly formed leadership commitment for your game plan?
- A. “Tested recovery” — owner: plant manager; deliverable: a completed manual-continuation exercise by day sixty
- B. “We are committed to world-class compliance”
- C. “Tested recovery” — a new legal duty under the PIA
- D. “Measured outcomes” — to be defined after the pilot succeeds
A. “Tested recovery” — owner: plant manager; deliverable: a completed manual-continuation exercise by day sixty — Each commitment needs a deliverable and an owner, and recovery needs a completed exercise. The commitments are disciplines rather than legal duties, and outcomes defined after success are retrospective claims.
Today’s Assignment
Assemble Sections 1–14 into one signed 90-Day PIA Game Plan for your company, with a one-page executive summary — this capstone will run well beyond the usual 900 words, and that is expected.
- Assemble all fourteen sections in order and make them consistent; where a later session changed your mind, correct the earlier section and note it.
- Name the first pilot as a business problem, without naming a product, and justify it on consequence, evidence readiness and owner commitment. State its acceptance decision as an operating right.
- State the baseline — time, rework, unresolved exceptions and specialist effort — with its source and period. If it has not been measured, make it a days 1–30 deliverable with an owner and a date.
- Name the sponsor and the process owner by role and name, and give the start date as an actual date.
- Set the day-30, day-60 and day-90 gates with the evidence each requires, and write the scale, narrow, repair or stop criteria now.
- List the twelve commitments, each with a concrete deliverable and an owner, and name the one your business is least likely to keep.
- Write the one-page executive summary — decision requested, obligation, pilot, baseline, sponsor, owner, start date, gates — then sign and date the plan.
YOUR GAME PLAN
Section 15 — The assembled, signed 90-Day PIA Game Plan with its executive summary.
How this is marked — the rubric
| All fourteen sections present, in order and internally consistent, with no placeholders and corrections acknowledged | 20 |
| A named first pilot stated as a business problem, justified on consequence, evidence readiness and owner commitment | 20 |
| A real baseline with source and period, or a chartered baseline task with owner and date, rather than a target or benchmark | 20 |
| Sponsor and owner named by role and name, an actual start date, and day-30, day-60 and day-90 gates with evidence criteria set in advance | 20 |
| All twelve commitments each carrying a deliverable and a named owner, and a signed one-page executive summary | 20 |
| Total · pass mark 70 | 100 |
Submitting session 15
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: 15, 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.FINAL EXAM
Your Final Exam is sat in this session. 15 questions, of which 3 are rebuilt from questions you got wrong on the practice papers. Sit it once your capstone is submitted.
Sit the Final Exam →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Pilot selectionNinety-day gatingCommitments into evidenceDay-ninety decisionGame plan assemblyBoard-ready summary