THE INTELLIGENT REFINERY LEADERSHIP PROGRAMME
AI earns its place in a refinery one decision at a time.
20-Session Programme · 25 Minutes Per Day · Free · For leaders of Nigeria’s modular and mega refineries
WEEK 4 OF 4 · ENTERPRISE EXECUTION, GOVERNANCE AND THE NINETY DAYS · SESSIONS 16–20 · WORK AT YOUR OWN PACE
Enterprise · Chapters 17 and 18SESSION 16
Critical Spares, Operating Data and Reconciliation
Why This Matters
Week 4 turns from individual operating chains to the enterprise machinery every one of them depends on. Two pieces of it fail quietly. A critical spare looks adequate on the inventory list until the day it is needed and the specialist, the lifting arrangement or the funded order is missing. A management number looks agreed until operations and finance bring different versions of it to the same review. This session gives you two disciplines: buy a complete recovery capability rather than a part, and reconcile records before anyone explains them. You will finish with one spare decision and the data contract behind your first use case.
25-Minute Module
0:00 – 5:00
A spare is worth the response it supports
- A critical spare has value because it supports a defined operating response; purchase price alone cannot establish its importance. Link each candidate to the installed asset, the approved specification and its replacement purpose, with compatibility and preservation requirements. A description that sounds similar to an installed part is not evidence for substitution.
- Ask maintenance for the consequence of absence. Can the refinery continue safely in a restricted mode? Is another installed unit available? Does recovery need a shutdown, specialist labour or a lifting arrangement?
- THE DOCTRINE
Separate planned consumption from contingency stock. Contingency stock needs a documented recovery rationale and periodic review, and no model should call a rarely used protective spare surplus solely because historical consumption is low.
Ask when a critical spare was last found in the store but could not be fitted in time.
5:00 – 11:00
The complete lead time, and an assumed stocking decision
- Record the complete route: quotation approval, payment, production, transport, documentation, customs where applicable, receipt, inspection and release for use. Treasury confirms currency and the payment date. Never substitute a generic international lead time for your actual import route, and never present a predicted arrival as a confirmed shipment.
Assumed input (teaching example only) Value Cost of the spare $25,000 Annual included holding and preservation cost $3,000 Assumed probability of a relevant failure in the year 20 percent Assumed incremental consequence without the spare Two days of lost contribution at $50,000 per day Simple expected avoided contribution loss 0.20 × 2 × $50,000 = $20,000 - Every input is assumed, and the book is clear that $20,000 is not sufficient to approve stock. It excludes cash timing, remaining value, alternative recovery routes and uncertainty in the failure estimate, and it assumes the spare actually removes the two days of delay.
- Test that assumption. If another available part restores the equipment within the same response time, the value of extra stock may be much lower. If specialist labour still needs five days to arrive, the spare may not remove the delay.
- THE DOCTRINE
Management should buy a complete recovery capability, with its dependencies understood. A part in the store is one dependency, not the recovery.
When the expected loss looks persuasive, ask who fits the part, and when.
11:00 – 16:00
Commercial authority, identity and time
- A purchasing assistant may prepare a requisition or compare approved quotations; financial commitment stays within the delegation of authority. Bank details, specifications and delivery instructions are controlled events. Text generated from an email chain must not silently overwrite a supplier master record or an engineering specification.
- Now the data. A refinery can have several accurate systems and still lack an agreed operating account. Management needs the relationships between them, not a claim that one database holds the whole truth.
- Choose identifiers that survive handovers: a tank, asset, batch or movement keeps a controlled identity across systems through an approved mapping, with its effective period recorded when configuration changes.
- Define the timestamp's meaning, time zone, reporting period and treatment of late entries. The time a result is entered may differ from the time the sample was collected; the time an invoice is posted may differ from the time the product moved. Never carry yesterday's usable inventory forward as today's confirmed stock without making that visible.
Freshness follows the decision. A monthly trend and an immediate dispatch choice need different evidence.
16:00 – 21:00
Reconcile before explaining, then write the contract
- When two reports disagree, examine scope, units, time and status before looking for a sophisticated cause. A mass quantity and a volume quantity need a common basis; a preliminary laboratory result is not the approved release record.
- THE DOCTRINE
Use a discrepancy register with an owner and a deadline, and preserve unresolved differences in the executive review instead of forcing a balanced number.
- Classify each discrepancy enough to route it: missing movement, identifier mismatch, timing difference, measurement concern or unresolved cause. Repeated differences at shift change justify investigating posting practices; that is a hypothesis, not evidence of a person's error.
- For the first use case, write a data contract in ordinary language: required fields, accepted sources, update frequency, quality checks, correction route, owner, and what the application does when an important condition is not met.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. The book's assumed case fits closely. A scheduling assistant needs current usable tank capacity, released-product status and a confirmed loading window. If capacity is stale, it withholds a definitive loading recommendation and may prepare an exception record for review.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. The same contract crosses historian tags, laboratory sample identities, maintenance asset codes and commercial movement references. The approved mapping and its effective dates become the hardest clause to keep current.
- Do not assign every data problem to IT. The business function defines meaning, IT supports integration and access, measurement specialists assess instruments and finance defines the financial boundary. Budget for that ownership after the project.
Withholding can be the correct output when the evidence does not support a choice.
21:00 – 25:00
Before you close the laptop
- Write down the spare whose absence would most damage your Session 1 problem, the step in its supply route nobody has confirmed, and the one management number operations and finance report differently.
- This session produces Section 16 of your 90-Day Intelligent Refinery Plan: one critical spare decision and the data contract behind your first use case.
- Measure on-time availability for approved work and the actual recovery consequence, not purchase-order count. For data, measure the time to produce an agreed account and how often discrepancies recur.
Reliable evidence should make the operating discussion shorter and more decisive.
Key Terms
Contingency stock
A spare held against unplanned failure rather than dated planned work. It needs a documented recovery rationale and periodic review, not a consumption history.
Complete lead time
The whole route from quotation approval and payment through production, transport, customs, receipt and inspection to release for use, with the funded commitment confirmed.
Discrepancy register
The owned, dated list of differences between records, each classified as missing movement, identifier mismatch, timing difference, measurement concern or unresolved cause.
Data contract
A plain-language agreement for one use case: fields, sources, update frequency, quality checks, correction route, owner, and what the application does when a condition fails.
Self-Check — Before You Submit
1. An analytics tool flags a protective spare for a critical pump as surplus because it has not been issued for several years. What should the materials manager do?
- A. Dispose of it, since years without issue prove it is surplus
- B. Keep it only if its price is above the disposal threshold
- C. Ask engineering and maintenance to confirm its recovery purpose and keep a documented contingency rationale
- D. Accept the tool's view, since consumption history is objective
C. Ask engineering and maintenance to confirm its recovery purpose and keep a documented contingency rationale — Contingency stock is justified by the recovery it supports, not by consumption. Price and consumption history are tempting only because they are easy to measure.
2. In the book's assumed stocking example, what does the $20,000 represent?
- A. A simple expected avoided contribution loss, 0.20 × 2 × $50,000, not sufficient on its own to approve stock
- B. The net saving after deducting the $25,000 cost and $3,000 holding cost
- C. The confirmed annual value of holding the spare
- D. The maximum price the refinery should pay for the part
A. A simple expected avoided contribution loss, 0.20 × 2 × $50,000, not sufficient on its own to approve stock — It is a simple expected value on assumed inputs. It excludes cash timing, remaining value, alternative routes and uncertainty, and the book does not net it against cost.
3. At month end, operations and finance report different stock figures for the same product tank. An analyst proposes a model to find the cause. What comes first?
- A. Build the model, since the pattern may be subtle
- B. Average the two figures and report the result as agreed stock
- C. Adopt finance's figure, because it is the reporting record
- D. Check scope, units, time and status, and log the difference in the discrepancy register with an owner and a deadline
D. Check scope, units, time and status, and log the difference in the discrepancy register with an owner and a deadline — Reconcile before explaining: most disagreements are scope, unit, time or status differences. Averaging forces a balanced number, and choosing one function's figure hides the difference rather than resolving it.
Today’s Assignment
Write Section 16 of your plan: one critical spare decision for your company, and the data contract behind your first use case.
- Choose one spare tied to the constraint in your Session 1 problem. Confirm with engineering its identity and recovery purpose, and with maintenance the complete response, including labour, lifting and any shutdown.
- Lay out its supply route to release for use, marking each step confirmed or assumed, including the funded commitment.
- Choose one disputed management number in your business. Trace its records across functions, explain the differences in one reconciliation, and name who can approve corrections.
- Write the data contract for your first use case, including what the application does when an important condition fails. Say honestly where your company forces a balanced number or carries stale stock forward.
- If you do not work at a refinery, apply this to the spare or number your work touches most directly.
YOUR 90-DAY PLAN
Section 16 — One critical spare decision and the data contract behind our first use case.
How this is marked — the rubric
| The spare is linked to its installed asset, specification and recovery purpose, with the complete response and dependencies named | 30 |
| The supply route is laid out step by step, with the funded commitment and each step marked confirmed or assumed | 20 |
| The disputed number is reconciled by scope, units, time and status, with each remaining discrepancy owned and dated | 25 |
| The data contract is complete, including what is withheld or routed as an exception when a condition fails | 25 |
| Total · pass mark 70 | 100 |
Submitting session 16
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: 16, 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
Critical-spare judgementComplete lead-time modellingCommercial authority controlReconciliation disciplineData contracts
Enterprise · Chapters 19 and 20SESSION 17
Cybersecurity, Controlled Access and Workforce Adoption
Why This Matters
An AI application in a refinery creates paths that did not exist before: copied process data, a connection between enterprise and operating environments, a supplier who can log in, an account with new privileges. It also depends on people who must recognise its output, judge it and carry the response through. This session puts both sides on one page. You will define what your application can reach and how it is stopped, and what each person using it must be able to do, including reject it. A refinery should grant only the authority it can explain, monitor and withdraw.
25-Minute Module
0:00 – 6:00
Start with what the application can reach
- Record the systems, data, accounts and network paths the use case requires, with read access separated from write access. Establish whether the application can only prepare a report or can initiate a transaction or an operating command; the consequences of compromise differ.
- Ask the security and operations teams to review the actual design together. A convenient enterprise integration should not decide the connection to OT by itself. Establish controlled interfaces, approved data flows and the people responsible for maintaining them.
- THE DOCTRINE
Permission should follow the defined purpose rather than the supplier's maximum available capability. An advisory trial may need only an approved copy of selected data, with no new path for commands.
- The book cites NIST SP 800-82 Revision 3, the Guide to Operational Technology Security, in its September 2023 final publication. It is U.S. guidance used as an international reference. It does not prescribe your Nigerian regulatory obligations or approve a particular architecture, and a Revision 4 draft is a separate document.
Ask the supplier for the list of accounts the product creates. Then ask which of them can write.
6:00 – 12:00
Supplier support, untrusted content and recovery
- For supplier support, specify who may connect, for what purpose, through which approved route and for how long, using the refinery's established identity, authentication and logging. Remove access when the task ends, and make remote support visible to the operating team where it can affect their systems.
- Record where copied data is held and whether sensitive operating, commercial or personal information is sent to an external service. The Nigeria Data Protection Act 2023 is relevant where personal data is processed; your legal and privacy specialists determine the obligations for the actual processing.
- A generative assistant can meet instructions embedded in documents, emails or retrieved pages. An instruction in a supplier document is not authority to disclose records, change a payment instruction or create an account. Test it with a misleading instruction planted in a test document.
- Plan failure. Know how the service is stopped and what work continues during an outage; the refinery should not lose its operating review because a nonessential supplier service is down. Test recovery in a controlled environment and keep the evidence.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. One person may hold several support roles. The book is explicit that a small installation needs the same clarity. Write down who connects, by what route, for how long, and who can withdraw that access on the night shift.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. The risk is many paths maintained by different teams. The access schedule has to be reviewed jointly by operations, security and the business owner, or the enterprise integration quietly sets the OT boundary.
Human review belongs at the consequential step, with the proposed action visible. A review after the action is an audit, not a control.
12:00 – 18:00
Competence by role, and the difficult cases
- An application becomes useful through the people who recognise its output, judge it and carry the response through. A general demonstration of AI features does not establish competence to use a refinery recommendation.
Role What competence means for this application Supervisor Recognise stale inputs and reject an unsupported recommendation Planner Compare the application's proposal with existing constraints Data owner Correct mappings and preserve the reason Technical custodian Restore the service - THE DOCTRINE
Distinguish user competence from approval authority. Completing a training course does not expand a person's delegation.
- Train on the live decision record: a normal recommendation, a missing-data case, a contradictory-record case and one the user should reject. In the book's assumed teaching scenario, a scheduling assistant recommends loading a tank whose release status is preliminary. The competent user identifies the status, consults the controlled record and holds the commercial commitment until the required release evidence is available.
- Include service failure. Participants should be able to continue the established workflow, retrieve the evidence they need and escalate support. Speed of acceptance is not the measure of good use.
The best training case is one your own team handled badly last quarter. Use it, with names removed.
18:00 – 22:00
Safe feedback, and the work that changed
- Give users a simple route to report an incorrect or unhelpful recommendation, recording input, output, consequence and actual response. Avoid judging users solely on whether they followed the model. A justified rejection can reveal a missing constraint; calling all rejection resistance to change conceals the operating problem.
Expressly assumed pilot Count Recommendations generated 40 Accepted 20 Rejected 12 Left unresolved 8 Accepted recommendations that received action 10 - THE DOCTRINE
Attendance shows participation. Usage shows interaction. Neither establishes that the application improved the decision.
- In that assumed case the main limitation may be response capacity rather than model accuracy. Investigate the gap between 20 accepted and 10 acted on before expanding training or buying another application. Do not use the counts to compare individuals: a team dealing with more abnormal situations may reasonably reject more.
Ask users what prevents them from acting, then give each of those constraints an owner.
22:00 – 25:00
Before you close the laptop
- Write down two things: the one path your application would create that nobody has yet reviewed, and the person who could revoke supplier access tomorrow morning.
- This session produces Section 17 of your 90-Day Intelligent Refinery Plan: the access boundary, recovery route and role competence for your application.
- Treat pausing the service, revoking supplier access and a working fallback as part of acceptance, not as housekeeping for later. Review capability, staffing and support alongside the pilot result.
An intelligent refinery keeps human judgement visible and provides the organisation required to use it well.
Key Terms
Access schedule
The list of every access an application needs, each entry with its purpose, owner, scope, expiry where appropriate and the consequence of misuse, reviewed with operations, security and the business owner.
Untrusted input
Content an assistant reads, such as a supplier document or email, which may contain embedded instructions. It is material to assess, never authority to act.
Fallback process
The controlled procedure and records that let the refinery keep working, including preparing its operating review, while the application is stopped or unavailable.
Role competence
What a specific user must be able to do with the application, such as rejecting an unsupported recommendation. It is separate from, and never expands, approval authority.
Self-Check — Before You Submit
1. A procurement assistant retrieves a supplier document containing a line instructing that payments be sent to a new bank account, and drafts an updated payment instruction. What should happen?
- A. Accept the draft, because the instruction came from the supplier's own document
- B. Treat the line as content to assess; a bank-detail change goes through the existing verification and approval arrangements
- C. Let the assistant update the supplier master record but notify finance afterwards
- D. Disable the assistant permanently and return to manual procurement
B. Treat the line as content to assess; a bank-detail change goes through the existing verification and approval arrangements — An embedded instruction is not authority, and bank-detail changes are controlled events. Notifying finance afterwards puts review after the consequential step; scrapping the assistant ignores that the fix is a restricted permission.
2. How does the book position NIST SP 800-82 Revision 3?
- A. As the cybersecurity standard Nigerian refineries are legally required to meet
- B. As an approval of the architecture a refinery chooses to follow it
- C. As replaced by the Revision 4 draft, which should be cited instead
- D. As U.S. guidance used as an international reference that does not prescribe Nigerian obligations
D. As U.S. guidance used as an international reference that does not prescribe Nigerian obligations — The book uses the September 2023 final publication as an international reference that recognises OT's performance, reliability and safety needs. It neither sets Nigerian obligations nor approves an architecture, and it treats the Revision 4 draft as a separate document.
3. In the book's assumed pilot, 40 recommendations produced 20 acceptances, 12 rejections and 8 unresolved, and only 10 accepted recommendations were acted on. What should leadership investigate first?
- A. Why accepted recommendations were not acted on, since response capacity may be the main limitation
- B. How to retrain the model, since accuracy must be the problem
- C. Which individuals rejected the most, so they can be retrained
- D. Whether a second application would produce more recommendations
A. Why accepted recommendations were not acted on, since response capacity may be the main limitation — Half the accepted recommendations received no action, which points to response capacity rather than accuracy. Ranking individuals ignores task difficulty, and more output only adds review workload.
Today’s Assignment
Write Section 17 of your plan: the access boundary, recovery route and role competence for your company's first AI application.
- Write the access schedule: every system, data source, account and network path the application needs, each with purpose, owner, scope, read or write, expiry where appropriate and the consequence of misuse.
- Describe supplier support: who may connect, by which approved route, for how long, how it is made visible to operations, and who revokes it. State whether personal data is processed and who will determine the obligations.
- State how the service is paused, how supplier access is revoked and how work continues without it, and how you will test that before acceptance.
- Write a competence requirement for each role in the use case, including a justified rejection and a fallback case. Say honestly where your company currently confuses training with authority.
- Ask two users what prevents them from acting on a recommendation, and assign an owner to each constraint. If you do not work at a refinery, apply this to the application your work touches most directly.
YOUR 90-DAY PLAN
Section 17 — The access boundary, recovery route and role competence for our application.
How this is marked — the rubric
| Access schedule complete, with read and write separated and each entry carrying purpose, owner, scope and consequence of misuse | 30 |
| Supplier access, pause, revocation and fallback stated concretely, with a test before acceptance | 25 |
| Competence written by role, distinguished from approval authority, and tested through rejection and fallback cases | 25 |
| User-reported constraints captured, each with a named owner | 20 |
| Total · pass mark 70 | 100 |
Submitting session 17
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: 17, 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
Access-path reviewSupplier access controlEmbedded-instruction testingRecovery and fallbackRole-based competence
Governance · Chapter 21SESSION 18
Investment Cases and Value Attribution
Why This Matters
Many AI investment cases explain a capability in the software rather than a change in the business. They count the same benefit two or three times, price the project as a licence, and mix dollar costs with naira cash flows without saying how. Finance then approves a number nobody can reproduce, and the programme loses credibility the first time someone asks for the ledger. This session gives you the honest version: the decision that will improve, the response the refinery can execute, the financial consequence, the counterfactual, the cash timing, and a benefit ledger in which each contribution is counted once. It ends with Pop Quiz 4.
25-Minute Module
0:00 – 5:00
A change in the business, on a boundary finance can reproduce
- State the decision that will improve, the response the refinery can execute and the financial consequence of that response. Include the alternative of repairing the existing workflow without AI.
- Finance needs a boundary it can reproduce: capital cost, recurring service cost, integration, staffing, training, technical review and the resources required to act on recommendations. A project priced as a licence may need considerable operating effort before it delivers a result.
- THE DOCTRINE
Higher saleable output, lower included operating cost, released working capital and earlier cash collection are different mechanisms. Record each with its own baseline and evidence, and count the same outcome once.
- Two traps recur. If a reliability intervention avoids downtime, the recovered output may already be inside the avoided contribution loss, so adding its revenue double counts. If inventory falls, the release is a working-capital change, not recurring annual profit.
- Give finance an explicit verification role. The sponsor explains the expected mechanism; finance assesses the boundary and the reported result; technical evidence establishes that the physical change occurred. A convincing narrative from either function alone is insufficient.
Ask the sponsor to name each mechanism in one line. If two lines describe the same barrel, one of them goes.
5:00 – 12:00
A transparent investment example
- The book's example is assumed solely for teaching. The annual amounts are assumed to include the project's recurring costs, and the results are before any excluded tax, residual-value or other effects.
Assumed item $60,000 case $40,000 case Initial investment $120,000 $120,000 Annual net cash benefit, three years $60,000 $40,000 Assumed discount rate 15 percent 15 percent Present value of the benefits approximately $136,994 approximately $91,329 Net present value approximately $16,994 approximately negative $28,671 - The same installation cost produces a different investment decision because the benefit assumption changed. Show that sensitivity before approval rather than discovering it after implementation.
- Simple payback in the $60,000 case is two years, assuming evenly realised annual net benefit. Payback omits the time value of money and outcomes after repayment. Use it alongside the cash-flow view and the uncertainty, never as a replacement for them.
If the case only works at the optimistic benefit, say so on the first page. Finance will find it anyway.
12:00 – 18:00
Counterfactual, currency and cash
- Ask what would have happened without the intervention. A plant-wide improvement may coincide with maintenance, a different campaign or a change in prices. Compare the pilot with a defensible baseline or suitable comparator, and explain the limits of that comparison.
- THE DOCTRINE
Where attribution cannot be cleanly separated, report a range or the amount that remains unverified. A bounded conclusion management can defend is more useful than a large number resting on unstated assumptions.
- Keep the evidence trail for rejected recommendations as well as accepted ones. False alerts, review time and unsuccessful actions belong in the result.
- State the currency and payment basis of every item. Identify which costs need foreign currency and when payment falls due. Use an approved planning assumption, test a reasonable downside supplied by finance, and never mix naira cash flows and dollar costs without a consistent conversion basis.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. Affordability often decides before attractiveness does. A project with a positive estimated net present value may still need cash the refinery cannot provide at the necessary time. Stage expenditure where a stage is a genuine learning gate, and include the cost of staging.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. The larger exposure is overlap: several projects claiming the same recovered output or the same released inventory. Finance has to check overlaps across entries and across projects, not only within one case.
Affordability and economic attractiveness are related but separate decisions. Put both on the page.
18:00 – 22:00
The benefit ledger
- Appendix C of the book sets out the ledger: one entry for each distinct benefit mechanism, with finance checking overlaps across entries and projects, and four amounts kept separate.
Amount What it means Expected What the mechanism should produce, on its stated baseline, counterfactual and quantity, price, currency and timing assumptions Observed The operating difference actually recorded, with the action completed and its supporting record Attributed The amount or range credited to the intervention, with the method and the concurrent changes affecting it Financially verified The result finance has confirmed, with the review date and the uncertainty that remains - THE DOCTRINE
Do not report an expected amount as realised simply because the application is in use.
- A measured operating improvement may remain financially unverified until the relevant movement, invoice, expense or cash record is available. That is not a failure. It is the ledger telling the truth about timing.
Put every benefit into the ledger before pilot approval, not after the first good week.
22:00 – 25:00
Before you close the laptop
- Write down the benefit mechanism your Session 1 problem would produce, its baseline, and the one cost you have not yet counted.
- This session produces Section 18 of your 90-Day Intelligent Refinery Plan: the honest investment case and benefit ledger for your first application.
- At the investment gate, approve the evidence and authority needed for the trial. At the scaling gate, approve only the result that has been established and the additional assumptions required for expansion. The programme should make capital allocation clearer, including when the right choice is to stop.
- Pop Quiz 4 — Execution and Value opens now: five questions on Sessions 16 to 18, closed book. Sit it before Session 19.
A case that cannot show its downside has not been finished.
Key Terms
Benefit mechanism
The specific route by which value arrives: higher saleable output, lower included operating cost, released working capital or earlier cash collection. Each needs its own baseline and evidence.
Counterfactual
What would have happened without the intervention, established through a defensible baseline or comparator whose limitations are stated.
Net present value
The present value of a project's net cash benefits at a stated discount rate, less the initial investment. It is sensitive to the benefit assumption and silent on affordability.
Financially verified benefit
A benefit finance has confirmed against the movement, invoice, expense or cash record, as distinct from one that is merely expected, observed or attributed.
Self-Check — Before You Submit
1. In the book's assumed investment example, with $120,000 invested, three annual net benefits of $40,000 and a 15 percent discount rate, what is the approximate net present value?
- A. Approximately $16,994
- B. Zero, because $120,000 is recovered in three years
- C. Approximately negative $28,671
- D. Approximately $91,329
C. Approximately negative $28,671 — The present value of the benefits is approximately $91,329, so the net present value is approximately negative $28,671. Zero is the tempting answer because the undiscounted sum equals the investment; $16,994 is the $60,000 case and $91,329 is the present value, not the net figure.
2. A reliability project's case claims the avoided contribution loss from downtime and then adds the revenue of the recovered output as a second benefit. What should finance do?
- A. Remove the added revenue, because the recovered output may already be inside the avoided contribution loss
- B. Accept both, since they are reported by different functions
- C. Halve each figure as a prudent compromise
- D. Accept both, but report the total as a range
A. Remove the added revenue, because the recovered output may already be inside the avoided contribution loss — The same outcome is being counted twice under two descriptions. A range or an arbitrary haircut still rests on the double count; the ledger needs one entry per distinct mechanism, with overlaps checked.
3. A proposed application shows a positive estimated net present value, but its main licence fee is payable in dollars at a date when treasury expects no foreign currency to be available. What is the right executive response?
- A. Approve it, because a positive net present value settles the decision
- B. Convert the fee to naira at today's rate and treat the problem as solved
- C. Reject it, because any foreign-currency cost is unacceptable
- D. Treat affordability as a separate decision: restate the cash timing and a finance downside, and stage expenditure if a stage is a genuine learning gate
D. Treat affordability as a separate decision: restate the cash timing and a finance downside, and stage expenditure if a stage is a genuine learning gate — Economic attractiveness and affordability are related but separate. A positive value does not create cash at the date it is needed, and an outright rejection ignores that staging or a revised payment basis may make the case workable.
Today’s Assignment
Write Section 18 of your plan: a one-page investment account and benefit ledger for your company's first AI application.
- State the decision, baseline, intervention, response resources, cash timing and downside on one page, and include the alternative of repairing the existing workflow without AI.
- List the full cost boundary: capital, recurring service, integration, staffing, training, technical review and the resources needed to act. Name the currency and payment date of each item.
- Run the investment arithmetic at two benefit assumptions, labelling every input as assumed and stating the discount rate finance gave you.
- Write one ledger entry per benefit mechanism, with baseline, counterfactual, and the expected, observed, attributed and financially verified amounts kept separate. Show where entries could overlap.
- Say honestly where your company's past cases have double counted or reported expected benefits as realised. If you do not work at a refinery, write the case for the refinery decision your business touches most directly.
YOUR 90-DAY PLAN
Section 18 — The honest investment case and benefit ledger for our first application.
How this is marked — the rubric
| The case explains a change in the business, including the repair-without-AI alternative and a complete cost boundary | 25 |
| Arithmetic shown at two benefit assumptions, inputs labelled as assumed, with currency and cash timing for each item | 25 |
| Each benefit mechanism has its own ledger entry and baseline, with overlaps identified and a contribution counted once | 30 |
| A counterfactual and a finance-supplied downside are stated, with unverified amounts reported as such | 20 |
| Total · pass mark 70 | 100 |
Submitting session 18
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: 18, 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 4 — Execution and Value 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 4 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Benefit-mechanism separationInvestment sensitivityCounterfactual thinkingCurrency and cash timingBenefit ledger discipline
Governance · Chapters 22 and 23SESSION 19
AI Governance, Pilots and Controlled Scaling
Why This Matters
A successful trial attracts new requests, and new requests are where AI programmes quietly lose control. Someone asks the assistant to interpret another product specification; someone connects another data source; a supplier update moves the model beyond the conditions the refinery accepted. Meanwhile a pilot whose acceptance criteria are written after the results is functioning as a sales plan. This session joins governance and piloting: the use-case register that keeps authority explicit, and the pilot charter that sets acceptance and stopping conditions before anyone sees a result. It closes with Practice Exam 4, the full-syllabus paper.
25-Minute Module
0:00 – 6:00
Keep a use-case register
- Governance starts with the decision an application may influence. A responsible-AI policy will not resolve who can release product, approve expenditure or change an operating condition.
Register field What a usable entry says Purpose and business owner The decision supported and its owner Technical custodian and users Who keeps it working, who uses it Evidence sources and permitted output What it reads from and what it may produce Read, propose, write What the application can see, recommend and change Approval and stopping Approval before each consequential action; how the service is stopped Exclusions In operational language: can draft a requisition, cannot commit funds - Classify consequence in terms the operating team understands. A draft meeting summary and a recommendation affecting product release carry different exposure. Avoid reducing the judgement to a numerical risk score that conceals important conditions.
- NIST's AI Risk Management Framework 1.0 organises its core around Govern, Map, Measure and Manage. The book uses it as a voluntary reference for structured risk work, not a refinery standard or a Nigerian legal requirement.
If an entry cannot say how the service is stopped, the entry is not finished.
6:00 – 11:00
Generative output, changes in use and incidents
- NIST's Generative Artificial Intelligence Profile discusses confabulation among the risks of generative AI: a fluent statement can be unsupported or incorrect. Use an approved source collection, show the evidence behind each answer, and test questions the collection cannot answer or where revisions conflict.
- THE DOCTRINE
Producing a draft is different from sending it to a customer, issuing a certificate or updating a controlled record. External actions need separate approval, enforced by the system as well as described in training.
- Govern changes in use. Record the versions that matter to the result: model, prompt or configuration, source collection, integration and data mapping, and which changes require renewed testing or technical review. Reuse of the same software is not approval for a new use.
- Review actual behaviour: a report tool can become an informal assurance tool when people stop checking the underlying record. For incidents, name someone to contain the issue and someone to decide whether service resumes, and let each one change a control, test, permission or practice.
Counting incidents without closing the response repeats the incomplete loop this programme has challenged from Session 2.
11:00 – 17:00
Bound the trial and agree acceptance before results
- A pilot is an investment in evidence. Write the uncertainty first: does the data support a dependable recommendation, can the team respond in time, can the effect be measured, is the support arrangement sustainable? The design must be capable of giving an unfavourable answer.
- Bound it to one decision, asset family, product or campaign, with permitted users, access and actions. A shadow-mode trial compares recommendations with actual decisions without execution authority, but users who see the output may change their behaviour. Set duration by the opportunities needed, and never extend it merely to find a favourable result.
- THE DOCTRINE
Set acceptance criteria before results, covering evidence, useful output, response, business consequence and control. A prediction metric supports technical assessment; it is not the whole acceptance decision.
- Include failure cases in a controlled setting: missing inputs, contradictory evidence, unfamiliar operating conditions, incorrect recommendations and service outage.
Assumed teaching pilot Count Recommendations created 50 Assessed as technically useful 30 Able to be acted on, given available response resources 15 - Explain that limit before proposing expansion. Increasing the number of recommendations may add review workload without improving the operating result.
Write the stopping condition first. It is the line every sponsor is tempted to leave until later.
17:00 – 22:00
Keep a complete result, and reassess each expansion
- Record accepted, rejected and unresolved outputs with their reasons, and include review effort, unsuccessful actions and support costs. Preserve the baseline and any change in operating conditions. Finance verifies the benefit mechanism; the technical owner confirms the physical evidence.
- THE DOCTRINE
An application may pass the information test and fail the economics test. That is a useful outcome if cost and authority stayed bounded. Failure to scale is not failure to learn.
- Use staged gates. Approval to connect data is not approval to rely on recommendations. Approval for advisory use is not approval for automated action. At each gate, name the new permission and the evidence supporting it.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. The next step is often a new campaign or a different crude rather than a second unit. A model validated on one campaign should not be presumed suitable for a materially different feedstock or mode; treat it as a new setting.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. Expansion usually means another unit with its own configuration, identifiers, measurement quality, staffing and approvals. Cost local adaptation and sustained support, or the programme only looks inexpensive.
- Two templates carry this work: the screening sheet in Appendix B, completed before procurement, where a favourable score never overrides an unmet technical, safety, legal or financial condition; and the pilot charter in Appendix E, agreed before the trial.
Record the reason for every deferral.
22:00 – 25:00
Before you close the laptop
- Write down the uncertainty your pilot must resolve, and the result that would make you stop it.
- This session produces Section 19 of your 90-Day Intelligent Refinery Plan: the use-case register entry and pilot charter, with acceptance agreed before results.
- Ask an independent reviewer what result would lead management to stop. If no plausible result can stop the project, the charter is functioning as a sales plan. At closure the decision is to stop, revise, continue narrowly or scale under stated conditions.
- Practice Exam 4 — Full Syllabus opens now: fifteen questions across Sessions 1 to 19. Sit it before Session 20. Three of the Final Exam's questions will be rebuilt from your own earlier mistakes, so read every returned explanation.
Preserve the result even if it is commercially disappointing. The Sequencing Discipline depends on it.
Key Terms
Use-case register
The record of each application's purpose, owners, users, sources, read-propose-write boundary, approvals, exclusions and stopping method.
Confabulation
The generative AI risk, discussed in NIST's Generative AI Profile, of fluent output that is unsupported or incorrect.
Shadow mode
A trial comparing the application's recommendations with actual decisions, without execution authority.
Stopping condition
A result, agreed in the pilot charter before the trial, that would lead management to stop. Without a plausible one, the charter is a sales plan.
Self-Check — Before You Submit
1. A shadow-mode trial on one product tank went well. The supplier proposes letting the application schedule loadings automatically across all product tanks next month, under the same licence. What should the executive require?
- A. Approval, because the software has already been validated
- B. A longer shadow trial on the same tank to build confidence
- C. A reassessment of each new setting and a gate naming the new permission, since advisory use is not approval for automated action
- D. A supplier warranty in place of renewed testing
C. A reassessment of each new setting and a gate naming the new permission, since advisory use is not approval for automated action — The proposal expands scope and authority at once. Each gate must name the new permission and its evidence; extending the original trial tests nothing new, and a warranty is not operating evidence.
2. Which statement about the NIST AI Risk Management Framework 1.0 matches how the book uses it?
- A. It is a voluntary reference for structured risk work, organised around Govern, Map, Measure and Manage
- B. It is a Nigerian legal requirement for refinery AI
- C. It replaces the six frameworks of the Reader's Toolkit
- D. It sets the numerical risk score every use case must pass
A. It is a voluntary reference for structured risk work, organised around Govern, Map, Measure and Manage — The book cites it as a voluntary framework, not a refinery standard or Nigerian legal requirement. The book also warns against reducing judgement to a single numerical risk score, and keeps the six frameworks as the author's own tools.
3. In the book's assumed teaching pilot, 50 recommendations were created, 30 were technically useful and only 15 could be acted on. The sponsor proposes tuning the model to produce more recommendations. What is the better next step?
- A. Approve the tuning, since more output means more value
- B. Stop the project, since half the useful recommendations were lost
- C. Lower the acceptance criteria so the pilot can pass
- D. Explain the response-resource limit first, since more recommendations may add review workload without improving the result
D. Explain the response-resource limit first, since more recommendations may add review workload without improving the result — The constraint is the response, not the volume of output. Moving the acceptance criteria after results defeats the charter, and stopping outright ignores that the pilot has revealed a fixable limit worth stating.
Today’s Assignment
Write Section 19 of your plan: the use-case register entry and pilot charter for your company's first application, with acceptance agreed before any result.
- Write the register entry: purpose, owners, users, sources, what it can read, propose and write, approvals, exclusions and the stopping method.
- State the uncertainty the pilot must resolve, and bound the trial: inclusions, exclusions, users, actions and whether it runs in shadow mode.
- Set acceptance criteria now, covering evidence, useful output, response, business consequence and control, and the failure cases you will test.
- Name the result that would make you stop, and the independent reviewer who will challenge it. Say honestly whether your company has ever stopped a project on evidence.
- State the gate that would follow a good result and the new permission it would grant. If you do not work at a refinery, write the charter for the application your business touches most directly.
YOUR 90-DAY PLAN
Section 19 — The use-case register entry and pilot charter, with acceptance agreed before results.
How this is marked — the rubric
| Register entry complete, with read, propose and write separated and the stopping method named | 25 |
| Trial bounded with a written uncertainty, inclusions and exclusions, and failure cases to be tested | 25 |
| Acceptance criteria set before results across evidence, output, response, consequence and control, with a plausible stopping condition | 30 |
| The next gate names one new permission and the evidence it requires | 20 |
| Total · pass mark 70 | 100 |
Submitting session 19
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: 19, 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 4 — 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 4 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Use-case governanceGenerative output controlChange-in-use disciplinePilot charter designStaged scaling gates
The Leadership Mandate · Chapter 24 · CapstoneSESSION 20
The First Ninety Days
Why This Matters
Nineteen sessions ago you chose one live refinery problem and described it in operating language before anyone attached an AI label to it. Today you put your name to what you have built around it. This session has three parts: the ninety days as the book sets them out, the records that keep the programme honest once the project team has gone home, and the capstone itself, nineteen sections assembled into one signed 90-Day Intelligent Refinery Plan. I have tried to be direct with you rather than encouraging, and I will stay that way for one more session. Then the Final Exam.
25-Minute Module
0:00 – 6:00
The ninety days, as an author's framework
- The aim is not the largest portfolio. It is one consequential decision, reliable evidence, assigned authority and a bounded intervention whose result can be assessed.
- These periods are an author planning framework, indicative rather than promised. Adapt them to your calendar, reviews, staffing and procurement; some applications should stop before a trial.
Period What should be true at the end Days one to fifteen Sponsor and business owner chosen; three recurring decisions compared on the Value-versus-Risk Map; baseline from actual records; one use case chosen at the first gate Days sixteen to thirty Data contract and decision register written; Autonomy Ladder rung set; access reviewed; investment boundary, benefit ledger and pilot charter drafted Days thirty-one to sixty Approved trial run within scope; rejections recorded with reasons; difficult cases tested; participants trained; baseline maintained Days sixty-one to ninety Trial closed against its charter; fallback demonstrated; consequence verified by finance; sponsor decides the next commitment - THE DOCTRINE
Do not let a procurement deadline replace readiness. A delayed software order can be a sensible result of disciplined preparation.
Move the dates if you must. Do not remove the gates.
6:00 – 11:00
Where the ninety days usually go wrong
- In the first fifteen days, reject an application whose main prerequisite is an operating change management cannot fund or authorise; a conventional repair may need to come first.
- Between days thirty-one and sixty, locate the delay: data, review, approval, physical execution or commercial completion, and give each material gap a corrective action. Do not keep tuning the model when the real limitation is an unavailable spare or an unfunded movement.
- By day ninety, write the executive account: decision improved, evidence, actions, outcome, cost and authority requested, failed cases included. If the application stops, keep its data definitions and lessons, and never remove an unfavourable result from the portfolio record.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. The book's conclusion is that a modular installation may begin with a controlled campaign record and a small number of consequential dependencies. One campaign decision, few records, owners you can name tonight.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. A large integrated refinery may need shared models and more extensive interfaces. That is a reason to keep the first decision narrow, not to start several.
The trial's result must remain intelligible after the enthusiasm of launch has passed.
11:00 – 16:00
The continuing cadence and its records
- After ninety days the programme should have a responsible business owner, sustained data ownership, a review route and an understood financial boundary. Decide which decisions belong in daily operating review, which benefits need monthly verification and which changes require a formal gate.
Record Its job Decision register (Appendix A) One record per decision: choice, evidence, action and learning, earlier versions preserved Data quality checklist (Appendix D) Identity, fitness for the decision and ownership, each item with an acceptance condition and owner Executive review pack (Appendix F) Operating, financial and programme accounts, closing with named decisions, owners and dates - THE DOCTRINE
An empty evidence field is a visible gap to resolve. It should not be filled with a confident narrative unsupported by the operating record.
- A failed checklist condition must change the application's behaviour: withhold, qualify or route an exception. Report a portfolio of decisions and verified results, not a catalogue of installed products.
An accurate report still needs management to authorise and resource the response.
16:00 – 21:00
What the doctrine asks of you
- The intelligent refinery is an operating organisation that can carry evidence into a decision and a decision into an accountable result. Software supports it; it does not supply the feedstock, repair the equipment, release the product or fund the movement.
- The six frameworks are the common language: the Loop connects response and learning; the Data-Foundation Precondition establishes evidence; the Autonomy Ladder separates competence from permission; the Value-versus-Risk Map shows consequence; the Sequencing Discipline makes expansion depend on proof; the Operations-Excellence Lens keeps the refinery's result in view.
- THE DOCTRINE
Approve a bounded trial when the case deserves evidence. Repair the foundation when the information is weak. Preserve technical authority when a fluent recommendation exceeds its scope. Stop a project when its economics or controls do not justify continuation.
- Take your live problem into the next operating review: state the decision, establish the evidence, name the person who can act and record the consequence.
Make the first gate real. Approval should require a decision the refinery can execute and evidence it can explain.
21:00 – 25:00
The capstone, the Final Exam, and a last word
- The capstone is Section 20: Sections 1 to 19 assembled into one signed 90-Day Intelligent Refinery Plan with a one-page executive summary on top. It will run well beyond the usual 900 words, and that is expected.
- Two things close the programme, in order. First the capstone, submitted as Session 20. Then the Final Exam, sat on this page: fifteen questions, three of them rebuilt from your own earlier mistakes, closed book, pass mark 70. The certificate of completion requires the capstone and the Final Exam at 70 or above.
- Send me the plan even if two sections embarrass you. I would rather mark an honest document than a tidy one.
- The responsibility of leadership is to carry a sound decision through the organisation. Evidence needs an owner. A recommendation needs judgement. An approved response needs people, resources and follow-up. Technology can help connect those steps, but you remain accountable for whether they happen.
- Begin with a problem your team can recognise. Let the evidence challenge the preferred answer. Preserve the authority of the people who understand the process, and measure the business consequence honestly.
The intelligent refinery is built through those choices, repeated with discipline.
Key Terms
90-Day Intelligent Refinery Plan
The signed document assembled from all nineteen sections, with a one-page executive summary on top.
First gate
The point, around day fifteen in the author's framework, where one use case is chosen, or a conventional repair is chosen first.
Executive review pack
The Appendix F pack of operating, financial and programme accounts that ends each meeting with named decisions and owners.
Continuing cadence
The rhythm after ninety days: daily decisions, monthly benefit verification and formal gates for change.
Self-Check — Before You Submit
1. On day twenty-five a supplier's quotation is about to expire, but the data contract is unfinished and tank identifiers are unconfirmed. The sponsor wants to sign to hold the price. What does the doctrine advise?
- A. Sign the order and finish the data contract during the trial
- B. Repair the foundation first; a delayed order can be a sensible result of disciplined preparation
- C. Sign the order but shorten the trial to recover the lost time
- D. Ask the supplier to complete the data contract as part of the order
B. Repair the foundation first; a delayed order can be a sensible result of disciplined preparation — A procurement deadline must not replace readiness. Signing first commits money before evidence can support the trial, and the data contract belongs to the business, not the supplier.
2. How does the book describe the four periods of the first ninety days?
- A. As implementation times a supplier should guarantee
- B. As a regulatory timetable for AI programmes in Nigerian refineries
- C. As fixed dates every refinery must meet
- D. As an author planning framework with indicative periods, adapted to the refinery
D. As an author planning framework with indicative periods, adapted to the refinery — The periods are indicative, not promised implementation times, and some gates end in a decision to stop. Reading them as a guarantee or a regulatory timetable mistakes an author's framework for an obligation.
3. On day fifty, users accept the maintenance recommendations, but most are not carried out because a critical spare has not arrived. The team proposes retuning the model. What should the sponsor do?
- A. Assign a corrective action to the spare's supply route, since the delay lies in physical execution
- B. Approve the retuning, because better recommendations will raise acceptance
- C. Extend the trial quietly until the spare arrives
- D. Close the trial early and report it as a model failure
A. Assign a corrective action to the spare's supply route, since the delay lies in physical execution — The book warns against tuning the model when the real limitation is an unavailable spare. The delay must be located and owned; hiding it or calling it a model failure misstates the result.
Today’s Assignment
Assemble Sections 1 to 19 into one signed 90-Day Intelligent Refinery Plan for your company's facility, with a one-page executive summary.
- Assemble all nineteen sections in order and edit them for consistency. Where a later session changed your mind, correct the earlier section and say so in a line.
- Name the first use case as a decision, not a product, with its deadline. State its baseline as a figure with its source record and period.
- Name the executive sponsor, process owner, evidence owner and technical custodian by role and name.
- Set the four phase gates with criteria written now, before any result, including the condition that would make you stop.
- State the pause and recovery route, and the benefit-ledger entry finance will verify: mechanism, baseline, counterfactual, currency and cash timing.
- Write the one-page executive summary, sign and date the plan, and name the commitment you are least confident your company will keep. The plan will run beyond 900 words; the summary may not.
- If you do not work at a refinery, assemble the plan around the refinery decision your work touches most directly.
YOUR 90-DAY PLAN
Section 20 — The assembled, signed 90-Day Intelligent Refinery Plan with its executive summary.
How this is marked — the rubric
| All nineteen sections present, in order and consistent, with no placeholders | 25 |
| The first use case stated as a decision with a real baseline and the four accountable roles named | 25 |
| Phase gates with criteria set in advance, a stopping condition, and a working pause and recovery route | 25 |
| A benefit-ledger entry finance can verify, and a signed one-page executive summary | 25 |
| Total · pass mark 70 | 100 |
Submitting session 20
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: 20, 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
Ninety-day sequencingGate disciplineExecutive review cadencePlan assemblyAccountable leadership