15-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigerian crude, petroleum-products and gas trading
WEEK 3 OF 3 · RISK, TREASURY, TRUST AND THE LEADERSHIP CHARGE · SESSIONS 11–15 · WORK AT YOUR OWN PACE
Part Five · Risk, Treasury and TrustSESSION 11
Market Risk, Hedging, Liquidity and Trade Finance
Why This Matters
Your decision now has a contract, a price clock, a margin and a voyage behind it. This session asks what it can cost you while you carry it, and whether you can pay on the day the bill arrives. Chapter 15 shows why a hedge protects against a defined movement without guaranteeing the margin. Chapter 16 shows why a profitable trade can run out of cash before the margin is collected. The discipline is the same: define the exposure, keep the residual visible, date the cash. By the end you will know the first day your decision could become infeasible.
25-Minute Module
0:00 – 5:00
Measure the exposure the contract creates
- One cargo may carry benchmark, grade differential, freight and currency exposure at the same time.
- The plain idea first. A sensitivity says how much a position's value changes when one thing it depends on moves by one unit, such as one dollar on the benchmark. Define the risk factors and their sensitivity units before any measure is reported.
- Adding each sensitivity times its factor's change approximates a simple linear portfolio; for nonlinear instruments, large moves or changing quantities, revalue through the accepted pricing engine.
- THE DOCTRINE
Risk begins with price formulas, quantities, fixing states, currencies and dates. Netting all values into one nominal number can obscure the risks that remain.
- For the decision you named in Session 1, list the factors your contract actually exposes you to and the date each stops moving.
If the exposure report has one number, ask how many factors were added together to make it.
5:00 – 10:00
A benchmark hedge leaves basis
- A hedge is an offsetting position: if the cargo loses value when the benchmark falls, a short benchmark position gains. But the cargo is not priced at the benchmark alone. The gap between what you hold and what you hedged with is basis; here, the grade differential.
Chapter 15 assumed case, 100,000 barrels At purchase At sale Change in value Physical inventory (benchmark plus differential) $80 + $2 = $82 $75 + $1 = $76 Falls $600,000 Short benchmark hedge, same illustrative quantity and timing $80 $75 Gains $500,000 Remaining loss: differential exposure $2 $1 $100,000, before fees, financing and other effects - Timing mismatch, volume change, quality adjustment and instrument settlement can create further differences.
- Every hedge recommendation shows exposure before and after, remaining basis, quantity and timing mismatch, execution cost, collateral consequence and the case that makes it wrong. Approval refers to that evidence version; a later amendment can require renewed review.
- THE DOCTRINE
A hedge works against the movement it was defined for. It does not guarantee the margin. Explain the residual rather than call the hedge a failure or the trade risk-free.
When a hedged trade still loses money, the first question is what was left unhedged, not who failed.
10:00 – 15:00
Tail measures and the risk engine
- Plain language first. Line up the losses a position could suffer over a chosen period, best to worst. Value at Risk is the point beyond which only a chosen small share of outcomes lie; expected shortfall is the average beyond that point. Neither predicts the worst possible loss. State the portfolio, horizon, confidence level, scenario method and treatment of nonlinear exposures.
Book's equal-probability teaching distribution Value Five losses negative $100,000; zero; $50,000; $150,000; $500,000 VaR at 80%, lower empirical quantile $150,000 Expected shortfall, average of the worst 20% $500,000 The book's warning Five observations are insufficient for a real tail-risk estimate - The Basel Committee's market-risk standards give an example of formal expected-shortfall modelling in bank regulation; they are not automatically the rules for a Nigerian commodity trader. Your business sets its own authorised risk policy within applicable requirements.
- AI may examine factor behaviour and exposure anomalies, and an optimiser may recommend hedge alternatives under limits; controlled engines calculate accepted measures. Independent risk and the authorised trader approve execution.
- Of a risk engine, ask: are contracts, windows, basis and hedges revalued under the same scenarios, separating mark-to-market loss from collateral cash? Do stresses break history, with the benchmark falling while the differential worsens? Predictive accuracy is not evidence of an executable edge after costs.
A small historical risk number can sit beside a large operational exposure it never modelled.
15:00 – 21:00
Profit and cash follow different clocks
- Supplier payments, freight, duties or charges, collateral and customer terms can create a peak funding requirement larger than the expected contribution. Treasury needs a dated cash ladder by currency and legal entity: a day-by-day table of cash, obligations and reliable receipts. A receivable is not cash. An undrawn limit is not available funding until its conditions and draw process are satisfied.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. The book's assumed case: 100,000 barrels in inventory, hedged short. The benchmark rises $10 a barrel. The product may now be worth more, but that value may stay unavailable until sale and collection.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The same move creates a simplified $1 million hedge loss and potential variation-margin cash need under the assumed settlement arrangements. The offset protects economic exposure without guaranteeing liquidity; the views reconcile only when timing is shown.
- A forecast distinguishes contractual due date, expected receipt and stressed receipt, and keeps binding obligations visible. The optimiser should not fund a cargo with a facility whose conditions are unresolved.
- Of the cash-path model, ask whether it runs day by day without counting an unapproved facility. For collections it may use a hazard: the chance an invoice is paid in this interval, given it is still unpaid. An invoice unpaid when observation ends is censored, not labelled late by an invented date. Summing hazards as payment probabilities double counts the receivable.
- THE DOCTRINE
Stress delayed collection, currency change, higher collateral and a smaller facility draw together. Report the first day the plan becomes infeasible. A month-end cash total can conceal an inability to meet a payment on day twelve.
Ask treasury for the worst day in the ladder, not the month-end balance.
21:00 – 25:00
Write the exposures and the cash ladder
- Before cargo approval, treasury signs off the base and downside cash ladders: peak need, committed funding, minimum headroom and conditions requiring action. Any AI narrative reconciles exactly with them.
- Model when foreign currency must be acquired and when each sale price fixes; a delayed pass-through makes the exposure time-dependent.
- Write down the factor you currently leave unhedged without saying so. Today's assignment produces Section 11. Execution and funding approvals stay with the people authorised to give them.
The plan is feasible only if it survives the day the collateral call and the late receipt arrive together.
Key Terms
Basis
The difference between what a business holds and what it hedged with, such as a grade differential left open by a benchmark hedge.
Value at Risk
A quantile of a defined loss distribution for a defined portfolio, horizon and confidence level; not a prediction of the worst possible loss.
Expected shortfall
The average of losses in the tail beyond the Value at Risk point, under a stated distributional treatment.
Cash ladder
A dated, daily view of cash by currency and legal entity, separating unrestricted cash, committed funding, receivables and obligations, showing peak need and the first infeasible day.
Self-Check — Before You Submit
1. In Chapter 15's assumed case, physical inventory falls $600,000 and the benchmark hedge gains $500,000. How should the remaining $100,000 loss be explained?
- A. The hedge failed
- B. Differential exposure the benchmark hedge did not cover
- C. Execution cost
- D. A rounding difference
B. Differential exposure the benchmark hedge did not cover — The hedge worked against the benchmark move; the differential narrowed from $2 to $1. Calling the hedge a failure is the tempting error: explain the residual.
2. A fictional product importer is short a benchmark hedge on 100,000 barrels. The benchmark rises $10. The trader says the inventory gain offsets the hedge loss. Your response?
- A. Agree, because the economic exposure is offset
- B. Close the physical position immediately
- C. Ask treasury for the dated cash ladder: the $1 million hedge loss may need variation margin now, while the inventory gain waits for sale and collection
- D. Ask for a higher Value at Risk limit
C. Ask treasury for the dated cash ladder: the $1 million hedge loss may need variation margin now, while the inventory gain waits for sale and collection — The offset protects economic exposure without guaranteeing liquidity. Agreeing is tempting because the economics net, but cash and profit follow different clocks.
3. Which statement about the Basel Committee's market-risk standards matches the book?
- A. A bank-regulation example of expected-shortfall modelling, not automatically a Nigerian trader's rules
- B. The capital rules for Nigerian commodity traders
- C. A requirement for VaR at 80%
- D. A replacement for company risk policy
A. A bank-regulation example of expected-shortfall modelling, not automatically a Nigerian trader's rules — The register uses Basel only as a bank-regulation example. Treating it as binding on a trader is the error; the business sets its own authorised policy.
Today’s Assignment
Write Section 11 of your plan: the exposures your decision's contract creates, what remains after the hedge or protection your company uses or would consider, and a dated cash ladder with its first infeasible day.
- List the exposures your decision creates (benchmark, differential, freight, currency, timing), each with its sensitivity unit and the date it stops moving. Give structure and size, not confidential prices or names.
- State the hedge or protection your company uses or would consider, and what would remain: basis, quantity and timing mismatch. If you do not hedge, say what remains open.
- Write the evidence a hedge recommendation must show, and who approves execution by role.
- Sketch a dated cash ladder by currency and legal entity: peak need, committed funding, minimum headroom and unresolved facility conditions.
- Run the joint stress (delayed collection, currency change, higher collateral, smaller draw); name the first infeasible day and the funded alternative. Say honestly where your company reports month-end cash instead.
YOUR 90-DAY PLAN
Section 11 — The exposures, the residual basis after hedging and the dated cash ladder with its first infeasible day.
How this is marked — the rubric
| Lists each exposure with a sensitivity unit and date, without netting them into one number | 25 |
| States the residual after hedging or protection, including basis and quantity or timing mismatch | 25 |
| Specifies the hedge evidence and the approval roles, keeping independent risk and the authorised trader for execution | 20 |
| Builds a dated ladder by currency and entity with peak need, committed funding, headroom and the first infeasible day under a joint stress | 30 |
| Total · pass mark 70 | 100 |
Submitting session 11
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 11, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Exposure definitionBasis after hedgingTail-measure literacyCash ladderJoint stressTreasury sign-off
Part Five · Risk, Treasury and TrustSESSION 12
Credit, Regulatory Evidence, Sanctions and Fraud
Why This Matters
A trade that is priced, hedged and funded can still fail at the counterparty, the regulator or the bank account. This session covers three exposures that are easy to discuss in general and hard to evidence in a particular trade: credit that begins before the invoice, regulatory requirements that depend on the actual transaction, and identity and payment details that can be changed by the wrong person. It ends with the book's own source register, which shows how an executive keeps external references honest. Throughout, an indicator is a question for investigation, never a finding against anyone.
25-Minute Module
0:00 – 5:00
Credit exposure begins before the invoice
- A customer can create exposure through outstanding receivables, delivered but unbilled product and accepted future commitments. Measure exposure according to your approved policy, and do not count the same delivered quantity both as an open commitment and as an invoice after the transition.
- Name the target before anyone shows you a score. Paying after the due date differs from default, dispute, restructuring or loss after recovery. A short history of settled invoices may support a late-payment advisory while offering weak evidence for rare default prediction.
- THE DOCTRINE
A model can rank review needs. The credit authority decides limits, security, terms and exceptions. A long customer relationship is not proof of capacity to pay.
- Apply it to your decision. If it creates customer exposure, find where that exposure starts. If it does not, find the counterparty your decision depends on paying or performing.
Ask which of the five targets the score predicts. If nobody can say, nobody can use it.
5:00 – 10:00
Estimate risk, then let the credit authority decide
- Expected loss in plain words: how much is at stake, how likely the defined credit event is over the defined horizon, and how much would be lost if it happened. The book's illustrative calculation:
Illustrative input Case 1 Case 2 Exposure $2 million $4 million Default probability over the defined horizon 3% 3% Loss-given-default assumption 40% 40% Expected loss $24,000 $48,000 - This is illustrative, not an impairment determination or a validated estimate for a real customer. Concentration, wrong-way risk and changing recovery conditions may invalidate the unchanged assumptions.
- Defaults are rare, so a model predicting that every customer pays can look accurate. Ask for calibration, missed consequential events and review burden; unnecessary restriction can damage a viable relationship. Of the model, ask: did regularised logistic regression come first, with a gradient-boosted classifier only if the data justify it, and were post-default notes kept out of approval-time features?
- In collection, AI can assemble invoice, acceptance, disputes and promised date into a brief; a payment promise remains a promise until cash is received and matched. Ranking by lateness alone can send staff where a call changes nothing.
Require a reason and an expiry for every material credit exception.
10:00 – 15:00
Regulatory applicability, identity and maritime evidence
- The regulatory record identifies activity, product, entity, facility, route, jurisdiction, instrument, requirement, evidence and owner. Current specialists determine applicability. A retrieved legal paragraph does not establish that a trade satisfies it, and an old portal manual is not a current fee schedule.
- Record contracting entity, beneficial-ownership evidence, bank beneficiary, intermediaries, vessel identity and product origin; names alone are insufficient. The FATF and Egmont Group Trade-Based Money Laundering Risk Indicators (March 2021) are indicators for investigation, not proof that a flagged party committed an offence.
- OFAC's maritime advisory of 16 April 2025, on Iranian oil sanctions evasion, discusses deceptive shipping, document manipulation and vessel-history due diligence. Use it for risk questions; applicable sanctions and consequences require current transaction-specific review.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. A fictional exporter's voyage record shows a gap in vessel-location data and a ship-to-ship transfer. AI may compare voyage events with documents and find an inconsistent identifier. A legitimate transfer and a data gap should not automatically become an allegation; material concerns are corroborated with permitted authoritative evidence.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The same cargo's payment and contract record is routed to qualified review. The daily exception report states what evidence is missing, what activity is paused under approved policy and who can resolve it. The expected margin stays separate from compliance status.
Write investigate, never guilty, on every indicator your systems raise.
15:00 – 21:00
Bank details, networks and the source register
- A changed beneficiary or account is a high-consequence action. A payment-data assistant flags it and routes it through independent verification using your accepted process, never the same email thread that requested the change. The approved payment master stays outside the language model's direct authority.
- A graph connecting shared addresses, ownership and accounts shows association, not necessarily misconduct. Of a ranking model, ask whether deterministic screening came first and whether it was tested on time-separated investigated cases. GraphSAGE's benchmark evidence is not evidence of sanctions-screening accuracy. Generated allegations are not evidence; authorised reviewers decide any block.
- THE DOCTRINE
An attractive opportunity cannot make an unresolved mandatory condition disappear. Compliance status stays separate from expected margin.
- Appendix D, reviewed on 11 October 2026, keeps each source with its identity, date and limited use. An external source is not evidence of a Nigerian company's results or endorsement of the doctrine.
Register entry Stated limited use S1 Petroleum Industry Act, 2021, Act No. 6 Sections 4 and 29: institutional distinction; no universal licence, levy or tax matrix reproduced S2 NMDPRA permit portal Portal functions only; current requirements need the approved process S3 ICC Introduction to Incoterms 2020 Title and payment fall outside the rules; contract and applicable law decide S4 API 2024 catalogue; S11 GIIGNL handbook, sixth edition, 2021 Existence and scope of the guidance; obtain the accepted edition S5 NIST AI RMF 1.0 (2023) and NIST AI 600-1 (July 2024) Voluntary cross-sector foundations S10 US EIA refining pages; S12 Basel market-risk standards Not Nigerian refinery baselines; a bank example, not a trader's requirement S13 FATF and Egmont Group (March 2021); S14 OFAC (16 April 2025) Indicators, not proof; stated maritime context only S6 to S9 and S15 to S22 research papers A mechanism only, never trading performance
A source without a date and a use boundary is a rumour with a citation.
21:00 – 25:00
Write the credit, regulatory and payment controls
- Before you close the laptop, write down where your decision's credit exposure begins and who in your business can change a bank beneficiary today, and how.
- Today's assignment produces Section 12 of your plan. Record each source your decision relies on as Appendix D does: identity, date, access date and the limited use you make of it.
- This is not legal, sanctions or compliance advice. Specialists determine applicability; the credit authority decides limits; authorised reviewers decide any block.
The control that matters is the one that holds when the request looks urgent and familiar.
Key Terms
Expected loss
Exposure multiplied by the probability of a defined credit event over a defined horizon and by the loss given that event; an illustrative risk measure, not an impairment determination.
Credit target
The precisely defined event a credit model predicts over a stated horizon: late payment, default, dispute, restructuring and loss after recovery are different targets.
Regulatory applicability record
Activity, product, entity, facility, route, jurisdiction, instrument, requirement, evidence and owner for the actual transaction, decided by current specialists.
Use boundary
The limited purpose for which an external source supports the doctrine, recorded with the source's identity and date.
Self-Check — Before You Submit
1. Using the book's illustrative assumptions (3% default probability, 40% loss-given-default), what is the expected loss on $4 million of exposure?
- A. $24,000
- B. $120,000
- C. $1.6 million
- D. $48,000
D. $48,000 — $4 million × 3% × 40% = $48,000. $24,000 is the $2 million case. The book adds that concentration and wrong-way risk may invalidate unchanged assumptions as exposure grows.
2. An email from a known supplier's address asks your fictional gas aggregator to pay this month's invoice to a new account, urgently. The payment assistant flags the change. What should happen?
- A. Reply to the email to confirm the new details
- B. Route the change through independent verification using the company's accepted process, outside that email thread, before the payment master changes
- C. Let the assistant update the payment master since it detected the change
- D. Pay half to each account
B. Route the change through independent verification using the company's accepted process, outside that email thread, before the payment master changes — Bank-detail changes are high-consequence actions and must never be verified through the thread that requested them. Replying to confirm feels diligent but uses the channel that may be compromised.
3. A graph model links a prospective buyer to an intermediary through a shared address. What does that connection establish?
- A. Association that a qualified reviewer checks against current authoritative evidence, with innocent explanations considered
- B. Sanctions exposure
- C. Misconduct by the buyer
- D. Nothing, so it should be ignored
A. Association that a qualified reviewer checks against current authoritative evidence, with innocent explanations considered — A connection is evidence of association, not misconduct; shared addresses can have innocent explanations. Ignoring it is the opposite error: it is a ranked lead for review.
Today’s Assignment
Write Section 12 of your plan: the credit exposure your decision creates and who approves it, the regulatory evidence record with its sources, and the controls on identity, provenance and payment changes in your company.
- Define where your decision's credit exposure begins before the invoice (receivables, delivered unbilled product, accepted commitments), the target any credit or collection model would predict, and who approves the limit or exception by role.
- Map the regulatory applicability of your decision across activity, product, entity, facility, route, jurisdiction, instrument, requirement, evidence and owner. Mark which entries a specialist has confirmed and which remain open. Do not state requirements a specialist has not confirmed.
- Record each source the decision relies on in the Appendix D form: identity, publication date, access date and the limited use you make of it.
- State how your company verifies counterparty identity, product provenance and bank-detail changes, and who may pause the activity. Say honestly where a change could today be accepted through the requesting email.
- Describe structure and size only; no counterparty names, account numbers or contract terms.
YOUR 90-DAY PLAN
Section 12 — Credit exposure and its approval, the regulatory evidence record and the controls on identity and payment changes.
How this is marked — the rubric
| Defines exposure before the invoice and a single named target, with the credit authority retaining the decision | 25 |
| Maps applicability across all ten fields with owners, separating confirmed from open items | 25 |
| Records sources with identity, dates and use boundary in the Appendix D form | 20 |
| States independent verification of identity, provenance and bank-detail changes, and who may pause, without treating any indicator as proof | 30 |
| Total · pass mark 70 | 100 |
Submitting session 12
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 12, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Credit exposureTarget definitionRegulatory applicabilitySanctions boundariesPayment-change controlSource discipline
Part Five · Risk, Treasury and TrustSESSION 13
Govern Models and Agents
Why This Matters
Every AI use you have designed so far may be connected to tools, documents and other systems, and some may be allowed to act. This session is about keeping that authority concrete. Chapter 19 asks you to govern the whole application, not just the model; to write down what an agent may and may not do; to test it with hostile and ordinary cases; and to make change, monitoring and retirement part of operations. The standard is simple to state and demanding to meet: six months from now, you should be able to replay what influenced a disputed cargo. At the end, Pop Quiz 3 opens.
25-Minute Module
0:00 – 5:00
Govern the whole application
- NIST's AI Risk Management Framework (AI RMF 1.0, 2023) organises risk work around Govern, Map, Measure and Manage. Its Generative Artificial Intelligence Profile (NIST AI 600-1, July 2024) addresses additional risks relevant to generative applications. These are voluntary cross-sector resources, not Nigerian trading approvals. Apply them through an operating register with named owners and evidence.
- Each application needs one register entry describing the whole of it, so that anyone accountable can read what it is for and what it may do:
Register field What the executive should be able to read Intended and excluded use The decision it supports, and the decisions it must not touch Data sources and dependencies Which accepted records, services and vendors it relies on Evaluation The evidence it was accepted on, including failed cases Authority What it may read, draft, recommend or execute, and who approves Monitoring Input quality, behaviour, corrections, unauthorised attempts and consequences Change process and recovery What triggers renewed review, the previous accepted version and the fallback staff can operate - THE DOCTRINE
A model version alone is not the application. Retrieval settings, prompts, tools, source permissions and reviewer workflow can change the result.
If the register names a model but not its tools and reviewers, it is describing a part, not the application.
5:00 – 11:00
Make agent permissions concrete
- In plain words, an agent is an application that repeatedly looks at the current situation, picks a permitted next step and calls a tool to carry it out. Its capability depends on the reasoning model and on its tool permissions. This is the Autonomy Ladder from Session 1 applied to tools: approval to summarise an agreement does not authorise a price commitment, hedge order, payment or cargo release.
- The book's example. A contract-review agent may read approved files, calculate through an approved service and draft an exception. It should not amend the accepted contract, release cargo or change a bank beneficiary. An authorised task-creation agent can open a review item within a defined scope, with duplicate prevention and a clear cancellation route.
- Around the language model sits a state machine: a fixed sequence of stages the work must pass through in order, like a cargo that cannot be released before its documents are accepted. A draft recommendation moves to validated evidence, then policy checks, then human acceptance where required. The model cannot skip a stage by describing the action as urgent. An execution service checks authority independently, using current trade state and accepted limits at the moment of action.
- An idempotent action is one that has the same effect whether it is sent once or twice, usually because it carries a unique transaction key. When a tool call times out, the outcome is unknown: it may have succeeded. Distinguish accepted request, confirmed completion, rejected request and unknown outcome, and investigate transaction state before retrying.
- THE DOCTRINE
If an action times out, determine whether it succeeded before retrying. An articulate agent that repeats a payment after a timeout is a failed operating design.
List what the agent may not do as carefully as what it may. The second list is the one that protects you.
11:00 – 16:00
Test hostile and ordinary conditions
- Documents and emails can contain instructions aimed at the assistant. Treat source content as evidence, not authority to change permissions. Test malicious text, conflicting clauses, stale data, wrong entities, denied access, interrupted tools, duplicate operations, malformed units and changed permissions.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. A fictional product distributor's agent retrieves a scanned inspection document containing a line addressed to the assistant, telling it to mark the parcel ready for release. The line remains document content. It cannot grant tool authority, and cargo release stays with the competent release authorities.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The same distributor's payment request times out. The agent reports a failure; the bank may have accepted it. The execution service holds the transaction key, records an unknown outcome and checks state before any retry, so one payment does not become two.
- Include ordinary cases. A system that refuses every question may avoid unsupported answers while failing to assist the business. Acceptance balances supported useful work with appropriate abstention and controlled escalation.
- Evaluate whether the complete system preserves its boundaries, not merely whether the model refuses a malicious sentence in isolation. Give the independent reviewer access to test failures and rejected recommendations, and require a release decision with scope and conditions.
A test set without ordinary cases measures caution, not usefulness.
16:00 – 21:00
Change, monitoring and retirement
- A new benchmark, contract type, destination, customer segment or action tool can exceed the validated scope. Assess the affected decisions, repeat the relevant tests, and retain the previous accepted version and a fallback staff can operate. Separate model weight changes from feature, retrieval, tool and policy changes: an unchanged language model can behave differently after the retriever or prompt changes.
- Monitor outcomes after their labels mature. Record data drift (the inputs look different), performance drift (the results get worse) and changed operating conditions separately. A feature change can occur without harm; a critical unit error can cause harm without any gradual drift signal.
- THE DOCTRINE
Drift detection signals change; it does not establish the cause or the remedy. The owner decides whether to investigate, recalibrate, restrict or retire the application.
- A vendor demonstration can support discovery; it cannot replace your company's own acceptance evidence for consequential trading use.
- Plan retirement: archive the evidence needed to replay material decisions, withdraw obsolete versions from new use and preserve historical references. Ask what model influenced a disputed cargo six months earlier and expect a reproducible answer, not a current model's explanation of an old decision.
Practise the fallback before the interruption, not during it.
21:00 – 25:00
Write the register entry and the permissions
- Before you close the laptop, write the one action your decision's AI use must never take, and the control that would stop it if someone, or some document, asked.
- Today's assignment produces Section 13 of your plan. If your company has no AI in use yet, write the entry for the application your proposal would introduce.
- Pop Quiz 3 — Risk, Treasury and Trust opens now: five questions on Sessions 11 to 13, closed book.
Authority you cannot replay is authority you cannot defend.
Key Terms
Application register
The operating record for an AI use: intended and excluded use, data sources, evaluation, authority, dependencies, monitoring, change process and recovery, with named owners.
Agent
An application that repeatedly observes state, selects a permitted next step and calls tools; its capability depends on both the model and its tool permissions.
Idempotent action
An action carrying a unique key so that sending it twice has the same effect as sending it once.
Unknown outcome
The state after a timeout when it is not established whether the action succeeded; investigated before any retry.
Self-Check — Before You Submit
1. What status does the book give NIST's AI Risk Management Framework and its generative-AI profile?
- A. Mandatory Nigerian trading approvals
- B. Certification that a model is safe
- C. Voluntary cross-sector resources, applied through an operating register with named owners and evidence
- D. Rules that replace company policy
C. Voluntary cross-sector resources, applied through an operating register with named owners and evidence — The book and its source register call them voluntary cross-sector resources, not Nigerian trading approvals. Treating them as certification is the tempting error; the company still needs its own register and acceptance evidence.
2. A fictional crude exporter's agent sends a payment instruction; the bank interface times out and the agent reports failure. The operator asks whether to resend. What is the right action?
- A. Treat the outcome as unknown and check transaction state before any retry
- B. Resend immediately, because the agent reported failure
- C. Resend with a slightly different amount so it is not a duplicate
- D. Ask the agent whether it succeeded
A. Treat the outcome as unknown and check transaction state before any retry — A timeout leaves the outcome unknown; the first attempt may have succeeded. Resending because the agent said it failed is the trap: a duplicated instruction can create a second commitment.
3. A monitoring dashboard shows input drift on a collection model. What does that establish?
- A. The model is wrong and must be retired
- B. Customers are paying later
- C. Nothing worth investigating
- D. That something changed; the owner decides whether to investigate, recalibrate, restrict or retire
D. That something changed; the owner decides whether to investigate, recalibrate, restrict or retire — Drift detection signals change; it does not establish cause or remedy. Retiring automatically confuses a signal with a diagnosis, and a feature change can occur without harm.
Today’s Assignment
Write Section 13 of your plan: the application register entry for the AI use your decision would rely on in your company, the agent's concrete permissions, and the testing, change, monitoring and fallback rules.
- Write the register entry: intended and excluded use, data sources, evaluation, authority, dependencies, monitoring, change process and recovery, with an owner by role for each.
- List concretely what any agent may and may not do, naming the tools, the stages of the state machine and where human acceptance is required. Say how a timed-out action is handled.
- Name at least three hostile and three ordinary test cases drawn from your company's own documents and workflow.
- List the changes that would trigger renewed review, the fallback staff can operate and who has practised it.
- Explain how a disputed decision could be replayed six months later. Say honestly whether your company could do that today for any current system.
YOUR 90-DAY PLAN
Section 13 — The application register entry, the agent's concrete permissions and the change, monitoring and fallback rules.
How this is marked — the rubric
| Completes every register field with an owner by role, distinguishing the application from the model version | 25 |
| States concrete may and may-not permissions, the state-machine stages and the handling of an unknown outcome | 25 |
| Names hostile and ordinary test cases from the company's own workflow | 20 |
| Sets change triggers, a practised fallback and a credible six-month replay route | 30 |
| Total · pass mark 70 | 100 |
Submitting session 13
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 13, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Pop Quiz 3 — Risk, Treasury and Trust opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Pop Quiz 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Application registerAgent permissionsHostile testingChange controlDrift responseDecision replay
Part Six · Execution and the Leadership ChargeSESSION 14
The Inspectable Investment Case and the Trading Intelligence Team
Why This Matters
By now you know what your decision needs. This session asks two questions your board will ask before it pays for anything. First, is the investment case inspectable: can someone else see the assumptions, the benefit counted once and the change that would make it negative? Second, who will own the result after the people who built it have moved on? Chapter 20 gives you a worked investment decision and a pilot design that can stop a weak case early. Chapter 21 gives you the team. A pilot that reveals an uneconomic use case has done its job. At the end, Practice Exam 3 opens.
25-Minute Module
0:00 – 5:00
Begin with a bounded benefit
- Choose a problem with observed volume, consequence and response capacity: repeated contract omissions, delayed reconciliation, unreliable cash forecasting or late discovery of delivery conflicts. State which decisions the application changes and which outcomes remain outside its influence.
- A time saving does not automatically reduce cash cost. Existing salaried staff may use released capacity elsewhere. Record capacity separately from realised expenditure reduction.
- An avoided-loss estimate depends on assumptions about events that did not occur. Preserve that uncertainty rather than booking every risk reduction as revenue.
- THE DOCTRINE
A time saving does not automatically reduce cash cost. Record capacity separately from realised expenditure reduction, and do not book every risk reduction as revenue.
Ask what would have to happen in the bank account for the benefit to be real.
5:00 – 11:00
A worked investment decision
- Discounting in plain words first. A naira received in three years is worth less to you today than a naira today, because money now can be used or earn a return meanwhile. A discount rate converts future amounts into today's value. Net present value adds up those converted benefits and subtracts the upfront cost. Positive means the assumed benefits exceed the cost in today's terms; negative means they do not.
Chapter 20 assumed case Base Lower improvement Initial cost ₦60 million ₦60 million Annual operating cost ₦20 million ₦20 million Independently accepted gross annual improvement ₦50 million ₦40 million Net annual benefit, three years ₦30 million ₦20 million Three-year annuity factor at an assumed 15% discount rate approximately 2.2832 approximately 2.2832 Simplified NPV positive ₦8.50 million negative ₦14.34 million - Recomputed: ₦30 million × 2.2832 is about ₦68.50 million, less ₦60 million, gives ₦8.50 million; ₦20 million × 2.2832 is about ₦45.66 million, less ₦60 million, gives negative ₦14.34 million. Break-even gross annual improvement is approximately ₦46.28 million under the same assumptions. The model excludes tax, financing, residual value and within-year timing.
- A fall of under ₦4 million a year in gross improvement turns the case negative. The book's conclusion: the investment is sensitive to adoption and the cost of review, not merely the accuracy of the model.
- THE DOCTRINE
A pilot that reveals an uneconomic use case has prevented wider expenditure. The leadership standard is the quality of that decision, not the number of applications launched.
Ask for the break-even number before the NPV. It tells you how much optimism the case needs.
11:00 – 16:00
Count value once and design the pilot
- A better arrival prediction may reduce delay, financing duration and customer disruption. Some effects are distinct; some describe the same outcome twice. Finance approves the boundary and overlap treatment. A benefits register keeps realised, partially supported and projected benefits separate, and separates realised savings, released capacity and hypothetical risk reduction. Charge the full cost: integration, data preparation, specialist review, support, licences, monitoring, correction, incidents and response.
- Define the comparator, scope, technical measures, operating outcomes and rejection conditions before deployment. Then follow the Sequencing Discipline: historical replay on as-of evidence, then shadow use, then a bounded advisory stage. Record all eligible decisions, including recommendations rejected or not acted on.
- Shadow output cannot establish realised benefit, because no action changed. A simple before-and-after result can be affected by different market conditions, customers or staff. Use a credible contemporaneous comparison where feasible, and do not attribute the entire improvement to AI without an accepted basis.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. A fictional distributor tests a depot-level stock intervention. Depots exchange stock through transfers, so a change at one depot can affect another, and splitting the trial invoice by invoice may be inappropriate.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The same trial predeclares its primary economic outcome and compares incremental contribution and cash against the accepted comparator. A lower forecast error or larger gross volume is an intermediate result, not economic evidence on its own.
Write the rejection condition before you see the first result.
16:00 – 21:00
Build the trading intelligence team
- For each consequential application assign an operating owner, a technical owner and an independent acceptance role. A committee can coordinate trading, risk, treasury, operations, legal, credit and finance, but it should not obscure who can accept a model, grant authority or suspend use.
- An assistant can review status reports and decision records for missing owners, unresolved assumptions and incomplete closure, and prepare coaching questions. A manager determines the assessment. Use one review sequence: identify the commitment, compare the accepted baseline, explain changed evidence, record the decision and follow closure.
- Training completion does not establish authority to interpret contracts, approve quality or execute hedges. Assess people on bounded tasks with reference evidence, including ambiguous cases where they must explain when they would escalate. Your company should retain understanding of its trading logic and controls even when external specialists supply models.
- The executive curriculum: read a model specification, an as-of snapshot, a calibration report and a constrained recommendation; recognise an unsupported target, an unavailable action and an unmeasured benefit. Executives do not need to become neural-network researchers to hold those boundaries. Analysts reconstruct one decision end to end and a second analyst reproduces it. Commercial and model specialists review jointly; neither review replaces the other.
A demonstration that depends on the original developer's memory is not an operating capability.
21:00 – 25:00
Write the case, the register and the owners
- Ask each functional leader for one recurring judgement that should remain human and one preparation task AI can improve, connected to the same operating record. Write down yours for your decision.
- Today's assignment produces Section 14 of your plan. Use your own labelled assumptions; where an input is unknown, leave it visibly open rather than choosing a number that makes the case work.
- Practice Exam 3 — Full Syllabus opens now: fifteen questions across Sessions 1 to 14, closed book, pass mark 70.
The case is inspectable when someone who dislikes it can find exactly where to disagree.
Key Terms
Net present value
Future net benefits converted to today's value at a stated discount rate, less the initial cost; in Chapter 20 a simplified figure excluding tax, financing, residual value and within-year timing.
Break-even improvement
The gross annual improvement at which net present value is zero under the same assumptions; about ₦46.28 million in the book's case.
Benefits register
The record that counts each benefit once and separates realised, partially supported and projected value, and capacity from cash.
Independent acceptance role
The person or function, separate from the operating and technical owners, who can accept an application for consequential use.
Self-Check — Before You Submit
1. In Chapter 20's assumed case, what is the simplified NPV if gross annual improvement is ₦40 million?
- A. Positive ₦8.50 million
- B. Negative ₦14.34 million
- C. Zero
- D. Positive ₦20 million
B. Negative ₦14.34 million — Net benefit falls to ₦20 million a year; ₦20 million × 2.2832 ≈ ₦45.66 million, less ₦60 million, is negative ₦14.34 million. Positive ₦8.50 million is the ₦50 million base case.
2. A fictional LPG merchant's pilot shows lower forecast error and trading volume rose in the same quarter, while market conditions also changed. The sponsor wants to book the volume growth as AI benefit. What should you decide?
- A. Book it, because the timing matches
- B. Book half of it as a prudent estimate
- C. Ignore the pilot
- D. Treat both as intermediate results; ask for a contemporaneous comparison and an accepted basis before attributing any contribution
D. Treat both as intermediate results; ask for a contemporaneous comparison and an accepted basis before attributing any contribution — Before-and-after results can be affected by different market conditions; lower forecast error and larger volume are intermediate results. Booking half sounds prudent but still attributes without an accepted basis.
3. Which statement about competence matches Chapter 21?
- A. Competence is demonstrated on bounded tasks with reference evidence; training completion does not establish authority
- B. Training completion establishes authority to execute hedges
- C. External specialists should own the trading logic
- D. Executives must learn to build neural networks
A. Competence is demonstrated on bounded tasks with reference evidence; training completion does not establish authority — The book separates training from authority and keeps trading logic understood internally. The executive bar is reading specifications and recognising unsupported targets, not becoming a researcher.
Today’s Assignment
Write Section 14 of your plan: the bounded benefit your decision could produce, an inspectable investment case from your own labelled assumptions, the benefits-register treatment, the pilot design and the owners who would survive turnover in your company.
- State the bounded benefit your decision could produce and the outcomes outside its influence. Separate capacity from cash.
- Build a simple inspectable investment case from your own labelled assumptions: initial cost, operating cost, gross improvement, horizon and discount rate. Show the sensitivity or break-even that would turn it negative.
- Set the benefits-register treatment: realised, partially supported and projected; and where one benefit could be counted twice.
- Design the pilot: comparator, scope, measures, rejection conditions, and the sequence of replay, shadow and bounded advisory.
- Name by role the operating owner, technical owner and independent acceptance role; one judgement that stays human and one preparation task AI can improve. Say honestly where your company today depends on someone who could not be replaced quickly.
YOUR 90-DAY PLAN
Section 14 — The inspectable investment case, the benefits register, the pilot design and the owners who survive turnover.
How this is marked — the rubric
| States a bounded benefit, the outcomes outside its influence and the capacity-versus-cash distinction | 20 |
| Builds an inspectable case from labelled assumptions with the sensitivity that turns it negative | 30 |
| Sets a benefits register that counts value once and separates realised from projected | 20 |
| Designs a pilot with comparator, rejection conditions and replay-shadow-advisory sequence, and names the three owners by role | 30 |
| Total · pass mark 70 | 100 |
Submitting session 14
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 14, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Practice Exam 3 — Full Syllabus opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Practice Exam 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Bounded benefitInvestment arithmeticBenefits registerPilot designTeam ownershipExecutive model literacy
CapstoneSESSION 15
The Executive Review: Your Signed 90-Day AI Trading Plan
Why This Matters
Fourteen sessions ago you wrote down one live decision and the latest time at which better information could change it. Since then you have built its record, forecast, contract terms, price clock, physical evidence, exposures, cash ladder, controls, governance and investment case. Today those fourteen sections become one signed 90-Day AI Trading Plan: three periods with the evidence each review must see, the Appendix A authority table and the decision records your decision needs. Then you sit the Final Exam.
25-Minute Module
0:00 – 5:00
The first ninety days
- Days one to thirty establish the decision boundary. Inventory the strategies and choose one use case. Confirm scope, evidence owners and the simpler comparator. Reconstruct representative completed trades, including difficult ones, and find the material identity, unit, timing and approval gaps. Create the application register and pilot charter, permissions and a manual fallback; agree reference cases before tuning; confirm capacity to act.
- Days thirty-one to sixty prove the evidence. Build controlled inputs and calculations, run as-of replay and examine failures with the operating owners. Test amendments, stale prices, interrupted tools and missed events; weigh reviewer burden with technical quality. Do not broaden scope because early examples look good. Independent review states what remains unproven.
- Days sixty-one to ninety test the response. Operate in shadow or bounded advisory mode. Preserve recommendation, evidence, human decision and outcome. Review weekly with trading, risk, treasury and operations, and confirm improved information arrives before useful choices close: the Operating-Intelligence Loop against your latest useful time.
- THE DOCTRINE
At day ninety, decide to expand, revise, retain limited use or stop. The deadline is a review discipline, not a promise that every trading model can be validated in three months.
- Rare credit events and extreme market behaviour require a different evidence horizon; say so if your decision depends on them.
Write the evidence each review must see before day one, so no one can choose it after the results arrive.
5:00 – 9:00
The next year and the leadership charge
- Expand first to similar decisions with compatible evidence and controls. Quarterly, compare the portfolio with trading priorities, retire what no longer improves decisions, and keep the benefits ledger and failed cases.
- Select the next model by the evidence the first pilot created, such as accepted clause labels from a contract pilot. Require reproducible baseline, validated challenger, useful advisory, then bounded action. A model approved for one depot has not automatically earned approval for a different energy product.
- THE DOCTRINE
Approve the evidence boundary before approving the model. Examine downside cash before celebrating expected margin. Preserve uncertainty where the facts remain unresolved. Require independent challenge where a consequential decision depends on the application.
- An intelligent trader can explain what it promised, holds, risks and must fund.
Build a dependency chain of evidence, not a collection of disconnected demonstrations.
9:00 – 14:00
Appendix A: who decides, and the six records
- Appendix A is the author's proposed set of decision records. Complete its authority table for your decision.
Decision Accountable function AI contribution Retained approval Trade commitment Trading and commercial leadership Evidence pack and feasible alternatives Delegated commitment authority Exposure and hedge limits Independent risk Reconciliation and anomaly advisory Risk policy and authorised execution Funding and collateral Treasury Dated forecasts and joint stresses Funding and draw approval Cargo and product release Operations and quality Readiness and evidence checks Competent release authorities Credit exception Credit authority Calibrated review ranking Credit decision and conditions Contract interpretation Legal and commercial Retrieved clauses and conflict identification Accepted specialist interpretation Final economics Finance Reconciliation preparation Accounting and benefit acceptance - A1 Trade approval: the trade's terms, fixing state and evidence version; expected and downside contribution, peak cash and credit utilisation; unresolved conditions with owners and deadlines. A2 Daily exception: cut-off, magnitude, evidence status, latest useful action time and owner; open until its closure condition is met.
- A3 Model acceptance: target, horizon, data boundary, untouched test period and comparator; calibration, missed cases, reviewer burden, authority limits, fallback and change triggers, failed cases kept. A4 Cash and collateral: one dated ladder per currency and entity, with the first infeasible date.
- A5 Source and compliance change: a detected change is not complete until the process is updated or an accepted decision says why not. A6 Benefit and learning: baseline, comparator, attribution and overlaps; value realised, partially supported or projected; failures kept; the next decision: continue, revise, restrict, expand or stop.
If a row of the table has no name beside it in your company, that is the first gap your plan closes.
14:00 – 20:00
Appendix C: one fictional cargo from approval to collection
- An entirely fictional 100,000-barrel crude cargo, financed at an assumed simple 12% a year on an ACT/365 teaching basis. Every figure is assumed.
Step Book's assumption Result Purchase cost Five observations averaging $80, plus $2 $8.2 million Sale revenue Five observations averaging $81, plus $3.50 $8.45 million; gross $250,000 Execution costs Freight $90,000; handling and insurance $25,000; other $10,000 $125,000 Financing, outcome $8.2 million for forty-five days $121,315.07 Simplified contribution, outcome $250,000 less $125,000 less $121,315.07 $3,684.93 Approval case Both averages $80; twenty days' funding $53,917.81 negative $28,917.81 Reconciliation Sale window plus $100,000; twenty-five extra days less $67,397.26 net plus $32,602.74, reconciling to $3,684.93 Price fixing (C4) Two open purchase observations rise to $82 average $80.80; cost up $80,000 - THE DOCTRINE
A favourable final number does not rewrite an unsound approval. The approval case was negative; a competent review should reject it, renegotiate or identify separately supported economics.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. In a separate sensitivity, accepted sale quantity is 99,800 barrels. At $84.50, revenue falls $16,900; contribution becomes negative $13,215.07. The 200 barrels are not automatically a physical loss; measurement and commercial review establish cause.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. If the customer pays twenty days late, extra financing of $53,917.81 makes contribution negative $50,232.88 with no price change. Treasury records timing, credit records exposure, finance accepts treatment; the delay is not booked as a default loss unless the defined accounting and risk treatment supports it.
- In C7, contract, formula, arrival, cash and risk services each do their part, and an optimiser compares feasible actions, which may include declining the trade. Authorised reviewers choose. One accepted transaction identity is stronger than a single language model generating the margin narrative.
Ask of every approved trade what the approval case said, not only what the outcome became.
20:00 – 25:00
Assemble, sign and sit the Final Exam
- The Conclusion gives AI an assignment: prepare better evidence, estimate uncertainty, compare feasible choices and support accountable follow-through. Leadership keeps purpose, authority, resources and acceptance.
- Begin with one consequential decision. Establish the record, test the simpler alternative, evaluate the model and observe the human response. Expand when that operating loop earns trust.
- Assemble Sections 1 to 14 under a one-page summary. The approvals in your plan stay with the people authorised to give them.
- 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 Trading Plan
Sections 1 to 14 assembled around one live decision under a one-page summary, signed and dated.
Day-ninety decision
Expand, revise, retain limited use or stop, decided on the evidence; a review discipline, not a promise of validation.
Responsibility and authority table
Appendix A's map of each decision to its accountable function, the AI contribution and the approval that stays with people.
Approval case
The economics expected at commitment, reconciled to the outcome, never rewritten by a favourable result.
Self-Check — Before You Submit
1. In Appendix C, the approval case was negative $28,917.81 and the outcome was positive $3,684.93. What does the book conclude?
- A. The approval was sound
- B. The forecasting model worked
- C. A thin result from an unhedged timing difference, which does not rewrite an unsound approval
- D. Funding was too conservative
C. A thin result from an unhedged timing difference, which does not rewrite an unsound approval — The sale window added $100,000 and extra funding cost $67,397.26. Treating the positive outcome as proof of a sound approval is the error the book warns against.
2. At day ninety, a fictional gas aggregator's credit-default advisory has seen no defaults. The sponsor wants to expand. What do you decide?
- A. Expand, because no defaults were missed
- B. Retain limited use or revise, recording that rare credit events need a longer evidence horizon
- C. Stop, because the pilot proved nothing
- D. Expand to all products at once
B. Retain limited use or revise, recording that rare credit events need a longer evidence horizon — Rare credit events require a different evidence horizon. Expanding treats missing evidence as proof; stopping ignores what the pilot showed about workflow and reviewer burden.
3. In Appendix A, who holds the retained approval for a credit exception?
- A. The credit authority: credit decision and conditions
- B. Treasury: funding and draw approval
- C. The AI model: calibrated review ranking
- D. Trading leadership: delegated commitment authority
A. The credit authority: credit decision and conditions — Calibrated review ranking is the AI contribution, not the approval. The trap is confusing the model's ranking with the credit authority's decision.
Today’s Assignment
Assemble Sections 1 to 14 into one signed 90-Day AI Trading Plan for your company, with a one-page summary, the completed Appendix A table and the decision records your decision needs.
- Assemble all fourteen sections, noting where a later session changed an earlier one.
- Write the one-page summary: decision, owner by role, latest useful time; then Days 1 to 30, 31 to 60 and 61 to 90 with the evidence each review must see.
- Complete the Appendix A table for your decision, with a role, not a department, in every applicable row.
- Draft from your own evidence at least A1 (or its equivalent), A3 and A6. Leave unknown fields visibly open.
- State the day-ninety rule (expand, revise, retain limited use or stop), the evidence for each, and any part needing a longer horizon.
- Name the next model or decision the pilot's evidence would support. Sign, date, and name the commitment you are least confident your business will keep.
YOUR 90-DAY PLAN
The signed 90-Day AI Trading Plan.
How this is marked — the rubric
| Sections assembled, signed and dated; summary names decision, owner and latest useful time | 20 |
| Three periods with the evidence each review must see | 20 |
| Appendix A table by role; A1, A3 and A6 drafted from own evidence, open fields shown | 30 |
| Day-ninety rule with evidence for each outcome and honest horizons | 15 |
| Next model or decision justified by the pilot's evidence | 15 |
| Total · pass mark 70 | 100 |
Submitting session 15
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 15, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.FINAL EXAM
Your Final Exam is sat in this session. 15 questions, of which 3 are rebuilt from questions you got wrong on the practice papers. Sit it once your capstone is submitted.
Sit the Final Exam →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Ninety-day planningAuthority mappingDecision recordsApproval-to-collection reconciliationDay-ninety decisionPlan assembly