THE INTELLIGENT PETROCHEMICAL COMPANY LEADERSHIP PROGRAMME
The decision behind the chemical.
20-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s petrochemical and specialty chemical sector
WEEK 4 OF 4 · AUTHORITY, GOVERNANCE AND THE FIRST NINETY DAYS · SESSIONS 16–20 · WORK AT YOUR OWN PACE
Part Five · Enterprise ExecutionSESSION 16
Workforce, Suppliers and Decision Authority
Why This Matters
By now your decision has a constraint, evidence, a cash calendar and a data contract. None of that matters if the person asked to review the recommendation has no time, no source evidence and no real power to say no. This session is about the people around the decision: your own staff, the experts whose knowledge sits in their heads, and the suppliers who will sell you the capability. I want you to leave able to tell a supplier exactly what you are buying, design a demonstration that is allowed to fail, and refuse to widen any permission until your company has the review capacity to carry it. A platform cannot create those arrangements by itself.
25-Minute Module
0:00 – 6:00
Design around the role and make review substantial
- A laboratory reviewer, a buyer and a customer service engineer each support a different decision, so each needs different information. Training should explain that decision, its evidence, the application's limitations and the escalation route. A general introduction to AI is useful background but insufficient for operating permission.
- Train on the actual workflow: find a wrong unit, a superseded specification or a missing approval, and say what you would do when the application answers confidently without the required source. Competence includes recognising when not to use the output.
- THE DOCTRINE
A button labelled approve does not demonstrate meaningful oversight. A reviewer needs the supporting evidence, relevant expertise and enough time to judge the action.
- Where the consequence warrants it, separate review from execution: the person confirming the technical evidence may differ from the person issuing the financial commitment. Preserve actual approval identities, never a shared account. Present the decision, source evidence, uncertainty and required approval directly, not in pages of generated prose.
- If management rewards only transaction speed, staff may bypass the controls that make the information reliable. Align expectations with dependable completion and responsible escalation.
Ask your reviewers how long they actually have per recommendation. The answer tells you whether review is real.
6:00 – 11:00
Resolve review capacity before expanding permission
- Return to Worked Case Two from Session 7. On the book's assumed 120-batch comparison, the quality-warning model recommends 27 early reviews. Management should calculate the hours, skills and decision timing those reviews require. If only one specialist can resolve the exceptions, the application can overload that role. The response may be a narrower scope, a clearer presentation or additional qualified capacity.
- An approval quota is a poor substitute for competent review. Staff should be able to challenge a recommendation without being penalised for reducing an adoption percentage; equally, a rejection needs its reason and evidence. Review accepted, modified and rejected recommendations together.
- The delegation record states the action, its scope, required evidence, value or quantity limits where relevant, approver and suspension conditions. Permission should survive a change in personnel through controlled roles and records, not one individual's informal understanding of what the assistant may do.
If your delegation lives in one manager's memory, it leaves when that manager does.
11:00 – 17:00
Buy a defined capability and a demonstration that can fail
- The supplier brief should identify the operating decision, records, users, permission boundary and acceptance test, and state the integration, support, documentation and training needed. A proposal should show what your company must provide and what the supplier will deliver.
- Evaluate the supplier on relevant work. A vendor with strong document automation may not have evidence for process modelling.
- THE DOCTRINE
A successful demonstration should show the application's response to uncertainty, not only its most polished output. An attractive demonstration is the beginning of due diligence.
- Give the supplier a representative controlled test set after agreeing the purpose and acceptance method. Preserve the difficult cases the book names: an incomplete lot identity, conflicting specifications, a stale measurement and a request outside permission.
- Separate product capability from implementation dependency. A supplier may demonstrate a valid prediction on a prepared dataset while relying on your team to build the live data pipeline and the review queue. A useful statement of work identifies the records your company supplies, the integration the supplier delivers and the people who accept each component.
- Ask how product revisions, changed data mappings and model updates are handled at the intended scale. An application that works only while its original developer is available creates a continuity problem the executive must price.
Write the four hard cases into the invitation before you see the first slide.
17:00 – 21:00
Contract for the result and keep the knowledge
- Acceptance should include correct operation in the defined scope, traceable records, documented configuration, support readiness and the ability to recover. Define ownership and permitted use of data, formulations, model configurations and generated outputs, and include an exit plan that leaves your company with its records and usable documentation. Avoid a contract that pays entirely for installation while leaving evaluation and adoption unfunded.
- Convert expert knowledge into controlled records with the expert's review. A transcript or generated summary is a draft of what was said, not an approved operating instruction. Where accounts conflict, preserve the question rather than letting an assistant blend them into a synthetic procedure.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional batch formulator records its senior shift chemist explaining a recurring raw-material behaviour. She reviews the draft, the technical owner approves its scope and date, and a conflicting older note is held as an open question.
IN THE MARGIN · THE PARALLEL- Margin. The same formulator's supplier contract ties payment to acceptance in the defined scope, not installation alone, and includes exit documentation. Finance budgets the internal review hours, a resource requirement even where no extra salary is paid.
- Build Nigerian capability through the real work: product-data ownership, laboratory interpretation, forecasting evaluation, security, integration and finance reconciliation, with digital specialists paired with operating people. A certificate can show completed learning; it does not establish authority to release a product or modify a process.
Every pilot should leave your own people more capable, not more dependent.
21:00 – 25:00
Write Section 16 before you close the laptop
- This session produces Section 16 of your plan: the role, substantial review, a demonstration that can fail and review capacity before permission.
- Write down the role that reviews your decision's recommendation, how many reviews per week you expect, and who in your company could actually do them.
- If your company has no AI in use yet, that is normal. Work from the decision or supplier proposal in front of you; the honest answer may be that a clearer workflow or no model at all would serve better.
- There is no paper today. Pop Quiz 4 opens after Session 18.
The uncomfortable number in this section is review hours. Write it anyway.
Key Terms
Substantial human review
Review by a person with the supporting evidence, relevant expertise and enough time to judge the action.
Delegation record
The controlled statement of an action, its scope, required evidence, limits, approver and suspension conditions.
Implementation dependency
Work a supplier's demonstration relies on but does not deliver, such as your live data pipeline or review queue.
Demonstration that can fail
A supplier test on a representative controlled set that includes difficult cases such as a request outside permission.
Self-Check — Before You Submit
1. A fictional specialist supplier's pilot produces far more recommendations each week than its one qualified reviewer can properly assess. The sponsor proposes a high approval target to show adoption. What should you do?
- A. Accept the target as evidence of confidence
- B. Reduce the permission or redesign the queue, resolving review capacity before widening use
- C. Let the reviewer approve in batches before dispatch
- D. Add a second application to pre-screen the first
B. Reduce the permission or redesign the queue, resolving review capacity before widening use — Where volume exceeds review capacity the book says to reduce permission or redesign the queue. An approval quota is a poor substitute for competent review.
2. Which item does the book say a supplier demonstration should include?
- A. The supplier's most successful reference project
- B. A dataset prepared by the supplier to show peak accuracy
- C. A request outside permission, alongside conflicting specifications and a stale measurement
- D. A live connection to your release system
C. A request outside permission, alongside conflicting specifications and a stale measurement — The book asks for difficult cases so the demonstration shows the response to uncertainty. A supplier-prepared dataset is the tempting choice, but it hides the implementation dependency and cannot fail in the ways that matter.
3. A generated summary of a retiring engineer's interviews is ready. What is its status?
- A. A draft of what was said, to be converted into a controlled record with the expert's review
- B. An approved operating instruction, since it came from the most experienced person
- C. A replacement for the current procedure where the two disagree
- D. Training material that confers operating permission
A. A draft of what was said, to be converted into a controlled record with the expert's review — The book treats a transcript or generated summary as a draft, not an approved instruction. Where accounts conflict the question is preserved for review, not blended into a synthetic procedure.
Today’s Assignment
Write Section 16 of your plan: the role, substantial review, the capability you would buy, a demonstration that can fail and the review capacity your company needs before any permission is widened.
- Name the role that reviews the recommendation for your decision, what that person must see and the workflow training they need. Say honestly where review in your company today is a ritual.
- Estimate review volume and the hours, skills and timing it requires, and who can do it.
- Write the supplier brief: operating decision, records, users, permission boundary and acceptance test, with what your business provides and what the supplier delivers.
- Design the demonstration with at least three of the book's difficult cases, and the result that would make you stop.
- Write one delegation record for the first permitted action, and the contract terms for acceptance, ownership and exit.
YOUR 90-DAY PLAN
Section 16 — The role, substantial review, a demonstration that can fail and review capacity before permission.
How this is marked — the rubric
| Names the reviewing role by title, what it must see and an honest account of current review practice | 25 |
| Quantifies review volume against available qualified capacity and states the response if capacity is short | 25 |
| Supplier brief and demonstration include the decision, acceptance test, implementation dependencies and named difficult cases | 30 |
| Delegation record and contract terms are specific, with suspension conditions and an exit plan | 20 |
| 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
Role-based designSubstantial reviewSupplier due diligenceDemonstration designDelegation recordsCapability building
Part Six · Governance and ScaleSESSION 17
Investment Cases, Benefit Attribution and AI Governance
Why This Matters
This is the session your finance director and your board will test you on. An investment paper for AI must meet the same standard as any other: another person should be able to reconstruct the decision from it. That means a baseline that is honest about today, benefits counted once and by their real nature, the full cost of running the thing, and governance that names the actual system rather than a principle. It also means agreeing, before results arrive, when you would suspend it or stop it. Stopping a weak pilot can be a sound decision. A sponsor's seniority or an investment already made should not determine the result.
25-Minute Module
0:00 – 5:00
Compare with a credible baseline
- The alternatives should include improving the workflow without AI where that is feasible. A rules-based check, a reconciled ledger or a clearer approval route may solve the immediate problem. The investment committee should know what additional value the analytical capability offers.
- The baseline must describe the current arrangement accurately: its staffing, procedures and existing safeguards. Do not compare a fully supported new system with an exaggerated version of the old one. Identify where the current process genuinely fails and what evidence shows the consequence.
- THE DOCTRINE
A convincing investment paper should allow another person to reconstruct the decision. The presence of AI does not change that standard.
- Bound the scope: business, product family, users, data and decisions. A pilot on one family is not an enterprise-wide case. Expansion costs belong in later decisions unless they are explicitly in the current request.
If the baseline flatters the proposal, the committee will eventually find out. Better that you find out first.
5:00 – 11:00
Count value once, by its nature, and test attribution
- Plain idea first. Benefit attribution assesses how the intervention contributed to an observed improvement, including the other changes that may explain it. A before-and-after comparison can be moved by product mix, feed quality, maintenance, prices and customer demand.
- Create a benefit ledger linking each claimed benefit to an operating change and evidence, with an owner and a finance reviewer, separating observed results, estimates and proposals. If better quality reduces rework and frees vessel capacity, check that the new campaign's contribution does not include the avoided rework cost again.
Result What it is What it must not be reported as One-time working capital release A cash release Recurring earnings Deferred purchase A deferral An avoided cost, unless the business establishes it will no longer be required Avoided interruption estimate An estimate A measured cash receipt Chapter 3 campaign: ₦600,000 Assumed incremental contribution before programme costs Part of a combined realised return Chapter 10 reliability: ₦440,000 Assumed expected net financial benefit under specified probabilities Part of a combined realised return - The book's four worked cases are separate hypothetical decisions with different boundaries, periods and evidence assumptions. Their benefits should never be added into a series-wide or enterprise return.
- Use a concurrent, matched or phased comparison where feasible. For production, reconcile accepted quantities at a consistent valuation before examining price changes. For labour, distinguish released staff capacity from expenditure actually avoided.
Ask of every benefit line: observed, estimated or proposed? Then ask which other line it overlaps.
11:00 – 15:00
Include the full operating cost and stage the funding
- Include integration, data preparation, technical evaluation, training, support, licences or service consumption, security and review workload. Internal staff time is a resource requirement even where no additional salary is paid. A low prototype cost can conceal the cost of dependable operation.
- Test the result if support cost rises, benefits are smaller or deployment takes longer. A case that pays only under its most favourable assumptions should say so.
- THE DOCTRINE
Fund the smallest stage that can resolve the next material uncertainty. A common platform does not establish common chemistry, measurement methods, customer acceptance or data readiness.
- The book's sequence: if the target is unreliable, fund measurement and record reconciliation before model development. If retrospective performance is credible but response capacity is unknown, fund a shadow workflow. If the live intervention is useful but attribution is unresolved, fund the comparison before broader scale.
- Show fixed implementation cost, recurring support, variable usage and operating-review workload separately. This structure also lets a useful control improvement continue when the AI component proves unattractive.
Each stage should end in a decision. If it does not, it is not a stage; it is drift.
15:00 – 21:00
Govern the actual system, its failures and its exit
- Register the application, purpose, owner, users, data sources and permissions, its evaluation scope and limitations, who can approve changes, who can suspend use, and the fallback.
- The book notes that NIST's Artificial Intelligence Risk Management Framework (AI RMF 1.0) is a voluntary framework for managing AI risks, and that its Govern, Map, Measure and Manage functions provide a useful external structure. The book's recommendation is to connect that structure to the specific application and decision, rather than claiming compliance from a completed generic checklist.
- Agree acceptance thresholds before testing failure modes, and state which defects require correction and which require suspension. Average performance should not conceal an unacceptable high-consequence failure. An assistant allowed to draft an internal laboratory summary does not inherit permission to send a customer certificate.
- THE DOCTRINE
Restarting the software does not establish that the decision system has recovered. After correction, re-evaluate the affected cases and scope before returning to service.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional continuous producer's lot-readiness assistant cites a superseded specification. That defect is a named suspension trigger, so the permission is withdrawn at once, the manual fallback runs, and the affected cases are re-evaluated before return to service.
IN THE MARGIN · THE PARALLEL- Margin. The same producer's committee asks for baseline, scope, evaluated evidence, active permission, full operating cost and next decision. A second pilot shows a smaller opportunity than expected; it is stopped, useful work preserved and access removed.
A governance committee that approves principles while unresolved applications keep running is not governing.
21:00 – 25:00
Write Section 17 before you close the laptop
- This session produces Section 17 of your plan: the credible baseline, value counted once, governance of the actual system and staged funding.
- Write the one alternative without AI that a sceptical finance director would ask you to try first, and what it would cost.
- Note which of your expected benefits is observed, estimated or merely proposed today. Most will be proposed. Say so.
- The next session drafts the ninety days, and Pop Quiz 4 opens at its end.
Write the suspension triggers now, while nobody is defending a running system.
Key Terms
Credible baseline
An accurate description of the current arrangement, including its staffing, procedures and safeguards, against which a proposal is compared.
Benefit ledger
A record linking each claimed benefit to an operating change and evidence, with owner and finance reviewer, separating observed, estimated and proposed results.
Benefit attribution
The assessment of how an intervention contributed to an observed outcome, including other changes that may explain it.
Return to service
Restoring a suspended permission only after correction and re-evaluation of the affected cases and intended scope, with the defect, changed configuration and responsible approval preserved.
Self-Check — Before You Submit
1. A business unit reports a combined AI return by adding the book's ₦600,000 campaign result and ₦440,000 reliability result. What is wrong?
- A. Nothing, provided both figures are recomputed
- B. The figures should be averaged rather than added
- C. Only the reliability figure should be discounted for risk
- D. They are different kinds of assumed result and must not be combined into a claimed realised return
D. They are different kinds of assumed result and must not be combined into a claimed realised return — One is an assumed incremental contribution before programme costs; the other an assumed expected net benefit under specified probabilities. Recomputing them correctly does not change their nature, boundaries or hypothetical status.
2. A fictional batch formulator's pilot has credible retrospective performance, but nobody knows whether the quality team can respond to warnings in time. What should the next funding stage buy?
- A. A shadow workflow to establish response capacity
- B. Enterprise licences, because the model has proved itself
- C. Additional model development to raise accuracy
- D. A before-and-after revenue study
A. A shadow workflow to establish response capacity — The book funds the smallest stage that resolves the next material uncertainty, which here is response capacity. Buying enterprise scope is the tempting error: retrospective performance does not establish the operating mechanism.
3. How does the book position NIST's AI RMF 1.0?
- A. A mandatory standard for AI in Nigerian chemical plants
- B. A voluntary framework whose four functions give external structure, connected to the specific application and decision
- C. A checklist whose completion demonstrates compliance
- D. A replacement for the company's governance committee
B. A voluntary framework whose four functions give external structure, connected to the specific application and decision — The book calls it voluntary and warns against claiming compliance from a completed generic checklist. Its value is as structure applied to the actual system and decision.
Today’s Assignment
Write Section 17 of your plan: the investment case for your decision against a credible baseline, with value counted once, the full operating cost, governance of the actual system and staged funding.
- Describe the current arrangement in your company honestly, with staffing, procedures and safeguards, and state at least one feasible alternative without AI.
- List each expected benefit as a ledger line: operating change, evidence, status (observed, estimated or proposed), owner and finance reviewer. Mark any two lines that could overlap.
- List the full cost: integration, data preparation, evaluation, training, support, licences or usage, security and review workload, separated into fixed, recurring and variable. Say what reverses the case.
- Write the governance register entry for the system: purpose, owner, users, data sources, permissions, who approves changes, who suspends use, fallback.
- Name the next material uncertainty, the smallest stage that resolves it, two suspension triggers and how return to service is decided.
YOUR 90-DAY PLAN
Section 17 — The credible baseline, value counted once, governance of the actual system and staged funding.
How this is marked — the rubric
| Baseline describes the current arrangement accurately and includes a feasible alternative without AI | 20 |
| Benefits listed by nature and status with owner and finance reviewer, overlaps identified, no combined hypothetical returns | 30 |
| Full operating cost separated into fixed, recurring and variable, with the assumptions that would reverse the case | 20 |
| Governance names the actual system, permissions, suspension triggers, return to service and the next funded stage | 30 |
| 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
Investment case disciplineBenefit attributionTotal operating costStaged fundingApplication governanceSuspension and closure
Part Six · Governance and ScaleSESSION 18
The First Ninety Days of AI Leadership
Why This Matters
You now have the parts. This session puts them on a calendar. The book proposes six fifteen-day stages, each with something it must establish and a gate, review or decision at its end. I want you to read that schedule the way the book means it: as the author's proposed leadership sequence, not a universal delivery promise. The pace depends on your business, your data, your technical review and your resources. If a gate is not met, the responsible decision is to resolve it or reduce the scope, never to pretend. By the end of today you will have drafted your own ninety days and the rhythm that follows them.
25-Minute Module
0:00 – 4:00
Read the schedule as a sequence, not a promise
- The objective of the first ninety days is to select a real decision, build adequate evidence, test a bounded intervention and reach a defensible conclusion about its next stage. A collection of disconnected demonstrations does not meet that objective.
- THE DOCTRINE
The ninety-day schedule is the author's proposed leadership sequence, not a universal delivery promise. If a required gate is not met, the responsible decision is to resolve it or reduce the scope.
- The programme reads this as the Sequencing Discipline from Session 1 made concrete: dependable data, a bounded pilot, human oversight where consequences warrant it, and expansion only on demonstrated value.
Promise your board the gates, not the dates.
4:00 – 12:00
The six stages and their gates
Days What the stage must establish What must be true at its end 1 to 15: Decision and mandate One recurring decision with a meaningful consequence; decision record and pilot charter; baseline and benefit evidence agreed; finance reviews the economic boundary A specific funded mandate, or unresolved authority and resources named explicitly 16 to 30: Data and workflow Data contract tested; identities reconciled; current workflow observed; access, security and fallback confirmed Day-30 gate: inputs and workflow adequate for the intended consequence 31 to 45: Bounded capability Minimum capability built and tested on representative and difficult records; role training prepared; technical owner approves use in scope Day-45 gate: fit for a shadow pilot, and the business can review its recommendations 46 to 60: Shadow mode Recommendations recorded without changing the live process; rejections examined; review workload tracked Day-60 gate: whether controlled live use is justified, with what it may do and who approves 61 to 75: Defined authority Approved scope operated; recommendation, decision, execution and outcome preserved; agreed suspension triggers used where necessary; benefit verification begun Day-75 review: useful enough to continue, support consumed, limitations named 76 to 90: Next stage Closure or expansion paper reconciling cost, evidence, benefits, permissions and open issues against the agreed baseline One of four outcomes, each with an owner and resource decision - Days 1 to 15 begin with your business-model record from Session 3. Meet the people who own the work, not only the executives requesting a presentation, and do not proceed as though an enthusiastic meeting supplied authority or resources.
- At day 30, a clean demonstration dataset is not enough if the live records cannot meet the data contract. At day 45, a model's test score does not settle workflow readiness. In shadow mode, record why rejections happened rather than labelling every rejection as poor adoption.
- In days 61 to 75, a high-consequence defect should not wait for the next portfolio meeting. If the intervention improves documentation and reduces investigation time, report that before claiming additional production or revenue.
Write the gate question before the stage starts, so nobody rewrites it after the results.
12:00 – 17:00
Match the ninety days to your type of evidence
- Different evidence runs on different clocks. The quality-warning case needs enough confirmed exceptions to evaluate its errors; the customer-trial case needs suitable conditions, measurements and customer agreement; the imported-order case needs a reconcilable transaction and cash calendar. Ninety days is a management sequence, not proof that all three will mature.
- At the first gate, state the evidence you can realistically obtain. A rare equipment failure may not occur during a short pilot. Historical replay and warning-workflow evaluation can establish part of the case while the interruption-prevention benefit remains uncertain. Report that difference rather than inventing an avoided failure.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional supplier with technical service plans a ninety-day pilot on a customer treatment record. By day 30 it learns the customer's site measurements will not be comparable until a later operating period. The gate is not met; the team narrows the scope to the controlled service record and preserves the changed assumption.
IN THE MARGIN · THE PARALLEL- Margin. The same supplier's finance reviewer revises the investment decision at that gate using actual cost, not the original estimate. The day-90 paper asks for a specified corrective stage with an owner and budget, rather than an extension with no conclusion.
- At each later gate, preserve the assumption that changed. If the pilot cannot obtain its data, reduce scope or fund the measurement work. If reviewer capacity is inadequate, resolve that before adding alerts, as Session 16 required.
The honest plan says which benefit will still be unproven on day ninety.
17:00 – 21:00
Decide the next stage and set the rhythm after day 90
- Choose one of four outcomes: operate the existing scope, expand a defined scope after further evaluation, carry out a specified corrective stage, or stop the application. Continuing indefinitely without a conclusion is not a fifth outcome.
- An expansion states the new product, site, customer group or action, the evidence needed and the cost, and preserves the earlier approval. The next ninety days begin from that decision rather than repeating the original demonstration.
- After day 90: frequent operational reviews for exceptions and actions, a weekly cross-functional review for commitments, and a periodic portfolio review for investment and permission. Match the cadence to your decision windows, not a generic reporting schedule. Keep the source register, benefit ledger and authority record current, and re-evaluate the application when its operating conditions change.
- THE DOCTRINE
At any stage the executive should answer three questions: which decision is better supported, which result has been verified and which permission is justified.
If your day-90 paper cannot be executed on the day it is approved, it is not finished.
21:00 – 25:00
Write Section 18, then sit Pop Quiz 4
- This session produces Section 18 of your plan: the ninety days stage by stage, the gates between them and the rhythm after day 90.
- Before closing, write the one gate your company is most likely to miss, and what you would cut from scope if it did.
- Pop Quiz 4 opens now: five questions on Sessions 16 to 18, closed book. Sit it before Session 19.
Sit the quiz without notes. The plan you sign on Session 20 should not need them either.
Key Terms
Gate
A condition stated at the end of a stage that must be met before the next begins; if it is not met, resolve it or reduce the scope.
Shadow mode
Evaluation in which recommendations are recorded without changing the approved live decision or execution process.
Funded mandate
The output of days 1 to 15: decision, owner, charter, baseline and committed resources, with unresolved authority named.
Closure or expansion paper
The day-90 paper reconciling cost, evidence, benefits and permissions against the baseline and asking for one of four outcomes.
Self-Check — Before You Submit
1. At day 30 a fictional batch formulator's live lot records cannot meet the data contract, though the supplier's demonstration dataset is clean. What is the responsible decision?
- A. Proceed to build, since the demonstration data proves feasibility
- B. Extend the schedule to day 120 and continue as planned
- C. Resolve the gap or reduce scope, for example a smaller product scope or a preparatory control improvement
- D. Move directly to shadow mode to gather more data
C. Resolve the gap or reduce scope, for example a smaller product scope or a preparatory control improvement — The day-30 gate asks whether live inputs and workflow are adequate. A clean demonstration dataset is the tempting reassurance, but the book says it is not enough when the live records fail the contract.
2. Which of these is not one of the four day-90 outcomes in Chapter 17?
- A. Operate the existing scope
- B. Stop the application
- C. Carry out a specified corrective stage
- D. Continue running while further evidence accumulates, without a conclusion
D. Continue running while further evidence accumulates, without a conclusion — The book says continuing indefinitely without a conclusion is not a fifth outcome. Stopping is a legitimate outcome, and a corrective stage must be specified with an owner and resources.
3. How should a leader present the ninety-day schedule to the executive committee?
- A. As the author's proposed leadership sequence, whose pace depends on the business, data, technical review and resources
- B. As a delivery commitment with fixed dates for each stage
- C. As the time needed to prove every benefit in the case
- D. As a standard industry timeline for AI pilots
A. As the author's proposed leadership sequence, whose pace depends on the business, data, technical review and resources — The book states the schedule is not a universal delivery promise. Treating it as proof that all evidence will mature in ninety days ignores that different evidence runs on different clocks.
Today’s Assignment
Write Section 18 of your plan: the ninety days for your company's decision stage by stage, with what each stage must establish, the gate before the next, and the operating rhythm after day 90.
- For each of the six fifteen-day stages, write what it must establish for your decision and the gate question at its end, in your own words and specific to your business.
- State which evidence for your decision runs on a longer clock than ninety days, and what you will report instead of an unproven benefit.
- Name the gate your company is most likely to miss today and why. Be specific about the weakness: data, reviewer time, authority, funding or the customer.
- State what you would cut from scope if that gate is missed, and who decides.
- Set the operating rhythm after day 90: the operational, weekly cross-functional and portfolio reviews, their owners, and what each one decides.
YOUR 90-DAY PLAN
Section 18 — The ninety days stage by stage, the gates between them and the rhythm after day 90.
How this is marked — the rubric
| All six stages written for the student's own decision, each with what it establishes and a specific gate question | 35 |
| Evidence clocks matched honestly, with any benefit that will remain unproven stated rather than assumed | 20 |
| Names the most likely missed gate, the reason and the scope reduction, with an accountable decider by role | 25 |
| Operating rhythm after day 90 defined with owners, cadence matched to decision windows and the four outcomes respected | 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 — Authority, Governance and the Ninety Days 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
Ninety-day sequencingGate disciplineShadow evaluationEvidence timingNext-stage decisionsOperating rhythm
The Operating RecordsSESSION 19
The Operating Records: Traceability, Exceptions, Trials, Cash and Model Evaluation
Why This Matters
A company may discover how weak its records are only when a complaint, an auditor or a cash shortfall forces a reconstruction. The book's appendices give you five records to do that reconstruction deliberately: the batch and lot traceability review, the quality exception log, the application trial record, the working capital review and the model evaluation record. The book treats its tools as the author's synthesis; Appendix A describes itself as the author's recommended review format, not a regulatory form. Today you learn what each one asks and which earlier session it serves, and then you complete the one that matters most for your decision from real evidence in your company. The record will expose something. That is its purpose.
25-Minute Module
0:00 – 4:00
Five records and the sessions they serve
- None of these records replaces a regulatory form, the application-specific technical plan or a competent determination, and a completed template does not demonstrate compliance.
Record Use it to Serves (programme mapping) A. Batch and lot traceability review Test whether the company can reconstruct the material path behind one batch or delivery Sessions 5, 8 and 15 B. Quality exception log Keep one record per exception, separating investigation from disposition Sessions 7 and 8 C. Application trial record Agree a trial with the customer and technical owners before it starts Sessions 9 and 10 D. Working capital review Test a campaign, order, customer programme or replenishment decision for cash Sessions 4 and 14 H. Model evaluation record Make the model's evidence reconstructable, alongside the pilot charter Sessions 7, 11 and 17 - The third column is the programme's mapping, not the book's.
Choose the record where a failure would hurt your company most, not the one you can fill fastest.
4:00 – 9:00
Appendix A: reconstruct the material path
- Establish the identity: review identifier and date; product and specification; lot or batch, with original and current identifiers and aliases; scope, such as receipt to production or receipt to dispatch; owner. Begin with a real record selected by the responsible reviewer.
- Reconstruct the path. Incoming material: supplier, manufacturer's lot where available, receipt date, quantity basis, storage location, documents and inspection, test and hold records, and when it became authorised for use. A batch: approved recipe or process record, actual contributing material lots, quantities, relevant equipment, approved deviations, linked sampling and laboratory evidence, and rework genealogy with the authority for its use. A dispatch: product and lot, quantity, packaging, customer order, specification, release evidence and delivery record, linked to contributing lots where repacked or combined.
Question Record the finding Can the incoming and outgoing identities be linked? Source records and any missing link Are quantity and unit bases consistent? Reconciliation method and unresolved difference Does the applicable specification match the decision? Version, scope and authorised reviewer Is release authority evidenced? Person, date and supporting disposition Are changes and exceptions preserved? Approval, scope and closure evidence Can affected stock and customers be identified? Search result and limitations - THE DOCTRINE
Do not mark traceability adequate solely because every field contains text. The reviewer should be able to reconstruct the physical path using the referenced evidence.
Test a repacked lot, a combined batch and a held material. The easy lot proves little.
9:00 – 14:00
Appendices B and C: exceptions and customer trials
- Quality exception log. Its fields: exception identifier; detection; affected material; requirement; original evidence; immediate status; investigation owner; evidence needed; proposed disposition; authorised disposition; commercial consequence; learning action. Original evidence is kept without silent replacement.
- Preserve the decision history: each repeat test with its reason, an approved correction kept apart from a new determination, and any customer concession with its customer authority and precise scope, so it never becomes a general specification revision. For a model warning, keep the version, input references, recommendation time and whether it was accepted, rejected or modified and why. Measure closure through completed disposition and verified follow-up.
- Application trial record. Define the trial: trial identity; objective; baseline; product; scope; authority; resources; commercial terms. Then establish the evaluation, record the work and conclude. The record gathers leadership and evidence requirements; it does not determine a product rate or operating procedure.
- THE DOCTRINE
Conclude a trial as accepted within the stated scope, not accepted, inconclusive or further evaluation required. Preserve the customer's agreed position rather than treating internal optimism as acceptance.
- An assistant preparing the trial report must cite the controlled source behind each value, never filling missing observations with an assumed trend.
A changed status is not a closed exception. Ask what happened to the stock, the documents and the customer.
14:00 – 19:00
Appendices D and H: cash and model evidence
- Working capital review. Agree the economic boundary with finance. Establish the commitment and link the source contract or approved commercial record. Build the cash calendar with the book's events: supplier deposit; shipment or delivery payment; production and service costs; customer deposit; customer acceptance; collection; funding. Calculate the cumulative net cash position, state the opening balance, and identify the largest funding requirement and when it occurs, as you did for the funding trough in Session 14.
- Test the downside, stating the effect on contribution and cash separately. Do not assign probabilities without a defensible basis; a range can be more honest. A positive margin is not an instruction to proceed if funding or technical evidence remains unavailable.
- Model evaluation record. Its record items: target and decision; scope; source coverage; input availability; development and test; baseline; metrics; threshold; uncertainty; version; live review; suspension. Keep hard cases: a wrong unit, missing identity, superseded specification, unfamiliar qualified condition and a request beyond permission. Report how many cases the model declines to evaluate, and attach the benefit-ledger entry rather than duplicating its value.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional batch formulator completes Appendix H for its quality-warning pilot. The hard-case set reveals that a batch with a missing lot identity still received a score. That defect is preserved explicitly and the abstention rule is added before shadow mode.
IN THE MARGIN · THE PARALLEL- Margin. The same formulator completes Appendix D for the imported raw material behind that product. The calendar shows the largest funding requirement falls before customer collection; treasury confirms what funding is actually available before the order is committed.
Excluding the hard cases from the score does not make the live workflow able to handle them.
19:00 – 25:00
Write Section 19, then sit Practice Exam 4
- This session produces Section 19 of your plan: the operating record that matters most for your decision, completed from real evidence.
- Choose one record. Fill it from your company's actual documents, not from memory. Where a field cannot be completed, leave it visibly open and name the owner and completion date.
- If your business does not run the activity a record covers, use the nearest equivalent decision or the counterpart you depend on.
- Practice Exam 4 opens now: fifteen questions across the full syllabus, Sessions 1 to 19. Sit it before Session 20 and the Final Exam.
The gap the record exposes is more valuable to your plan than the fields it fills.
Key Terms
Batch genealogy
The record linking a batch to the contributing material lots, quantities and relevant processing history.
Disposition
The authorised decision concerning material or product following the relevant evidence and review.
Abstention
An application's declared decision not to provide a recommendation when required evidence or approved conditions are absent; the workflow still needs a defined response.
Recommended review format
The author's structure for reconstructing a decision from evidence; not a regulatory form and not a substitute for a competent determination.
Self-Check — Before You Submit
1. A fictional specialist supplier's traceability review has text in every field, but nobody can link two incoming lots to the batch they supposedly fed. What is the status?
- A. Adequate, because the template is complete
- B. Not adequate: the gap, its consequence, owner and completion date must be recorded
- C. Adequate, provided the release was authorised
- D. Not adequate, so a predictive model should be built to infer the missing links
B. Not adequate: the gap, its consequence, owner and completion date must be recorded — The book says traceability is not adequate solely because every field contains text. Building a model is the tempting fix, but a useful next step may be repairing identifiers before attempting prediction.
2. A fictional supplier with technical service ends a customer trial. Internal results look favourable, but the customer's operating conditions changed mid-trial and the customer has not stated its position. Which conclusion fits Appendix C?
- A. Accepted within the stated scope
- B. Accepted, with expansion to other sites
- C. Not accepted, since the customer has not signed
- D. Inconclusive or further evaluation required, preserving the customer's agreed position
D. Inconclusive or further evaluation required, preserving the customer's agreed position — Appendix C names changes that could make a result inconclusive and preserves the customer's agreed position rather than treating internal optimism as acceptance. Declaring acceptance is the tempting error; the customer has not rejected it either.
3. In the model evaluation record, what should happen to cases the model declines to evaluate?
- A. Report their number and identify the fallback and its workload
- B. Exclude them, since they do not affect accuracy
- C. Assign them the most common prediction
- D. Delete them from the hard-case set
A. Report their number and identify the fallback and its workload — Appendix H says excluding declined cases from accuracy does not show the live workflow can handle them. Excluding them is tempting because the score looks better, but the workload remains.
Today’s Assignment
Write Section 19 of your plan: complete the one operating record that matters most for your company's decision from real evidence, and state what it exposed.
- Name the record you chose (lot traceability review, quality exception log, application trial record, working capital review or model evaluation record) and why it matters most for your decision.
- Complete it from real documents in your business, using the appendix's fields exactly. Do not invent fields, and do not fill a field from memory.
- Leave every field you cannot evidence visibly open, with the owner and completion date.
- State in one paragraph what the record exposed about your company that you did not know or had not admitted.
- State the next action the record requires, such as repairing identifiers, preserving original evidence or confirming funding, and who owns it.
YOUR 90-DAY PLAN
Section 19 — The operating record that matters most for our decision, completed from real evidence.
How this is marked — the rubric
| Chooses the record that fits the decision and explains why it matters most | 15 |
| Completes the appendix fields exactly as the book sets them, from real company evidence, with no invented fields | 35 |
| Leaves unevidenced fields visibly open with owner and completion date | 20 |
| States honestly what the record exposed and the next action with an accountable owner by role | 30 |
| 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
Lot traceabilityException disciplineTrial recordsCash calendarsModel evaluationEvidence reconstruction
CapstoneSESSION 20
The Executive Review: Your Signed 90-Day AI Leadership Plan
Why This Matters
Nineteen sessions ago you named your company and one recurring decision with a meaningful consequence. Since then you have given it an owner, a denominator, defined material, a binding constraint, protected release, a customer, a failure label, a cash calendar, a data contract, reviewers, an investment case, ninety days and an operating record. Today you assemble those nineteen sections into one signed 90-Day AI Leadership Plan, using the book's pilot charter and five acceptance gates, the benefit ledger, the source and obligation register and, where relevant, the Nigerian procurement and commitment review. You end by asking your executive review for one decision. Then you sit the Final Exam.
25-Minute Module
0:00 – 4:00
Carry the decision through
- The Conclusion's argument in full. A chemical company's performance emerges from connected decisions. Suitable material must be procured, production must be feasible, quality evidence must support disposition and the customer application must be understood. Delivery and collection then determine the commercial result.
- AI can help the company see patterns, compare options and organise evidence. The improvement depends on the people authorised to decide, the resources to act and the discipline to examine the outcome.
- THE DOCTRINE
A recommendation that remains on a screen has not closed the operating loop.
- The leadership task is practical: begin with the business model and one important decision; establish an adequate data foundation; keep technical and commercial authority explicit; test against a credible alternative; count the value once and preserve the cash consequences. Ask the six framework questions in the meeting itself.
- The book's closing definition: the intelligent petrochemical company is a company that can explain its decisions and learn from their consequences.
Read your plan against those six questions before you sign it.
4:00 – 10:00
Appendix E: the pilot charter
- Write the charter before developing the application and agree acceptance criteria before examining results. The fields below are the book's; your Sections 1 to 19 supply the content.
Field Entry required Field Entry required Decision Actual choice and operating deadline Capability Retrieval, prediction, optimisation or execution proposed Sponsor and owner Funding authority and accountable decision owner Permissions What the application may read, create, change and send Baseline Current arrangement, evidence and consequence Resources Budget, staff, technical review and support Alternatives Include a feasible control improvement without AI Evaluation Representative cases, metrics and acceptance criteria Scope Product, site, users, records and period Protection Approval, suspension triggers and fallback Benefit evidence Quantity, economic boundary and finance reviewer Closing decision Date and evidence required to continue, expand or stop - Programme mapping, not the book's: Sections 1 and 2 supply the decision and owner; 4 and 17 the baseline, alternatives and benefit evidence; 15 and 16 the permissions and protection; 18 the closing decision.
If the alternatives field holds only AI options, the charter is not finished.
10:00 – 15:00
The five acceptance gates and the gate result
Gate What it confirms Closing requirement in Appendix E One: Input readiness Identities, units, source quality, update timing and required fields; missing and conflicting data tested; conditions in which the application must abstain The owner of each source accepts the relevant data contract Two: Technical fitness The capability against its intended use; a valid evaluation arrangement for prediction; source accuracy and appropriate treatment of conflicts for document work A proposed physical intervention needs its separate technical evaluation and approval Three: Workflow readiness The output reaches a competent person in time and a feasible response exists; workload, training and required resources evaluated Test the recommendation against the actual action; the appendix names no accepting role, so your plan names one (programme addition) Four: Authority and protection Permissions, approval identities, access controls, suspension and recovery; attempted actions outside scope tested Preserve the evaluation version and the responsible approvals Five: Economics and support Cost, cash requirement and benefit categories reconciled; support and supplier exit confirmed; assumptions that could reverse the case Finance accepts the calculation boundary and the evidence for benefit verification - Record each gate as passed, passed with specified conditions or not passed. Name the approving authority and evidence, and preserve the decision and review date.
- THE DOCTRINE
A conditional approval should state what remains prohibited until the condition is met.
- Name every approving authority by role in your company.
Mark a gate not passed when it is not passed. The plan is stronger for it.
15:00 – 20:00
Ledger, register and commitment review
- Benefit ledger (Appendix F), one entry per benefit mechanism: benefit identity; operating mechanism; baseline and comparison; physical quantity; economic category; calculation; attribution; status; review; related entries. In the meeting, begin with decisions and exceptions, then verified value, and make unsupported execution permission visible.
- Source and obligation register (Appendix G): source identity; version; location; status; scope; relevant claim; applicability; evidence required; owner and review. A draft should not be presented as an enacted rule, and updating the assistant's collection does not establish that the business has implemented a change.
- Nigerian procurement and commitment review (Appendix I), where your decision involves a chemical commitment: chemical identity; intended use; classification; regulatory route; trade documents; shipment readiness; usable site date; currency commitment; cash calendar; downside; commitment authority; closure. Preserve conflicting sources: the header-date question in NAFDAC's guidelines for obtaining a permit to import industrial and laboratory chemicals, from Session 5, needs clarification before a current obligation is inferred, and SON's SONCAP FAQ shows why intended use matters to an exemption claim.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional continuous producer's gate four passes with a specified condition: the lot-readiness check may advise but may not change any hold status. Release stays with the designated authority.
IN THE MARGIN · THE PARALLEL- Margin. Its ledger shows reduced investigation time as observed, a one-time working capital release as estimated and added contribution as proposed, the last two linked as related entries.
An order is commercially complete only under the acceptance and collection boundary you set for the review.
20:00 – 25:00
Assemble, sign and ask for a decision
- Assemble Sections 1 to 19 in order, noting in one line wherever a later session changed an earlier one.
- On top goes a one-page summary: the decision, its owner and authority; the pilot charter and the five gates with their results; the ninety days stage by stage; the benefit ledger; the source and obligation register; the procurement commitment review where relevant; and the decision asked.
- Ask the executive review for one of Chapter 17's four outcomes: operate the existing scope, expand a defined scope after further evaluation, carry out a specified corrective stage, or stop the application, with owner, resources and next review date. Stopping is a sound decision if the evidence supports it.
- Sign and date the plan. Then sit the Final Exam: fifteen questions, twelve from the paper and three rebuilt from your own earlier mistakes, closed book, pass mark 70.
Sign only what you would be prepared to review honestly on day ninety.
Key Terms
90-Day AI Leadership Plan
Sections 1 to 19 assembled around one recurring decision under a signed one-page summary that asks for one decision.
Pilot charter
The record giving a pilot its decision purpose, written before development, with acceptance criteria agreed before results are examined.
Acceptance gate
One of five checks recorded as passed, passed with specified conditions or not passed, with approving authority and evidence.
Source and obligation register
A register of references, instruments, specifications and customer requirements that keeps the source record distinct from the applicability decision.
Self-Check — Before You Submit
1. Gate four is passed with a specified condition. What must the record also state?
- A. The expected financial return
- B. What remains prohibited until the condition is met
- C. The date the application will reach full scale
- D. The supplier's reference customers
B. What remains prohibited until the condition is met — Appendix E requires a conditional approval to state what remains prohibited until the condition is met. Without that, a conditional pass quietly becomes an unconditional one.
2. A fictional specialist supplier's assistant has been updated with a revised customer specification. The quality manager reports the change as implemented. What is missing?
- A. Nothing; the assistant now reflects the new source
- B. A new pilot charter
- C. Retraining the model on the revised specification
- D. The affected procedures, applications, documents and roles identified, with a competent reviewer and action assigned
D. The affected procedures, applications, documents and roles identified, with a competent reviewer and action assigned — Appendix G says updating the assistant's document collection does not establish that the business has implemented the change. Treating the updated collection as implementation is the trap.
3. Your day-90 evidence shows the application improved documentation, but its benefit is smaller than its operating cost. The sponsor wants to keep it running while more data accumulates. What do you ask the executive review for?
- A. One of the four outcomes, such as a specified corrective stage or stopping, with owner, resources and review date
- B. Continued operation without a decision
- C. Expansion to a second site to spread the cost
- D. A combined return using the book's worked-case figures
A. One of the four outcomes, such as a specified corrective stage or stopping, with owner, resources and review date — Continuing indefinitely without a conclusion is not a fifth outcome, and an investment already made should not determine the result. A stopped pilot can be a sound investment decision.
Today’s Assignment
Assemble Sections 1 to 19 into one signed 90-Day AI Leadership Plan for your company, with a one-page summary on which your executive review can decide.
- Assemble all nineteen sections in order, noting in one line wherever a later session changed an earlier one.
- Write the one-page summary: the decision, its owner and authority, and the decision you ask of the executive review.
- Complete the Appendix E pilot charter and record each of the five gates as passed, passed with specified conditions or not passed, with evidence and approving authority. Say honestly which gates your company would not pass today.
- Attach the ninety days, the benefit ledger entries with status, the source and obligation register entries, and the Appendix I review if your decision involves a chemical commitment.
- Ask for one of the four outcomes with owner, resources and next review date.
- Sign and date the plan, and name the commitment you are least confident your business will keep.
YOUR 90-DAY PLAN
The signed 90-Day AI Leadership Plan.
How this is marked — the rubric
| Sections 1 to 19 assembled, consistent, signed and dated, with a one-page summary naming the decision, owner and authority | 20 |
| Pilot charter complete in the book's fields, five gates recorded honestly with evidence and approving authority | 30 |
| Ninety days, benefit ledger and source register attached, with benefits by status and no double counting | 25 |
| One of the four outcomes requested with owner, resources and review date, and the least certain commitment named | 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
Plan assemblyPilot charteringGate decisionsBenefit verificationSource disciplineExecutive decision