15-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s gas and industrial energy businesses
WEEK 3 OF 3 · INTEGRITY, GOVERNANCE AND SCALE · SESSIONS 11–15 · WORK AT YOUR OWN PACE
Chapter 15 · Integrity and Enterprise ExecutionSESSION 11
Measurement and Custody Transfer
Why This Matters
Every invoice your company issues or receives rests on a quantity, and that quantity rests on a chain: instrument, verification, conversion, sometimes composition, calculation version and accepted transaction. If one link is wrong, a precise formula produces a precise wrong invoice. This session teaches you to define the quantity before anyone values it, follow the chain link by link, and keep analytics in the role the book gives it: review, reconciliation and exception detection, never manufacturing accepted measurement evidence. You will finish by reconstructing one quantity your decision depends on, so that a reviewer could reproduce it.
25-Minute Module
0:00 – 5:00
Define the quantity before calculating its value
- Custody transfer is an engineering and commercial responsibility. The measurement chain establishes a quantity on a defined basis; the agreement makes it a billable transaction.
- THE DOCTRINE
AI can support review, reconciliation and exception detection. It should not manufacture accepted measurement evidence.
- The plain idea first. Two records both described as cubic metres can represent different quantities if their conditions differ; the worked conversion below shows how far apart they can be.
- So state which quantity you mean: volume at operating conditions, volume at an agreed base condition, mass or energy. Preserve with it the reference conditions, composition basis, method, time interval and source status.
- Standards have scopes. OIML R 137-1 & 2:2012 addresses gas-meter requirements within its scope and explicitly excludes, among other categories, liquefied-state gases and CNG dispensers (OIML, R 137-1 & 2:2012, section 2). It is not a universal acceptance standard for every gas transaction; the technical owner establishes the applicable arrangement for the actual use.
If two departments quote the same quantity in cubic metres, ask on what conditions before you ask why they differ.
5:00 – 10:00
Follow the measurement chain
- Connect the primary measurement to every link that turns it into a transaction, preserving changed configuration and substitutions.
Link in the chain What must be preserved Primary measurement Instrument identity and the relevant interval Calibration or verification The evidence for the instrument actually used Conversion Method, inputs and reference conditions Composition, where required Analyser status and its relationship to the measured stream Calculation The calculation version applied Accepted transaction Accepted opening and closing readings traceable to the interval - A stale or unrepresentative composition can affect an energy calculation while the volumetric meter appears sound. If another source substitutes for a missing result, retain that fact and its authority.
- ISO 6976:2016 describes calculation of calorific values, density, relative density and Wobbe indices from composition. The book uses only its official scope and does not establish that any facility complies (ISO, 6976:2016, official abstract).
A meter that reads correctly is one link. Ask to see the other five.
10:00 – 17:00
Two worked cases: conditions and energy
- The PTZ conversion (pressure, temperature, compressibility) restates operating volume at base conditions: operating volume × (operating absolute pressure ÷ base pressure) × (base absolute temperature ÷ operating temperature) × (base ÷ operating compressibility factor). OIML R 140 presents this relationship (OIML, R 140:2007, section 7.3.2.3).
Illustrative input (assumed) Value Operating conditions 1,000 cubic metres at 6 bar absolute, 300 K, compressibility factor 0.95 Base conditions 1.01325 bar, 288.15 K, compressibility 1 Result (recomputed) 1,000 × 6/1.01325 × 288.15/300 × 1/0.95 ≈ 5,987 standard cubic metres - These base conditions and compressibility values are assumptions, not a universal contractual basis. Absolute pressure and temperature are required: a gauge-pressure record cannot be inserted as absolute pressure without conversion. The actual method, gas properties, instrument and configuration need competent acceptance.
- Now the energy case, also assumed. Accepted volume of 10,000 standard cubic metres at 38 MJ per standard cubic metre gives 380,000 MJ, or 380 GJ; at an assumed ₦10,000 per GJ, calculated revenue is ₦3.8 million. At an accepted 36 MJ, energy is 360 GJ and revenue ₦3.6 million on the same assumed price: on these assumptions, a difference of 2 MJ per standard cubic metre moves calculated revenue by ₦200,000. This is a teaching example, not a gas-quality specification or a Nigerian tariff.
- THE DOCTRINE
Do not compare an energy-price invoice with a volume-price calculation and classify the difference as missing revenue. A precise formula applied to the wrong commercial unit remains a wrong invoice.
A conversion error travels into the delivery balance, the invoice and the benefit case. Find it at the first link.
17:00 – 21:00
Give analytics its proper role
- Start with explicit checks: units, reference conditions, missing readings, stale composition, duplicate transactions and changed configuration. Only then consider anomaly ranking, where the volume of records justifies it. A learned pattern can suggest investigation; it cannot establish that a quantity is wrong, and an unmatched quantity is not proof of loss or misconduct.
- Evaluate false investigations as well as accepted findings, so that a model cannot appear successful by flagging every transaction. Record whether each finding produced a legitimate correction, a confirmed no-change decision or an unresolved technical question, and keep the original and revised records linked.
- IN THE PHYSICAL SUPPLY · THE WORKED CASE
- Physical supply. At a fictional processing plant's delivery point, the volumetric meter appears sound but the analyser composition used for the period is stale. The energy calculation is affected; the measurement owner decides whether a substitute value applies, and the substitution and its authority are retained.
AT THE CUSTOMER PROMISE · THE PARALLEL- Customer promise. A fictional industrial customer queries the invoice for the same period. Commercial reconstructs the signed basis first: energy or volume, gross or net calorific value, reference conditions.
- Correcting the accepted quantity requires the applicable evidence and authority. It is the same loop, seen from both sides of the delivery point.
Ask how many flagged items ended in a legitimate correction. That number, not the flag count, is the measure.
21:00 – 25:00
Reconstruct one billable quantity
- Choose one transaction that bears on your decision and trace it from instrument evidence to accepted quantity, energy where relevant, agreed price and invoice. Then ask a reviewer to reproduce it without the preparer's explanation.
- Resolve ambiguous conversions or missing versions before automating any review. If your company buys gas rather than sells it, reconstruct a quantity you were billed for: the same chain, read from the customer side.
- THE DOCTRINE
A model-generated estimate should remain an estimate until the governing process accepts it. Claimed cash recovery needs accepted transaction and collection evidence.
- Better measurement control may show as fewer disputes, less reconstruction or corrected billing: three different claims, each with its own evidence.
- Before you close the laptop, write down the quantity your decision depends on and the link in its chain you are least sure of. Today's assignment produces Section 11 of your plan.
If the reviewer needs you in the room to reproduce it, the record is not yet complete.
Key Terms
Custody transfer
The handover of a measured quantity as the basis of a billable transaction; an engineering and commercial responsibility that analytics may support but not replace.
Reference (base) conditions
The agreed pressure and temperature at which a volume is stated. Two cubic-metre records on different conditions are different quantities.
PTZ conversion
Restating operating volume at base conditions using absolute pressure, absolute temperature and compressibility; OIML R 140 presents the relationship.
Measurement chain
The links from primary measurement through instrument identity, verification, conversion, composition and calculation version to the accepted transaction.
Self-Check — Before You Submit
1. A fictional pipeline team enters a gauge-pressure reading directly into the PTZ conversion as operating pressure. What is wrong?
- A. The relationship requires absolute pressure; a gauge record cannot be inserted without the appropriate conversion
- B. Nothing, provided the compressibility factor is correct
- C. Only temperature needs to be absolute
- D. The base pressure should be changed to match the gauge
A. The relationship requires absolute pressure; a gauge record cannot be inserted without the appropriate conversion — The book requires absolute pressure and absolute temperature. A correct compressibility factor does not repair a wrong pressure basis, and the error travels into the balance, invoice and benefit case.
2. Your analyst compares a customer's energy-price invoice with an internal volume-price calculation and reports the difference as missing revenue. What do you do first?
- A. Issue a supplementary invoice for the difference
- B. Ask a model to rank the largest discrepancies
- C. Reconstruct the signed commercial basis before classifying any difference
- D. Report it to the board as a recovery opportunity
C. Reconstruct the signed commercial basis before classifying any difference — Reconstruct the signed basis first: a precise formula applied to the wrong commercial unit remains a wrong invoice. Ranking discrepancies only ranks a comparison that may itself be invalid.
3. What does the book say about the scope of OIML R 137-1 & 2:2012?
- A. It is a universal acceptance standard for every gas transaction
- B. It explicitly excludes, among other categories, liquefied-state gases and CNG dispensers
- C. Citing it establishes that a facility complies
- D. It sets the Nigerian tariff basis for gas measurement
B. It explicitly excludes, among other categories, liquefied-state gases and CNG dispensers — The book cites section 2 for the exclusions and warns against using the scope as a universal acceptance standard. Treating a citation as proof of compliance is the trap the book names.
Today’s Assignment
Write Section 11 of your plan: define the quantity your decision depends on, follow its measurement chain and reconstruct one accepted quantity in your company so that a reviewer could reproduce it.
- Define the quantity your decision depends on: operating volume, base volume, mass or energy, with reference conditions, composition basis, method, time interval and source status.
- Follow its measurement chain from instrument to accepted transaction: identity, verification evidence, conversion, composition where required, calculation version and substitutions. Name by role the technical owner who establishes the arrangement; claim no compliance with any standard.
- Reconstruct one billable or accepted quantity step by step on the agreed commercial basis, and say whether a reviewer could reproduce it without you. If not, say exactly what is missing.
- List the explicit checks that must precede any anomaly ranking, and name by role who authorises a correction and how original and revised records stay linked.
- Be honest about where your company compares quantities on different bases today. If no AI is in use, work from the current reconciliation; an industrial energy user works from the quantities your business is billed for.
YOUR 90-DAY PLAN
Section 11 — The quantity defined, its measurement chain followed and one accepted quantity reconstructed.
How this is marked — the rubric
| Defines the quantity with basis, reference conditions, method, interval and source status | 25 |
| Follows the chain to the accepted transaction, names the technical owner and claims no compliance | 25 |
| Reconstructs one quantity reproducibly, or states precisely what prevents it | 30 |
| States the checks before anomaly ranking and who authorises a correction | 20 |
| Total · pass mark 70 | 100 |
Submitting session 11
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 11, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Quantity definitionMeasurement chainConversion literacyEnergy versus volumeException reviewCorrection authority
Chapters 16 and 17 · Integrity and Enterprise ExecutionSESSION 12
Asset Integrity, Maintenance and Spares
Why This Matters
A warning that cannot be acted on in time protects nothing, and a safety pack that reads as reassuring can hide the one control that is not in place. This session covers the integrity, spares and safety evidence on which your decision may silently depend. Chapter 16 links a concern to its failure consequence and the response actually available, then follows the response chain to its longest controllable delay. Chapter 17 sets the boundary for information support and the controlled change any AI application needs, even one that never writes to a controller. These decisions remain with competent, authorised owners.
25-Minute Module
0:00 – 5:00
Integrity is a controlled technical judgement
- Integrity decisions concern whether equipment can continue to fulfil its intended duty within accepted conditions. AI can help organise evidence, identify patterns or prioritise review.
- THE DOCTRINE
AI does not become an inspection certificate, fitness assessment or maintenance authority through fluent output.
- Comparison needs a stable equipment hierarchy linking facility, system, equipment and component, with service, duty, configuration, inspection, maintenance, failure, restriction and accepted readiness. Free-text work orders without these identities are hard to compare dependably.
- ISO 14224:2016 provides a basis for collecting and exchanging reliability and maintenance data for equipment in petroleum, petrochemical and natural-gas industries. Its official abstract supports the need for structured data; it does not establish compliance or fitness for a particular asset (ISO, 14224:2016, official abstract).
When a summary sounds certain about equipment, ask who signed the judgement it rests on.
5:00 – 11:00
Prioritise consequence and the response available
- Link each concern to the affected duty and system consequence. A cheap component may be critical because its absence prevents an accepted delivery; a costly item may have redundancy. Commercial materiality and technical criticality are related but distinct.
- Record the task type: inspection, diagnosis, preventive work, corrective repair, or a statutory or accepted technical requirement. A model ranking cannot defer necessary work because another task has greater forecast contribution.
- Keep work status granular: requested, approved, funded, procured, received, installed, tested and accepted are different states. A report that treats receipt of a part as restored equipment readiness overstates the response capability.
- For spares, use accepted part identity, applicability, quantity, condition, storage and availability. A material code does not establish compatibility; an apparent alternative needs engineering acceptance before the plan treats it as available, especially where textual similarity proposed it.
- Record lead time in components (approval, order, production or availability, transport, receipt inspection and acceptance), not one optimistic total. An estimated dispatch date is not the date the part becomes usable on the asset.
Ask which state your last critical repair was really in when the report called it done.
11:00 – 16:00
Worked example: stock or expedite
Item (fictional case, assumed) Calculation Result Annual holding cost of a compatible spare ₦8 million × 18% ₦1.44 million Exposure hours that would create unrecoverable loss 60 hours × 30% 18 hours Protected contribution 18 hours × ₦100,000 ₦1.8 million Stocking, before other costs and consequences ₦1.8 million − ₦1.44 million ₦360,000 a year Accepted expedite arrangement Expected annual cost ₦900,000 - If the expedite arrangement provides equivalent availability, it may be preferable in this simplified comparison. If it cannot meet a necessary integrity or service condition, it is not an equivalent alternative. The ₦8 million purchase also needs funding: holding cost is not the full initial cash outlay.
- THE DOCTRINE
Spare decisions require compatibility, availability, consequence and financing, rather than a general criticality score alone. The teaching example does not recommend a maintenance interval or stock level.
- IN THE PHYSICAL SUPPLY · THE WORKED CASE
- Physical supply. At a fictional compression station, engineering establishes the spare's compatibility and the supply owner the expedite arrangement's real lead-time components. Until the part is installed, tested and accepted, readiness has not been restored.
AT THE CUSTOMER PROMISE · THE PARALLEL- Customer promise. Finance sees the same decision as an ₦8 million funding call and a protected contribution that counts only hours that would otherwise be unrecoverable. If a delivery can be recovered later, its contribution may be delayed rather than permanently lost.
- Evaluate a maintenance recommendation on missed important tasks, unnecessary tasks, specialist review burden and accepted execution. Never train on later work-order closure while claiming an earlier risk prediction: the model may learn the closure workflow, not deterioration.
Confirm the response capability before you select the cheaper number.
16:00 – 21:00
Safety evidence and controlled change
- IOGP's process-safety work addresses major-event prevention and the use of indicators to examine performance. It is an industry reference, not a replacement for applicable Nigerian obligations or the company's accepted safety arrangements (IOGP, Process Safety, official workstream page).
- A completed form can be useful evidence of a check, but it does not by itself establish that the barrier worked. An exception assistant can flag missing or expired evidence with a rules layer and summarise records with source links. It should never invent a test result, close a defect or turn a confidence score into an operational permission.
- The reassuring summary, a fictional review. A monthly pack states that all critical controls are current. One inspection task was closed when its paperwork moved department, while the required test remains outstanding, and a document assistant repeats the closure. The defect lies in the definition of closure; a larger model will not reliably correct a record that represents the wrong state.
- THE DOCTRINE
An application can change operation without writing to a controller. Consider incorrect, absent and delayed output, and a user accepting it without adequate review.
- A material model, source, prompt, threshold or tool-permission change may change the accepted use: establish which changes require re-evaluation and who releases the version. In emergency readiness, a plausible generated instruction is not a substitute for an approved procedure.
If the pack shows fewer red statuses this month, ask why they changed before you accept the improvement.
21:00 – 25:00
Close one response chain
- Select the maintenance or integrity issue that bears on your decision. Trace the warning, accepted task, compatible part, funding, execution, test and readiness, and identify the longest controllable delay.
- Fund the capability at that delay: better part records, an accepted supply arrangement, a different approval route or an analytical advisory. Improve the complete response, not the number of alerts.
- Then reconstruct one safety-critical status your decision relies on: trace it to the physical evidence, approval and current restriction, and have the relevant owner review the trace. A trustworthy system can make the pack less reassuring, because it makes the unresolved condition visible.
- Before you close the laptop, write down the state in which your response chain most often stalls. Today's assignment produces Section 12 of your plan.
Never count fewer warnings as an improvement until you know why they disappeared.
Key Terms
Equipment hierarchy
A stable structure linking facility, system, equipment and component, so that service, failure, restriction and readiness records can be compared dependably.
Longest controllable delay
The slowest step in a response chain that management can change; where funding a capability improves the whole response.
Lead-time components
Approval, order, production or availability, transport, receipt inspection and acceptance, recorded separately rather than as one optimistic total.
Controlled change
Review of an application's operating effect, failure modes, fallback and approvals, and of which changes require re-evaluation.
Self-Check — Before You Submit
1. In the book's assumed stock-or-expedite case, what is the simplified annual result of stocking the spare, before other costs and consequences?
- A. ₦1.8 million
- B. ₦1.44 million
- C. ₦360,000
- D. ₦900,000
C. ₦360,000 — Protected contribution of ₦1.8 million (18 unrecoverable hours × ₦100,000) less ₦1.44 million holding cost gives ₦360,000. Taking ₦1.8 million as the result forgets the holding cost, and the ₦8 million purchase still needs funding.
2. A fictional monthly safety pack reports all critical controls current, but one inspection task was closed when its paperwork moved department, with the required test outstanding. The vendor proposes a larger model. What is the right response?
- A. Correct the closure transaction, establish the actual barrier status through the responsible authorities, and add this failure mode to the application test
- B. Adopt the larger model so the summary becomes more accurate
- C. Accept the pack, because the form was completed
- D. Ask the assistant to regenerate the summary
A. Correct the closure transaction, establish the actual barrier status through the responsible authorities, and add this failure mode to the application test — The defect is in the definition and evidence of closure, so a larger model will not reliably correct it. A completed form does not by itself establish that the barrier worked.
3. A maintenance-planning assistant has read-only access and never writes to a controller. Does it need controlled change?
- A. No, read-only access removes the need
- B. Yes: it can alter maintenance priorities, response speed, task sequence or the interpretation of a warning, so its operating effect must be assessed
- C. Only if the vendor changes the model
- D. Only once it has been certified
B. Yes: it can alter maintenance priorities, response speed, task sequence or the interpretation of a warning, so its operating effect must be assessed — Chapter 17 says assess the operating effect, not just read-only access. Waiting for a vendor change ignores the incorrect, absent and delayed outputs the application can produce today.
Today’s Assignment
Write Section 12 of your plan: link your company's decision to its failure consequence and available response, close its response chain and write the controlled-change review the proposed application would need.
- Link your decision to its failure consequence and the response actually available: affected duty, system consequence, task type, and the conditions not available for economic trade-off.
- Set out the work states your response chain passes through, from requested to accepted, with lead-time components. Identify the longest controllable delay and the capability that needs funding there.
- State the spares or expedite question your decision raises in four terms: compatibility, availability, consequence and financing. Use your company's figures only where evidenced; otherwise mark them as assumptions.
- Write the controlled-change review: intended use, affected users, consequences of incorrect, absent and delayed output and of unreviewed acceptance, fallback, and who releases a version. Name by role the integrity, safety and technical owners who would review it.
- Be candid about one status your business reports as complete on paperwork alone. Write no procedures, limits or inspection results. If no AI is in use, apply this to the proposal in front of you.
YOUR 90-DAY PLAN
Section 12 — Consequence and available response, the response chain closed and controlled change for the application.
How this is marked — the rubric
| Links the decision to failure consequence and available response, with conditions outside trade-off | 20 |
| Sets out work states and lead-time components and identifies the longest controllable delay | 30 |
| States the spares question in terms of compatibility, availability, consequence and financing, assumptions labelled | 20 |
| Writes a controlled-change review covering incorrect, absent, delayed and unreviewed output, with owners by role | 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
Integrity evidenceResponse-chain diagnosisSpares economicsBarrier evidence reviewControlled change
Chapters 18 and 19 · Integrity, Governance and ScaleSESSION 13
Data Foundations and Cybersecurity
Why This Matters
Many AI applications fail quietly at the joins: a stale feed, an equipment name that does not match across systems, a service account with more authority than any user, a contract that buys features rather than an outcome. This session gives you the controls beneath your decision. From Chapter 18: a data contract written for the decision, three application boundaries, evaluation built from real questions and a test of what happens when the service fails. From Chapter 19: buying an accepted capability, writing acceptance into the contract, pricing the full operating cost and naming who holds each decision right. Then Pop Quiz 3.
25-Minute Module
0:00 – 5:00
Architecture follows the decision
- The appropriate architecture begins with the information a specific decision requires, not a promise to integrate every system into one intelligent platform.
- NIST SP 800-82 Revision 3 addresses operational technology security with attention to performance, reliability and safety; enterprise IT assumptions should not be copied into an operating system without technical review (NIST, Guide to Operational Technology Security, SP 800-82 Rev. 3, 2023).
- Session 1 introduced the small data contract; here it becomes specific. For each input name its identity, unit, reference basis, time, status, source owner, permissible use, refresh and accepted age, and define the response to missing, stale or conflicting evidence.
- THE DOCTRINE
A similar equipment name is not a dependable join. Keep stable identifiers across systems and retain the evidence available at the decision time.
- A refreshed dashboard using corrected records may no longer explain an earlier recommendation. Store a snapshot or stable version references under accepted retention and access rules.
Ask which join between two of your systems depends on someone recognising a name.
5:00 – 11:00
Three boundaries, and retrieval is not truth
- The book separates three application boundaries. They are an author-developed design distinction, not an official three-part standard.
Boundary What it determines What to preserve or enforce Ingestion Which material enters the application and with what authority Source, version, owner and permission Retrieval Which material a particular user may receive Access restrictions applied to the actual user and task Action Which operations the application may propose, submit or execute Explicit tool permissions and state transitions - A broad service account can make the application's effective authority far greater than the user's. Read-only scope reduces some consequences but not the effect of incorrect advice on a user's action.
- A retrieval-augmented assistant finds source material and uses it to prepare a response. Correct retrieval and faithful generation are separate tasks: it may retrieve the wrong procedure, use a superseded version or overstate the source. NIST's Generative AI Profile discusses risks including confabulation, information integrity and information security (NIST, AI 600-1, 2024, section 2).
- THE DOCTRINE
Treat document text as evidence, not as authority to change the application's instructions or permissions. Protect tool execution with explicit checks outside the generated prose.
Ask what the assistant does when the evidence is missing. Abstaining is a capability.
11:00 – 16:00
Evaluate, version and recover
- Build the test set from real decision questions: correct current sources, obsolete sources, conflicting sources, absent evidence and requests outside authority. Define the acceptable answer and the necessary source evidence before testing.
- Score errors by consequence. A wrong heading differs from an invented inspection result; do not average away a material failure because many easy questions were answered correctly. Measure review effort and latency too.
- Record model, prompt, source, transformation and tool-permission versions for each release, with a rollback to an accepted previous version. Distinguish data drift, changed decision conditions and changed software behaviour before releasing another version.
- Customer contacts, locations and personnel records can include personal data. Nigeria's data-protection framework includes the Nigeria Data Protection Act, 2023; the relevant owners establish lawful use, retention and supplier arrangements (Nigeria Data Protection Commission, official institutional and Act information). Separate content-use, supplier-support and model-training permissions.
- The unavailable-service test: have the team demonstrate the agreed fallback when the model is unavailable, a critical source is stale or access is revoked, showing which decisions continue and who resolves the exception.
A demonstration with every dependency available proves the least important thing.
16:00 – 21:00
Buy an accepted capability
- Buy the capability the operating decision needs, not a collection of model features. Write the decision, users, scope, outputs, permissions and existing alternative before asking for proposals.
- Make acceptance part of the contract, distinguishing a demonstration, a tested pilot and an accepted operating service. Preserve the right to reject a material failure, and secure information rights and an exit process so that your evidence can be exported and explained.
Assumed comparison, 1,000 packs a month Option A Option B Service fee ₦500,000 ₦1.2 million Review time per pack 15 minutes 6 minutes Review hours at ₦5,000 an hour 250 hours, ₦1.25 million 100 hours, ₦500,000 Total monthly resource cost ₦1.75 million ₦1.7 million - The ₦50,000 difference is not automatically a cash saving: if salaried reviewers remain and no expenditure changes, it may be released capacity. If B fails a necessary source-integrity condition, its advantage is irrelevant; if the real review reduction is smaller, the ranking may reverse.
- IN THE PHYSICAL SUPPLY · THE WORKED CASE
- Physical supply. For a fictional compressor-group advisory, acceptance concerns identified episodes, useful lead time and review burden within a defined regime. It cannot write to the controller; any later integration needs separate technical and security acceptance.
AT THE CUSTOMER PROMISE · THE PARALLEL- Customer promise. The supplier's guarantee may cover service availability, while the investment depends on customer acceptance and commercial execution. The guarantee is not the company's realised benefit.
- Use a decision-rights record naming proposer, technical checker, commercial approver, releaser, executor and outcome reviewer; an application account should not silently combine them. The Nigerian Oil and Gas Industry Content Development Act, 2010, with NCDMB's official overview, is examined through responsible owners; the book infers no approval or statutory percentage.
A low per-call price can hide the most expensive line in the budget: human verification.
21:00 – 25:00
The reviewable procurement brief
- Attach the use-case screen, pilot acceptance gates and response dependencies to the request. Require proposals to state exclusions and assumptions in the same decision terms, and obtain technical, information, commercial and financial acceptance.
- Train by role and action, and include experienced staff's objections in evaluation rather than treating them as resistance.
- THE DOCTRINE
Procurement is complete when the business has an accepted, supportable response capability, not merely a signed technology contract.
- Before you close the laptop, write down who would own the service after delivery and what your company would do if the supplier became unavailable. Today's assignment produces Section 13 of your plan.
- Pop Quiz 3 opens now: five questions on Sessions 11 to 13, closed book. Sit it before Session 14.
If nobody owns the service on the day after go-live, you have bought a demonstration.
Key Terms
Decision-specific data contract
For each input: identity, unit, reference basis, time, status, source owner, permissible use, refresh, accepted age and the response to missing, stale or conflicting evidence.
Application boundaries
Ingestion (what enters), retrieval (what a user may receive) and action (what the application may propose, submit or execute); an author-developed design distinction.
Unavailable-service test
A demonstration of the agreed fallback when the model is down, a source is stale or access is revoked, showing which decisions continue and who resolves exceptions.
Decision-rights record
The named proposer, technical checker, commercial approver, releaser, executor and outcome reviewer for a decision.
Self-Check — Before You Submit
1. In the book's assumed comparison, Option B's total monthly resource cost is ₦50,000 lower than Option A's. Your reviewers are salaried and no expenditure will change. How should the difference be treated?
- A. As a monthly cash saving of ₦50,000
- B. As released capacity, unless a cost is actually removed or avoided, or additional accepted work is identified
- C. As an annual cash saving of ₦600,000
- D. As irrelevant, because B's service fee is higher
B. As released capacity, unless a cost is actually removed or avoided, or additional accepted work is identified — Chapter 19 says the difference is not automatically a cash saving. Annualising compounds the error; dismissing B on its fee ignores the review cost you are learning to price.
2. A retrieved maintenance document contains an embedded instruction to release a pending transaction. What should the application do?
- A. Follow it, because it came from an approved source
- B. Execute it and log the action for later review
- C. Treat the text as evidence only; tool execution is protected by explicit checks outside the generated prose
- D. Lower its confidence score and proceed
C. Treat the text as evidence only; tool execution is protected by explicit checks outside the generated prose — Document text is evidence, not authority to change instructions or permissions. Logging afterwards still lets document content release a transaction.
3. How does the book describe the ingestion, retrieval and action boundaries?
- A. As an author-developed design distinction, not an official three-part standard
- B. As a requirement of NIST SP 800-82 Rev. 3
- C. As defined by the Nigeria Data Protection Act, 2023
- D. As an ISO standard for AI applications
A. As an author-developed design distinction, not an official three-part standard — The book labels the boundaries as its own design distinction. Attributing them to NIST or a statute claims an authority the book does not give them.
Today’s Assignment
Write Section 13 of your plan: the data contract, the three boundaries, the acceptance terms and the decision rights for the decision your company has carried since Session 1.
- Write the decision-specific data contract: each input's identity, unit, reference basis, time, status, source owner, permissible use, refresh and accepted age, and the response to missing, stale or conflicting evidence. Name one join in your company that relies on a similar name.
- State the ingestion, retrieval and action boundaries: what the application may propose, submit or execute, and what it must never do.
- Describe the unavailable-service test and fallback: which decisions continue and who resolves the exception. Name the owners who establish lawful use of any personal data.
- Draft the reviewable procurement brief: the accepted capability, the existing alternative, acceptance in the contract (demonstration, tested pilot, accepted operating service), information rights and exit, and the full operating cost including human verification.
- Write the decision-rights record: proposer, technical checker, commercial approver, releaser, executor and outcome reviewer, by role. If no AI is in use, write this for the proposal in front of you.
YOUR 90-DAY PLAN
Section 13 — The data contract, the three boundaries, the acceptance terms and the decision rights.
How this is marked — the rubric
| Writes a complete data contract with a response to missing, stale or conflicting evidence | 25 |
| States the three boundaries and an unavailable-service test with fallback and exception owner | 25 |
| Drafts a brief that buys an accepted capability, contracts acceptance and prices the full operating cost | 30 |
| Names all six decision rights by role without combining them in an application account | 20 |
| 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 — Measurement, Integrity and Enterprise Controls 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
Data contractsApplication boundariesRetrieval evaluationService resilienceCapability procurementDecision rights
Chapter 20 · Governance and ScaleSESSION 14
Investment Cases and Verified Benefits
Why This Matters
An investment committee does not need a bigger number. It needs to know what would have happened anyway, which cash flows change, what happens on the downside, whether the company can fund the gap before benefits arrive, and how the claim could be shown to be wrong. This session teaches the counterfactual and incremental case from Chapter 20, the benefit ledger from Appendix F and the decision-terms evaluation from Appendix H. Every figure here is the book's assumed teaching case, not a result. Then Practice Exam 3.
25-Minute Module
0:00 – 5:00
Value begins with a counterfactual
- The counterfactual is what would plausibly happen without the intervention: the existing rule, a simpler automation, an equipment repair or another feasible alternative. Value is measured against that alternative.
- THE DOCTRINE
Separate a demonstrated technical improvement from an estimated business opportunity. A model may identify a condition accurately while the actual avoidability of loss remains uncertain.
- The investment can still proceed as a bounded pilot, with the uncertainty explicitly priced and tested.
- Build the incremental cash flows: currency, horizon, timing, initial expenditure, recurring and response cost, working capital and tax or terminal assumptions, each marked measured, assumed or estimated. Keep accounting contribution and project cash flow distinct, use a consistent nominal or real basis, and let finance set the valuation basis so that financing is not counted twice.
Ask what the existing rule would have done. If nobody knows, the baseline is the first deliverable.
5:00 – 11:00
Worked NPV, downside and the funding trough
- The plain idea of net present value (NPV): money received later is worth less than money today, so each future amount is divided down by the discount rate for each year of waiting, and the initial spend is subtracted. A positive result clears that rate on its own assumptions, nothing more.
- The book's assumed case: ₦30 million initial expenditure; ₦26 million annual gross benefit less ₦12 million recurring cost gives ₦14 million net at each year-end for three years; 20% discount rate; no tax, terminal value or extra working capital. NPV = −30 + 14/1.20 + 14/1.20² + 14/1.20³ (₦ million) ≈ negative ₦509,259.
Scenario (assumed) Annual net cash Approximate NPV Downside ₦10 million Negative ₦8.935 million Base ₦14 million Negative ₦0.509 million Upside ₦18 million Positive ₦7.917 million Break-even About ₦14.242 million Zero - Undiscounted payback is about 2.14 years, yet the discounted case is negative. If initial cost were US$20,000, an illustrative ₦1,500 per dollar gives ₦30 million; ₦2,000 gives ₦40 million and, at ₦14 million net, an NPV of approximately negative ₦10.509 million. These exchange rates are scenario assumptions, not current rates or forecasts.
- THE DOCTRINE
A positive NPV does not establish that the company can fund the project before benefits arrive. Review the funding trough.
Ask which single assumption would flip the sign. That is what the pilot must test.
11:00 – 16:00
One effect, counted once
- Contribution, avoided cost, released working capital, accelerated cash, released capacity and reduced exposure are different mechanisms. Released inventory is not an annual profit saving; faster invoicing does not automatically increase collection.
- One event can have several consequences. An early warning, a changed schedule and a faster spare may all support the same protected delivery: allocate that outcome once.
- Appendix F keeps a benefit ledger by mechanism and period, separating potential, provisional and accepted results. Its illustrative entry: 40 hours × 8,000 units an hour × NGN120 = NGN38.4 million gross; at an assumed 50% avoidable, NGN19.2 million; less NGN12 million cost, NGN7.2 million net; at 20% avoidability, negative NGN4.32 million. It repeats the Chapter 5 case from Session 5 so the ledger traces to its explanation. It is not a second benefit.
- Expected avoided loss is an estimate, not a realised receipt. Claims move from opportunity to estimated, provisional, accepted or rejected as evidence warrants. Finance reviews time basis, currency, attribution and duplication; the operating owner accepts whether the response changed the physical result.
If the same delivery appears in three benefit lines, ask which one owns it.
16:00 – 21:00
Evaluate in decision terms
- Begin with a falsifiable proposition: decision, comparator, intervention, outcome and boundary, stated so it can fail. Specify the unit of evaluation (thousands of sensor samples around one event are not thousands of faults) and replay only the evidence available at the time.
Assumed teaching counts: 200 episodes, 10 faults Existing process Proposed model Relevant faults correctly flagged 5 8 Non-fault episodes incorrectly flagged 10 12 Relevant faults missed 5 2 Recall for relevant faults 50% 80% Precision among flags 33.3% 40% Overall classification accuracy 92.5% 93% Total flags requiring review 15 20 - Flagging nothing would score 95% accuracy and miss every fault, so accuracy cannot carry a rare-event decision. On the book's second assumption, three additional episodes support a feasible response: 3 × NGN600,000 = NGN1.8 million, less five extra reviews at NGN80,000 (NGN400,000) and NGN1 million application cost, leaves NGN400,000 provisional net potential. At NGN400,000 per episode it is negative NGN200,000. This is not realised savings.
- Forecast error and operating loss can rank models differently: two assumed forecasts with equal mean absolute error of ten units cost NGN2,400 (A) and NGN800 (B) when under-forecasting costs NGN100 a unit and over-forecasting NGN20.
- IN THE PHYSICAL SUPPLY · THE WORKED CASE
- Physical supply. A fictional compressor advisory runs in shadow operation beside the accepted procedure, revealing evidence failures, infeasible actions and review effort. Required protection and maintenance are never withheld to manufacture a control group.
AT THE CUSTOMER PROMISE · THE PARALLEL- Customer promise. The committee receives the same evidence as provisional potential, not realised benefit: shadow operation cannot fully establish realised benefits because the actions were not carried out.
Proceed for this scope, with these controls and these unresolved questions, is a stronger conclusion than a claim that the technology works.
21:00 – 25:00
The investment committee challenge
- Require the sponsor to explain the decision, counterfactual, economic mechanism, feasible response, evidence, downside and funding in one coherent case. Ask which assumption would change the recommendation and what the next expenditure will resolve.
- Compare with a credible alternative: a model that beats an outdated manual process may still lose to a simple rule or a record repair. The board receives failed assumptions as well as successes; approval names scope, budget, conditions, review trigger and authority.
- THE DOCTRINE
An attractive spreadsheet becomes an investment only when it is connected to accepted operating capability and an inspectable financial record.
- Before you close the laptop, write the counterfactual for your decision in one sentence. Today's assignment produces Section 14 of your plan.
- Practice Exam 3 opens now: fifteen questions across the full syllabus, Sessions 1 to 14. Sit it before Session 15.
A revised or stopped investment can protect capital. Say so in the room.
Key Terms
Counterfactual
What would plausibly happen without the intervention: the existing rule, a simpler automation, a repair or another feasible alternative.
Funding trough
The cash the company must find before benefits arrive; a positive NPV does not show it can be funded.
Benefit ledger
A record of claimed benefits by mechanism and period, with evidence state, attribution, overlap checks and reviewer, so each effect enters the total once.
Falsifiable proposition
A statement of decision, comparator, intervention, outcome and boundary that the evaluation could prove wrong.
Self-Check — Before You Submit
1. In the book's assumed case, undiscounted payback is about 2.14 years. What is the NPV at a 20% discount rate over the three-year horizon?
- A. Approximately positive ₦12 million
- B. Approximately negative ₦509,259
- C. Approximately positive ₦7.917 million
- D. Zero, because payback falls within the horizon
B. Approximately negative ₦509,259 — −30 + 14/1.20 + 14/1.20² + 14/1.20³ (₦ million) ≈ negative ₦509,259. Payback inside the horizon ignores the time value of money; ₦7.917 million is the upside case.
2. A sponsor presents the Appendix F ledger entry and the Chapter 5 compressor case as two separate benefits and adds them together. What do you say?
- A. Accept both, because each is documented
- B. Keep both but halve the total as a prudent allowance
- C. They are the same case repeated so the ledger traces to its explanation; count it once, and it remains a teaching calculation with no realised result
- D. Replace both with the upside NPV
C. They are the same case repeated so the ledger traces to its explanation; count it once, and it remains a teaching calculation with no realised result — Appendix F says the calculation repeats the Chapter 5 case and is not a second benefit. Halving sounds prudent but still double counts one effect.
3. An evaluation team proposes switching off a required protection on one compressor to create a clean control group. What does Appendix H say?
- A. Never withhold required protection or necessary maintenance to manufacture a control group; use matched decisions, phased introduction or unaffected activities where acceptable
- B. It is acceptable for the pilot period only
- C. It is acceptable if engineering is informed
- D. It is required for valid attribution
A. Never withhold required protection or necessary maintenance to manufacture a control group; use matched decisions, phased introduction or unaffected activities where acceptable — Appendix H allows matched decisions, phased introduction or contemporaneous unaffected activities where operationally acceptable. Informing engineering does not make withholding required protection acceptable.
Today’s Assignment
Write Section 14 of your plan: the counterfactual, the incremental case with its downside and funding trough, the benefit ledger and a falsifiable evaluation for your company's decision, ready for the investment committee challenge.
- State the counterfactual for your decision: the existing rule, a simpler automation, a repair or another feasible alternative.
- Build the incremental cash flows with each assumption marked measured, assumed or estimated, a downside case and the funding trough. Use your finance owner's discount rate; the book's 20% is a teaching assumption.
- Enter each claimed benefit in a ledger by mechanism, with its evidence state and an overlap check showing no effect is counted twice.
- Write a falsifiable proposition with its unit of evaluation, comparator (existing process and a simpler alternative) and replay of the evidence available at the time. Say what result would make you stop.
- Answer the investment committee challenge in writing, naming the assumption that would change your recommendation and one benefit your business once counted as cash that was really capacity or timing. If no AI is in use, build the case for the proposal in front of you.
YOUR 90-DAY PLAN
Section 14 — The counterfactual, the incremental case, the benefit ledger and a falsifiable evaluation.
How this is marked — the rubric
| States a credible counterfactual and incremental cash flows with assumptions labelled, a downside and the funding trough | 30 |
| Enters benefits in a ledger by mechanism with evidence state and overlap check | 25 |
| Writes a falsifiable proposition with unit, comparator and replay of evidence available at the time | 25 |
| Answers the committee challenge, naming the assumption that would change the recommendation | 20 |
| Total · pass mark 70 | 100 |
Submitting session 14
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 14, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Practice Exam 3 — Full Syllabus opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Practice Exam 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Counterfactual reasoningIncremental cash flowsNPV and downsideBenefit ledgerDecision-terms evaluationCommittee challenge
CapstoneSESSION 15
The 90-Day Review: Your Signed 90-Day AI Leadership Plan
Why This Matters
Fourteen sessions ago you chose one real operating decision that affects a customer promise and costs your company time, money or service. Since then you have defined its promise, found its constraint, followed its quantity, closed its response chain, written its data contract and tested its investment case. Today you assemble those fourteen sections into one signed 90-Day AI Leadership Plan, built on the book's decision register and pilot charter and laid out in Chapter 21's four stages. The plan must be able to produce a decision on day ninety: continue, expand, revise or stop. Then you sit the Final Exam.
25-Minute Module
0:00 – 4:00
Ninety days should establish an operating capability
- The first 90 days should connect a useful decision with evidence, people, authority and an accepted review. The result may be a bounded deployment, a revised pilot or a decision to stop.
- THE DOCTRINE
Ninety days should not be measured only by workshops, demonstrations and licence activation. Choose one material problem with a feasible response.
- Good candidates have an accountable owner, recurring decisions, dependable evidence and an outcome observable within the horizon. The most financially dramatic problem may be a poor first candidate if no legitimate intervention is available. Test your Session 1 decision against that.
If no feasible action would follow the output, redefine the use or choose another problem before you select technology.
4:00 – 11:00
The four stages and what each must produce
Stage (Chapter 21) The work What it must produce Plan sections (programme mapping) Days 1 to 15: frame the decision Decision, alternative, boundary, timing, evidence and authority; past decisions reconstructed Outcome and acceptance owners; initial budget; unresolved conditions owned Sections 1 to 3 Days 16 to 30: baseline and tests Records reconciled; data contract; baseline and simpler alternative Gates set before results are seen; change control, stop authority and fallback Sections 4, 6, 11, 13 Days 31 to 60: shadow decisions and the response chain Method run beside the accepted procedure; disagreement classified Whether the warning reaches the right person, is checked, funded and completed in time; review burden Sections 5, 7, 9, 10, 12 Days 61 to 90: decide and establish ownership Bounded use under explicit authority where gates pass Review pack with both ledgers and a recommendation; ownership transferred Sections 8, 14 - The first three columns follow Chapter 21; the last is this programme's mapping, not the book's.
- If a reconstructed past decision shows the intervention would not have changed the action, revise the use before selecting technology. Where an important failure occurs, reduce scope or correct the cause before increasing exposure.
- Transfer means funded support, source maintenance, incident review, version release and periodic re-evaluation. A team that dissolves after the pilot leaves a demonstration, not a dependable capability.
Each stage ends in a decision. A stage that ends in a status update has not finished.
11:00 – 16:00
The register and the charter are the plan's spine
- The decision register (Appendix A) lets another competent person reconstruct the choice: decision and deadline, customer promise, current alternative, evidence snapshot, proposed intervention, hard constraints, decision owner, technical and commercial acceptance, decision and reason, execution reference, outcome acceptance and review trigger. Its completed entries are assumed teaching examples, not an actual Nigerian facility, company or project result.
- THE DOCTRINE
A rejected recommendation remains part of the record. Do not rewrite the original recommendation after the result becomes known; add a correction entry with its own timestamp and owner.
- Separate the decision from the action: instruction, executor, due time, acknowledgement and completion evidence. A sound decision can fail in execution; a completed action can leave the customer outcome unaccepted.
Pilot charter field (Appendix E) Completed illustrative entry (assumed) Business decision Whether a reliability advisory deserves a bounded operating deployment Scope One compressor group and one named response chain Advisory authority Recommendations only; existing protection and release controls retained Stop conditions Uncontrolled action, unacceptable evidence failure or material operating interference Fallback Existing approved process with assigned recovery owner Final decision Expand, restrict, redesign or stop based on the agreed evidence - Set numerical thresholds with the accountable owners: the book supplies no universal minimum accuracy, recall, warning time or return thresholds. Report independent events; ten favourable examples from one equipment episode are not ten independent demonstrations. An extension needs a specific unresolved question, because more time alone does not repair an untestable design.
If your charter could not end in a stop, it is not a test.
16:00 – 20:00
The sponsor's scorecard and the 90-day review
- Report four linked layers: technical (dependable outputs), workflow (review and execution), operating (accepted delivery, quality, equipment or stock) and economic (incremental cost and accepted financial consequence).
- Avoid measures that reward silence: fewer alerts may reflect missed events, fewer overrides excessive trust, more completed tasks administrative closure. Record no-use decisions and rejected recommendations, so the company does not keep funding an application whose missing response capability is unchanged.
- IN THE PHYSICAL SUPPLY · THE WORKED CASE
- Physical supply. At the review, the owner reconstructs one accepted outcome: for a fictional compressor group, the warning, the authorised inspection, the response and the accepted equipment state, with the evidence, alternatives and authority at each point.
AT THE CUSTOMER PROMISE · THE PARALLEL- Customer promise. The owner then reconstructs one rejected recommendation, as in Appendix A's illustrative entry: the existing allocation retained because a proposed receiving window was unconfirmed, with the reason, authority and consequence preserved.
- THE DOCTRINE
Leadership earns the result by funding the response, resolving dependencies and preserving the evidence. That discipline is the operating model beneath the technology.
Two reconstructions, one accepted and one rejected, tell you more than twenty slides.
20:00 – 25:00
Carry the promise through, then sign
- The Conclusion: the intelligent gas business is measured at the connection between physical supply and an accepted commercial commitment. Product, equipment, transport, receiving capability, measurement, billing and cash must remain linked.
- Different value chains need their own evidence and acceptance; common discipline connects them without pretending they are interchangeable. AI has earned its place when the business can explain what changed, why the action was accepted and how the result was established.
- THE DOCTRINE
The customer promise remains the test. Dependable intelligence should help people keep it through safe delivery, usable supply and an economic result the company can defend.
- Assemble Sections 1 to 14 under a one-page summary, noting where a later session changed an earlier one. Ask the review for one decision (continue, expand, revise or stop) with next scope, expenditure, authority, conditions and the operating owner who inherits the service.
- Sign and date the plan. Then sit the Final Exam: fifteen questions, three rebuilt from your own earlier mistakes, closed book, pass mark 70.
Sign only what you would be prepared to reconstruct honestly on day ninety.
Key Terms
90-Day AI Leadership Plan
Sections 1 to 14 assembled around one decision under a one-page summary, signed and dated, ending in a decision asked of the 90-day review.
Decision register
One record per decision that lets another competent person reconstruct the choice, including rejected recommendations and later corrections with their own timestamps.
Pilot charter
The scope, authority, tests, acceptance, stop conditions, fallback and final decision that make a pilot capable of producing a decision.
Sponsor scorecard
Four linked layers (technical, workflow, operating and economic) read through evidence so that no measure rewards silence.
Self-Check — Before You Submit
1. Which results does Chapter 21 treat as legitimate outcomes of the first 90 days?
- A. Only a full deployment across the business
- B. A bounded deployment, a revised pilot or a decision to stop
- C. Licence activation across all users
- D. A completed series of workshops and demonstrations
B. A bounded deployment, a revised pilot or a decision to stop — Chapter 21 names a bounded deployment, a revised pilot or a decision to stop, and warns against measuring the period by workshops, demonstrations and licence activation. Full deployment is not required for success.
2. At the 90-day review your sponsor reports that alerts fell by half and overrides almost stopped, and asks to expand. What do you ask for first?
- A. Approval to expand on the improved trend
- B. A higher-accuracy model from the vendor
- C. Evidence of why they changed, since fewer alerts may reflect missed events and fewer overrides excessive trust, read across all four scorecard layers
- D. Removal of the override function to simplify use
C. Evidence of why they changed, since fewer alerts may reflect missed events and fewer overrides excessive trust, read across all four scorecard layers — These are measures that reward silence. Expanding on the trend treats an unexplained change as success; read the scorecard through evidence.
3. Your pilot charter needs a minimum recall for relevant faults. Where does that threshold come from?
- A. The accountable owners, using the actual consequences and available evidence; the book supplies no universal threshold
- B. The 80% recall in Appendix H's illustrative table
- C. An industry average quoted by the vendor
- D. The supplier's contractual performance guarantee
A. The accountable owners, using the actual consequences and available evidence; the book supplies no universal threshold — Appendix E says thresholds depend on the failure mechanism, operating context and cost of an incorrect response. Appendix H's 80% is an assumed teaching count, not a target.
Today’s Assignment
Assemble Sections 1 to 14 into one signed 90-Day AI Leadership Plan for your company, with a one-page summary on which your 90-day review can decide.
- Assemble all fourteen sections in order, noting in one line wherever a later session changed an earlier one.
- Write the one-page summary: the decision, its owner by role, the authority needed and the decision you will ask of the review.
- Complete the Appendix A decision register entry for your decision and the Appendix E pilot charter: scope, existing alternative, advisory authority, evidence, tests, acceptance, stop conditions and fallback. Set thresholds with named owners; do not borrow the book's illustrative figures.
- Lay out the ninety days in the book's four stages, with the evidence and decision each must produce.
- Define the sponsor scorecard's four layers, the measure you will report in each and the one that could reward silence; attach the benefit ledger from Section 14.
- State the decision asked of the 90-day review (continue, expand, revise or stop) with next scope, expenditure, authority, conditions and the operating owner who inherits the service. Sign and date the plan, and name the commitment your company is least likely to keep.
YOUR 90-DAY PLAN
The signed 90-Day AI Leadership Plan.
How this is marked — the rubric
| Sections 1 to 14 assembled, consistent, signed and dated, with a one-page summary of decision, owner and authority | 20 |
| Complete decision register entry and pilot charter with stop conditions, fallback and owner-set thresholds | 25 |
| The ninety days in four stages, each with its evidence and decision | 20 |
| Four-layer scorecard resisting measures that reward silence, with the benefit ledger attached | 15 |
| One review decision with next scope, expenditure, authority, conditions and inheriting owner | 20 |
| Total · pass mark 70 | 100 |
Submitting session 15
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 15, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.FINAL EXAM
Your Final Exam is sat in this session. 15 questions, of which 3 are rebuilt from questions you got wrong on the practice papers. Sit it once your capstone is submitted.
Sit the Final Exam →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Plan assemblyDecision registerPilot charterNinety-day sequencingSponsor scorecardReview decision