15-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s power and energy sector
WEEK 1 OF 3 · DEFINE THE DECISION AND CHOOSE THE OPPORTUNITY · SESSIONS 1–5 · WORK AT YOUR OWN PACE
Orientation · The Decision-Rich Power SectorSESSION 1
The Decision-Rich Power Sector and the Reader's Toolkit
Why This Matters
Every power business is already rich in decisions. A maintenance decision affects the availability of a plant; a metering decision affects the credibility of a bill; a collection decision affects cash, service and customer trust. Artificial intelligence enters your business through those decisions, and its value depends on what your people can do with the output. Over fifteen sessions you will build a signed 90-Day AI Readiness Plan for your own company around one unresolved decision you choose today, one that currently costs your company time, money or service. A model may help you rank or explain it. Carrying it through to verified work is leadership, and that is what this programme trains.
25-Minute Module
0:00 – 5:00
The argument, and the finish line
- Electricity is an operating business with consequences well beyond the company supplying it, and this programme is for the executive who signs the cheque and remains accountable for the result.
- THE DOCTRINE
Where a line needs reinforcement, a transformer needs replacement or a plant lacks fuel, a model cannot supply the missing asset or resource. Treating AI as a substitute for engineering would give it a job it cannot perform.
- The three weeks: 1 Define the Decision and Choose the Opportunity; 2 Foundations, Procurement and the First Intervention; 3 Value, Governance and Execution. The finish line: a signed 90-Day AI Readiness Plan, fourteen sections written one per session, with the Appendix B scorecard self-scored on evidence, signed in Session 15.
- House rules: write about your own company; bring a real decision; no value claim without a baseline; no decision without a named owner; report a blocker the same day. The pass mark is 70.
- Said once: this is leadership training, not investment, legal or engineering advice. The frameworks and the scorecard are my proposals, not regulatory tests or validated industry benchmarks. Engineering, financial and legal approvals remain with the people authorised to give them.
Ask which decision is waiting before you ask what to buy.
5:00 – 9:00
One sector, different businesses, one starting point
- The book covers generation, transmission, distribution, electricity supply, mini-grids and the organisations that develop and finance energy projects. Their responsibilities and economics differ. Where an example comes from a part of the sector you do not run, apply it to the nearest equivalent decision in your business, or to the counterpart you depend on.
- The IEA discusses applications in forecasting, operations and innovation (IEA, Energy and AI, 2025). That establishes possibilities. A Nigerian operator must demonstrate its own operating and financial case.
- The book's sources were checked on 9 October 2026. A sector statistic stays attached to its reporting period; a regulation is read within its jurisdiction.
- My recommendation is to begin with one bounded decision, establish the baseline, protect customers and assets, and investigate at a cost the company can control. Waiting for a settled technology postpones learning; buying a broad platform first creates a different problem.
Possibility is not a business case; your own records are.
9:00 – 18:00
The six frameworks, as questions
Framework In brief The question it forces 1. Operating-Intelligence Loop Sense, decide, act, learn Where does our loop stop, and who owns that stage? 2. Data-Foundation Precondition Information fit for the decision Does the evidence meet this decision's floor? 3. Autonomy Ladder Inform, copilot, agent What authority has this task earned? 4. Value-versus-Risk Map Value beside the consequence of a wrong output What does a wrong answer cost? 5. Sequencing Discipline Bounded test against a baseline, then expand on results What evidence comes before the next commitment? 6. Operations-Excellence Lens Activity, operating outcome and financial value kept separate Which outcome can we observe and defend? - The loop. A dashboard usually strengthens sensing; a model may improve ranking. Neither establishes that work was completed. If crews cannot mobilise, a better ranking may leave the backlog unchanged. Fix the broken stage and name its owner before attributing value to the model.
- The data floor and the ladder. Identity, time, units and provenance are operating requirements; power, the rate of delivery at a moment, must stay distinct from energy, the amount delivered over a period. A missing critical input should produce a visible hold. On the ladder, drafting a work request differs from approving maintenance; technical capability alone does not confer authority.
- THE DOCTRINE
A high potential return does not remove an unacceptable exposure. Where safety, legality or essential control remains unresolved, treat it as a gate, not a weakness averaged into an attractive score.
- The sequence and the lens. Stopping can be a sound result. Alerts generated are activity; a verified finding and completed response are an operating outcome; attributable cash or contribution after incremental costs is a financial result.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. In a fictional generating company, a condition signal leads to an engineering review, an approved inspection and a verified finding. The loop breaks at act if no crew is available, and at learn if the finding never returns to the equipment record.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. In a fictional distribution company, an unusual account pattern leads to an investigation, a confirmed record correction and a reconciled payment. It breaks at learn if the correction never reaches the billing record. The same loop, seen twice.
Find the broken stage before you price the clever one.
18:00 – 22:00
Use the toolkit in an executive meeting
- Write one unresolved decision at the top of a page. Name the process, population, decision date and authorised owner. Describe the current method and the alternative of no change. Then walk the six frameworks in turn.
The decision record contains Why Request and alternatives What was asked, including no change Evidence and uncertainty What is known and not yet resolved Accountable owner A named role, not a department Operating and cash consequences What changes in the work and the money Next review date When the choice is examined again - Preserve supporting records under controlled identifiers. If the recommendation changes, retain the earlier version and explain why.
- THE DOCTRINE
Another qualified reviewer should be able to reconstruct the choice without relying on the memory of the people in the room.
A good operating review ends with a decision and a named owner.
22:00 – 25:00
Write it down, and Section 1
- Before you close the laptop, write one sentence naming your decision and one naming the loop stage where it stops.
- Choose a decision that costs your company time, money or service now, not one that sounds like an AI project. It may later need only a reminder, a better form or a clearer rule.
- Section 1 of your plan is Our company, the one decision we will carry, and where its loop breaks.
The decision you choose today is the one you will defend in Session 15.
Key Terms
Operating-Intelligence Loop
Sense, decide, act and learn. The loop closes when a result returns to the people responsible for the process.
Evidence floor
The minimum evidence a process owner agrees a decision needs, stating which defects can be controlled and which prevent an output.
Autonomy Ladder
Systems that inform, copilots that draft or suggest, agents that execute authorised steps: descriptions of authority, not product certifications.
Decision record
Request, alternatives, evidence and uncertainty, accountable owner, operating and cash consequences and next review date, kept so another reviewer can reconstruct the choice.
Self-Check — Before You Submit
1. A fictional distribution company is offered a model that ranks meter inspections more accurately, but crews cannot mobilise for half the existing queue. What comes first?
- A. Approve the model, since better ranking always helps
- B. Fix the broken act stage and name its owner before attributing value to the model
- C. Ask the vendor for a higher accuracy figure
- D. Automate account action instead of inspecting
B. Fix the broken act stage and name its owner before attributing value to the model — The loop breaks at act. A better ranking may leave the backlog unchanged. A higher accuracy figure is tempting but does nothing for the stage that is failing.
2. Under the Operations-Excellence Lens, which is an operating outcome?
- A. Alerts generated in a month
- B. The model's reported accuracy
- C. A verified finding and a completed response
- D. Dashboard log-ins
C. A verified finding and a completed response — Alerts generated are activity, however many there are. A verified finding with a completed response is the operating outcome; attributable cash after incremental costs is the financial result.
3. A candidate use scores well on value, but a safety question about the action attached to its output is unresolved. How does the Value-versus-Risk Map treat it?
- A. As a gate to resolve before the use proceeds
- B. As one weakness averaged against the value score
- C. As acceptable if the return is high enough
- D. As a matter for the technology team
A. As a gate to resolve before the use proceeds — Unresolved safety, legality or essential control is a gate, not a weakness averaged into an attractive score. A high potential return does not remove an unacceptable exposure.
Today’s Assignment
Name your company, its part of the power and energy sector, and the one unresolved decision you will carry through this programme, and show where its Operating-Intelligence Loop breaks.
- Your company. Its name, its part of the power and energy sector and what it must deliver.
- The decision. One unresolved decision that costs your company time, money or service, with its process, population, decision date and authorised owner by role.
- Current method and no change. How it is decided today, and what happens if nothing changes.
- Where the loop breaks. Walk it through sense, decide, act and learn. Name the stage where it stops, why, and who owns that stage.
- The honest part. What your company gets wrong about this decision today. No AI in use yet is normal.
- Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 1 — Our company, the one decision we will carry, and where its loop breaks.
How this is marked — the rubric
| Names the company, its part of the sector and what it must deliver | 15 |
| States one bounded decision with process, population, decision date and an owner by role, not a department | 30 |
| Describes the current method and the consequence of no change | 15 |
| Locates the loop break at a named stage with a reason and an owner | 25 |
| Names a real weakness in the student's own company | 15 |
| Total · pass mark 70 | 100 |
Submitting session 1
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: 1, 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
Decision-first thinkingThe six toolkit frameworksLocating the broken loop stageWriting a decision recordSeparating activity from outcome
Chapters 1 to 4 · Define the DecisionSESSION 2
From Power-Poor to Decision-Rich
Why This Matters
An executive can receive more reports and become less certain about the business. One report describes energy delivered, another invoices raised, a third collections against balances that include earlier arrears. Each may be correct within its own definition and together they may still fail to answer the decision before you. This session is about becoming decision-rich: a common account of what happened, an explanation of what can be changed and a route for making the change. You will read the regulator's published figures with care, define your decision so that it can actually change, and design a review that ends in work rather than in more charts. Most of this needs no technology at all.
25-Minute Module
0:00 – 5:00
Start with the operating evidence
- A plant report can show availability without explaining whether the equipment could have generated under the fuel and network conditions of the period. A dashboard helps only when its records are reconciled and someone has authority to act.
- Locate where an operating loss becomes an executive decision. A distribution company may need to distinguish a billing defect from a collection problem before assigning field staff; a generator, recurring equipment failure from fuel restrictions; a developer, whether a delay sits with an approval, a contractor or an unresolved company decision. Each calls for different evidence and a different owner.
- Describe the action in ordinary language: which assets will be inspected, which records corrected, which obligation escalated. A proposal labelled digital transformation answers none of these.
- THE DOCTRINE
The proposed action should be specific enough to audit after the event.
If you cannot say which assets or records will change, you do not yet have a proposal.
5:00 – 11:00
Read sector figures with care
- Three measures first, in plain words. Billing efficiency concerns the stage at which energy received becomes bills. Collection efficiency concerns the stage at which bills become cash. ATC&C, Aggregate Technical, Commercial and Collection loss, folds the technical, commercial and collection losses into one figure. Billing and collection are connected, but they are different stages of the commercial chain.
Aggregate DisCo measure (percent) Q1 2026 Q2 2026 Billing efficiency 79.24 78.67 Collection efficiency 78.95 81.06 ATC&C loss 37.44 36.23 - Source: the book's Figure 1.2, from NERC's Second Quarter 2026 Report, sections 2.3.3 to 2.3.5 (reporting period April to June 2026), with Q1 2026 beside Q2 2026. Higher billing and collection efficiency and lower ATC&C loss are favourable.
- Read the mechanism. Between the quarters collection efficiency rose while billing efficiency fell. An improved collection ratio does not establish that meter records improved; a lower aggregate loss ratio does not identify which investment caused the change. Nothing here attributes any movement to AI or to any programme.
- THE DOCTRINE
A board can use an external report to locate questions. It cannot substitute a sector aggregate for its own reconciled baseline.
- Your company should identify its reporting perimeter, its measurement period and its transaction records. If internal figures disagree with a regulatory return, that inconsistency is itself an operating issue to resolve before a model is asked to explain performance.
When two of your own reports disagree, the disagreement is the first finding.
11:00 – 17:00
Define a decision that can change
- Take inspection prioritisation as a candidate. The first question is whether the company has inspection capacity, a defensible process for selecting cases and a record of what inspectors found. A risk score can help rank cases. It cannot establish the outcome of an inspection that never happens.
Decision brief Evidence required before funding Operating decision A specific action, its frequency and the authority to take it Current performance A dated baseline with agreed definitions and exclusions Proposed improvement The mechanism connecting better evidence to a better action Delivery capacity People, equipment, access and budget needed to respond Control Review, appeal, safety approval or fallback appropriate to the consequence Result An observable outcome and a finance-approved value measure - The brief also names the alternative already in use: a manual queue, a fixed inspection interval or an engineering threshold. AI must earn its place against that alternative.
- The owner is an operating manager responsible for the affected process. The technology team prepares evidence and implements; it should not quietly become responsible for commercial policy or engineering approvals.
- A brief that exposes a missing dependency is doing useful work. If crews are unavailable, fund that capacity, choose another use or defer the model.
The most valuable line in a brief is often the one that says what is missing.
17:00 – 21:00
Conflicting objectives, and a review that leads to work
- Power businesses face objectives that cannot always improve together. Require one primary outcome and the protective measures relevant to it. An improvement in one metric cannot excuse an unacceptable consequence elsewhere.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. For maintenance in a fictional generating company, the outcome may be reduced failure exposure with approved operating limits preserved. Maintenance may require a planned outage, so the trade-off is stated, not hidden.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. For inspections in a fictional distribution company, verified useful findings may be the outcome, with wrongful accusations, complaints and crew safety as constraints. A collection drive may raise receipts while generating legitimate disputes if the bills are wrong.
- Approval states which decisions can be delegated, which need specialist sign-off and which remain outside. Begin the operating review with exceptions requiring a decision: source record, interpretation, proposed response and owner.
- THE DOCTRINE
Separate unresolved evidence from unresolved decisions. If a record is disputed, establish the record. If the record is trusted but two departments disagree on priority, management must resolve the trade-off.
- Where the company cannot identify an outage reliably, a generated answer should not invent a restoration time. Ask who is absent from the evidence: customers unable to use a digital channel, assets without inspection histories, events recorded only in another system.
Asking a model for more analysis can delay both kinds of unresolved work.
21:00 – 25:00
Write it down, and Section 2
- Before you close the laptop, list every report your decision currently relies on, and mark the two that most often disagree.
- Then write your decision as a brief that could change: the action, its frequency, the authority, the current alternative and the observable result.
- Section 2 of your plan is Our reports, the decision defined so it can change, and the review that leads to work.
Reconcile before you model.
Key Terms
Billing efficiency
A reported measure of the stage at which energy received becomes bills; a different stage of the commercial chain from collection.
ATC&C loss
Aggregate Technical, Commercial and Collection loss: one combined loss figure across the technical, commercial and collection stages. Lower is favourable.
Reconciled baseline
The company's own dated measure of current performance, with agreed definitions, perimeter, period and exclusions, against which any change is judged.
Protective measure
A constraint that must hold while the primary outcome is pursued, such as approved operating limits, crew safety or wrongful accusations avoided.
Self-Check — Before You Submit
1. Between Q1 and Q2 2026, Figure 1.2 shows aggregate DisCo collection efficiency rising from 78.95 to 81.06 percent. What may a board conclude?
- A. Meter records across the sector improved
- B. A specific investment caused the improvement
- C. AI adoption in the sector is working
- D. It may use the figure to locate questions, but not as a substitute for its own reconciled baseline
D. It may use the figure to locate questions, but not as a substitute for its own reconciled baseline — An improved collection ratio does not establish that meter records improved, and an aggregate change does not identify its cause. Attributing it to AI is the tempting error the book explicitly rules out.
2. In a fictional generating company, the maintenance and operations teams trust the same equipment record but disagree on which unit to take out for maintenance first. What is needed?
- A. Management must resolve the trade-off
- B. Further model analysis to settle the disagreement
- C. A new data-quality project
- D. A larger dashboard
A. Management must resolve the trade-off — The record is trusted, so this is an unresolved decision, not unresolved evidence. Asking a model for more analysis is tempting but delays the decision management must take.
3. Who should own the decision in a decision brief for inspection prioritisation?
- A. The technology team that introduced the tool
- B. The vendor supplying the model
- C. An operating manager responsible for the affected process
- D. The board's audit committee
C. An operating manager responsible for the affected process — The operating manager owns it. The technology team prepares evidence and implements, but should not quietly become responsible for commercial policy or engineering approvals.
Today’s Assignment
List the reports your decision relies on and where they disagree, define the decision so it can change, name its conflicting objectives and set a review that leads to work.
- The reports. Every report your decision relies on in your company, with definition, perimeter and period. Where two disagree, say how.
- The decision brief. Your decision against the six rows of the brief, naming the alternative already in use.
- Conflicting objectives. One primary outcome, its protective measures, and what can be delegated and what needs specialist sign-off.
- The review. How it will begin with exceptions and separate unresolved evidence from unresolved decisions.
- The honest part. Where your business relies on an unreconciled figure, or reviews charts instead of assigning work. No AI in use yet is normal.
- Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 2 — Our reports, the decision defined so it can change, and the review that leads to work.
How this is marked — the rubric
| Lists the reports with definitions and identifies a real disagreement between them | 25 |
| Completes the decision brief with a dated baseline, the current alternative and an operating owner by role | 30 |
| Names one primary outcome with specific protective measures and delegation boundaries | 25 |
| Designs a review that starts with exceptions and separates evidence from decisions, with an honest weakness named | 20 |
| Total · pass mark 70 | 100 |
Submitting session 2
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: 2, 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
Reading sector figures with careBilling versus collectionWriting a decision briefPrimary outcome and protective measuresReviews that lead to work
Chapters 1 to 4 · Define the DecisionSESSION 3
The Real Constraint Is Decisions, Not Megawatts
Why This Matters
The habit this session challenges is a common and expensive one: treating every performance problem as a request for more physical capacity. Generation, fuel, transmission capacity, distribution equipment and safe operating limits remain indispensable, and where the constraint is physical it must be addressed through engineering, commercial arrangements and investment. But many shortfalls in your company may sit elsewhere: in a contract, in poor information, in delayed authority or in a known action nobody executes. Before you approve any investment, AI or otherwise, you need to know which. Today you locate the constraint behind your decision, trace the whole transaction and decide what level of automation, if any, the decision deserves.
25-Minute Module
0:00 – 5:00
Locate the constraint before choosing the investment
- Before approving an investment, establish whether the observed shortfall comes from a physical limit, a contract restriction, poor information, delayed authority or failure to execute a known action. More than one may apply. The analysis should show how they interact rather than force the problem into one category.
- THE DOCTRINE
An algorithm cannot make an overloaded circuit safe by describing it more accurately.
- Two clocks, in plain words. Decision latency is the time between a usable signal and an authorised response: gathering evidence, reconciling reports, assigning responsibility, waiting for approval. Execution latency is the time from approval to completed work.
- Measure them separately. Otherwise management can buy a faster recommendation service for a problem caused by spare parts, site access or procurement. The book's Figure 2.1 is the author's diagnostic framework: faster analysis does not remove a physical or contractual constraint.
Ask whether the delay is in deciding or in doing before you fund either.
5:00 – 11:00
Trace the whole transaction
- Each stage of a power transaction has its own record and its own authority. A problem at one stage can appear downstream as a different problem.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. For a gas-fired generator, the chain may include fuel nomination, confirmation of supply, unit availability, permitted dispatch, metering and settlement. A forecasting tool can support fuel planning, but a nomination must still respect the contract and operating requirements.
- A model trained only on actual generation can mistake fuel interruption or curtailment for equipment failure, and recommend maintenance for the wrong cause. Preserve why output was restricted. An unknown cause should remain unknown until it is established.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. For a distribution company, trace energy received, energy measured, accounts billed, payments received and allocation to the correct balance.
- An incorrect customer-to-meter mapping can look like unusual consumption. A payment allocated to the wrong account can make a paying customer appear delinquent. Prioritising that customer more aggressively would compound the record defect.
- In both halves a downstream symptom can be an upstream record defect.
- Ask your team to trace a sample of complete transactions before anyone proposes automation. Include ordinary cases and known exceptions. The purpose is to find where evidence changes form, where decisions wait and where responsibility becomes unclear.
- This exercise is diagnostic. It is not a statistically sufficient evaluation of a model. It helps define the evaluation that is needed.
Follow one real transaction end to end before you read another dashboard.
11:00 – 16:00
Measure delay against the correct clock
- A timestamp is useful only when its meaning is agreed. The start of an outage, the time a complaint was logged and the time information reached the control room can differ. A work order created after a repair is not evidence that the repair happened at the order's creation time.
Delay Operating question Possible management response Signal to usable record Was the event captured and validated promptly? Improve capture, telemetry or reconciliation Record to decision Could the owner decide, and did the evidence arrive in time? Adjust authority, review timing or presentation Decision to mobilisation Were resources and access available? Address crews, parts, procurement or permissions Mobilisation to close Was the work completed and verified? Improve execution, testing and closure evidence - These categories are a proposed diagnostic. They do not assert that any Nigerian operator has a particular distribution of delays; your company must populate them from its own records.
- Record pauses with reasons. An engineering hold for a safety concern is different from a request lost between departments. Combining them as avoidable delay would encourage the wrong corrective action.
- Look at the spread, not only the average. A median can describe routine cases while a few long delays carry the greatest exposure.
A small number of long delays can carry the greatest exposure.
16:00 – 21:00
Match automation to the decision, and test the funding choice
- Not every delay needs machine learning. A reminder can resolve a missing response. A better form can prevent incomplete information. A clear rule can route a routine exception. When the decision is stable and its rule explicit, a conventional workflow may be cheaper and easier to maintain than a learned model.
- Machine learning becomes worth investigating when useful patterns are not captured by the simpler method. Work backwards from the time available: a recommendation arriving after a schedule is locked is too late, however strong its retrospective accuracy. Keep a fallback its users have practised.
- THE DOCTRINE
An investment committee should receive a constraint statement: what prevents the desired outcome today, what the proposal can change and which obstacles will remain.
- If better inspection ranking is expected to create value, show inspection capacity and the route from finding to correction. If an outage warning is expected to reduce disruption, show the response time and the actions engineering will permit. These dependencies belong in the investment case, not in a later implementation discussion.
- Compare the proposal with other uses of the same capital: metering, spare-parts arrangements, a revised approval process. Trace upstream to the earliest decision you can usefully change now; a missing spare part may trace back to stocking policy. Keep the measures of separate interventions separate.
A physical investment may be necessary while better decisions set its timing.
21:00 – 25:00
Write it down, Section 3, and Pop Quiz 1
- Before you close the laptop, write which of the five causes sits behind your decision, and whether its delay is decision latency, execution latency or both.
- Section 3 of your plan is The constraint located, the whole transaction traced and the right level of automation.
- Pop Quiz 1 opens now: five questions, closed book, on Sessions 1 to 3. Sit it before you start Session 4.
The cheapest fix that works is a good result, not a lesser one.
Key Terms
Decision latency
The time between a usable signal and an authorised response, including gathering evidence, reconciling reports, assigning responsibility and waiting for approval.
Execution latency
The time from approval to completed work, which can be held up by spare parts, site access or procurement rather than by information.
Constraint statement
A statement to the investment committee of what prevents the desired outcome today, what the proposal can change and which obstacles will remain.
Fallback
The practised procedure staff follow when a new service is unavailable: return to the current method, defer the decision or escalate.
Self-Check — Before You Submit
1. Approved repairs in a fictional distribution company wait weeks for spare parts. A vendor offers a faster recommendation service. What does the book's diagnosis suggest?
- A. Buy it, since faster recommendations shorten every delay
- B. The constraint is execution latency; a faster recommendation will not remove it
- C. Replace the maintenance team
- D. Ask the vendor to add spare-parts forecasting to the same model
B. The constraint is execution latency; a faster recommendation will not remove it — Separating decision and execution latency stops management buying a faster recommendation for a spare-parts problem. A parts forecast is tempting, but the earlier issue may be stocking policy.
2. A routine exception is decided by a stable, explicit rule, but cases are often routed to the wrong desk. What should be tried first?
- A. A learned model trained on past routing
- B. A copilot that drafts routing suggestions
- C. A larger data platform
- D. A clear rule in a conventional workflow
D. A clear rule in a conventional workflow — When the decision is stable and its governing rule is explicit, a conventional workflow may be cheaper and easier to maintain than a learned model. Machine learning becomes worth investigating when useful patterns are not captured adequately by the simpler method.
3. Why might a model trained only on actual generation recommend the wrong maintenance?
- A. It may mistake fuel interruption or curtailment for equipment failure
- B. Generation data is never accurate
- C. Models cannot be used in generation
- D. Maintenance records are always complete
A. It may mistake fuel interruption or curtailment for equipment failure — Without preserving why output was restricted, low output from fuel or dispatch limits can look like equipment failure. The fix is to keep maintenance records, operating conditions and restrictions in the evidence, not to abandon models in generation altogether.
Today’s Assignment
Locate the constraint behind your decision, trace one whole transaction, measure its delay against the correct clock, choose the right level of automation and test the funding choice.
- The constraint. Which of the five causes apply to your decision in your company: physical limit, contract restriction, poor information, delayed authority or failure to execute a known action. Show how they interact.
- The whole transaction. Trace one complete transaction, and one known exception, from first evidence to settlement, cash or closure. Say where the evidence changes form and where responsibility becomes unclear.
- The clock. Place the delay in the four categories, separating decision latency from execution latency, and say what you know about the tail of the queue.
- The level of automation. Whether the decision needs a model or a simpler method: a reminder, a better form or a clear rule. Name the fallback.
- The funding test. A short constraint statement and the competing interventions for the same capital. Be honest about what your company has so far tried to buy its way out of.
- Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 3 — The constraint located, the whole transaction traced and the right level of automation.
How this is marked — the rubric
| Names the binding constraint among the five causes and shows how causes interact | 25 |
| Traces a whole transaction and an exception with the stages where evidence and responsibility change | 25 |
| Separates decision latency from execution latency and addresses the tail, not only the average | 20 |
| Justifies the level of automation against a simpler method and writes a constraint statement with competing interventions | 30 |
| Total · pass mark 70 | 100 |
Submitting session 3
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: 3, 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 1 — Decisions and Constraints 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 1 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Locating the binding constraintDecision versus execution latencyTracing the whole transactionRight-sizing automationConstraint statements for investment committees
Chapters 1 to 4 · Define the DecisionSESSION 4
What AI Is and What It Is Not
Why This Matters
AI is a broad label, and broad labels are where poor investment decisions hide. A proposal that says it will use AI tells your investment committee almost nothing about what the system estimates, how it could be wrong, or what control it needs. This session replaces the label with the function: a forecast, a classifier, an anomaly detector, a language model or an optimisation method. Each needs a different test. You do not need to train a model. You do need enough understanding to decide whether your company can rely on one, under what conditions and with which controls, and to explain its limits to your board in plain language.
25-Minute Module
0:00 – 5:00
Understand the output before trusting it
- In an executive proposal, replace the word AI with what the system does. A demand forecast estimates a future quantity. A classifier assigns a case to a category. An anomaly detector flags a departure from an expected pattern. A language model generates text. An optimisation method selects a solution against an objective and constraints, and may or may not involve machine learning.
- Each needs a different test: a forecast, an evaluation period and horizon; a classifier, defined labels and evidence of errors; an anomaly detector, a process to establish whether an alert is meaningful; generated text, verification against sources; an optimisation, scrutiny of objective, constraints and assumptions.
- THE DOCTRINE
The supplier should show the input, the output and the proposed operating action separately. If management cannot explain what the service estimates or produces, it will struggle to design the control needed to use it.
- The book's Figure 3.1 is the author's classification for executive use. A polished demonstration can leave all three of input, output and action unclear.
Ask what it estimates before you ask how well.
5:00 – 10:00
Treat a prediction as conditional evidence
- A prediction is evidence that holds only under conditions, above all the records it learnt from. A dataset of completed maintenance orders may omit assets never inspected. More records do not help if they repeat the same selection process.
- A prediction is not a causal explanation. An association between asset age and recorded failure does not establish that replacing equipment by age alone is the best intervention. Engineering retains responsibility for treatment.
- Customer payment history has similar limits. A relationship with future payment does not determine whether a bill is accurate or lawful. A model trained on disputed accounts may learn the consequences of billing defects and present them as customer risk.
- THE DOCTRINE
A label such as high confidence does not establish calibration against outcomes. Ask how often the outcome occurred among comparable scores, and whether that holds in the intended population.
A model's pattern reflects what your company chose to record.
10:00 – 15:00
Put generated language under review
- Language models can help retrieve and summarise operating material, prepare explanations and organise documents. They can also produce plausible statements that are false. NIST discusses this risk as confabulation in its Generative Artificial Intelligence Profile (NIST AI 600-1, section 2.2, 2024). Fluent wording does not establish that a regulation, technical limit or financial figure is correct.
- A regulatory assistant should identify the source and version behind its answer, so the user can open the document and find the passage. The responsible reviewer still assesses applicability. An older order can remain relevant to a past decision while a newer order governs a current one.
- For operating procedures, define a restrictive task: locate an approved instruction, or draft a summary for review. The assistant should not invent a switching instruction, revise a protection setting or resolve contradictory procedures on its own authority. Missing or ambiguous evidence goes to the qualified owner.
- Test with incomplete, contradictory and misleading material. Treat retrieved content as evidence to analyse, not permission to change the task or disclose information.
Fluency is not evidence; a source the reviewer can open is.
15:00 – 21:00
Evaluate against the operating alternative, and explain the limitation
- An accuracy figure without context is not a business case. Test only with information available at the time of the decision. If later corrections leak into the inputs, a retrospective test can look strong while providing no usable prediction live. Time-dependent tasks generally need chronological evaluation.
- Compare the model with the current rule, a simpler statistical method and the manual process, where they are meaningful alternatives. For alerts, management needs the workload created, the useful findings and the significant cases missed.
Output Evaluation question Control in use Forecast Is error acceptable at the required horizon? Show uncertainty and refresh when conditions change Classification Which cases are missed or incorrectly flagged? Require evidence before consequential action Anomaly alert Does investigation establish a useful exception? Record findings and unresolved cases Generated text Are statements supported by applicable sources? Provide source access and accountable review Optimised plan Are objectives and constraints correct and feasible? Obtain engineering or commercial approval - ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. A maintenance prediction is not an engineering diagnosis. In a fictional transmission business, an incorrect flag can consume scarce inspection capacity and a missed defect can leave exposure untreated.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. A suspicious pattern is not a proven irregularity. In a fictional electricity supplier, an incorrect flag can cause unfair customer attention. Both sides ask the same question: what will the user do if it is wrong?
- For the board: NIST's AI Risk Management Framework 1.0, published in 2023, is voluntary guidance that can structure questions. It does not make an operation compliant with Nigerian electricity requirements. Tell the board what the service does, its evidence, known limitations and prohibited uses.
A score that ignores investigation capacity is detached from the decision.
21:00 – 25:00
Write it down, and Section 4
- Before you close the laptop, write the one word that replaces AI in your decision: forecast, classification, anomaly alert, generated text or optimised plan. If none fits, write that down too; it may mean the decision needs no model.
- Then write one sentence for your board on what the user will do when the output is wrong.
- Section 4 of your plan is The output we would use, its test against the current method, and the limitation in board language.
Keeping distinctions visible is what makes accountability practical.
Key Terms
Classifier
A method that assigns a case to a category; it needs defined labels and evidence of which cases it misses or flags incorrectly.
Calibration
Whether displayed scores match reality: how often the outcome actually occurred among comparable scores in the intended population.
Confabulation
NIST's term (AI 600-1, 2024) for a generative system producing plausible statements that are false.
Chronological evaluation
Testing a model only on information available at the time of each decision, in time order, so later corrections cannot flatter the result.
Self-Check — Before You Submit
1. A vendor reports excellent accuracy for a fault-prediction model. You learn that the test inputs included corrections entered after each work order closed. What is the problem?
- A. Accuracy figures are never relevant
- B. The model should be tested on more data of the same kind
- C. The test used information not available at the time of the decision, so live performance is unproven
- D. The model needs a language interface
C. The test used information not available at the time of the decision, so live performance is unproven — If later corrections enter the inputs, a retrospective test can look strong with no usable live prediction. More data assembled the same way repeats the flaw.
2. Which statement about the NIST AI Risk Management Framework 1.0 matches the book?
- A. It is voluntary guidance that can structure questions but does not make an operation compliant with Nigerian electricity requirements
- B. It is a Nigerian regulatory requirement for DisCos
- C. Following it makes an AI service compliant
- D. It replaces engineering approval
A. It is voluntary guidance that can structure questions but does not make an operation compliant with Nigerian electricity requirements — It is voluntary guidance published in 2023. Treating it as a compliance badge is the tempting error; local duties must still be mapped to the actual use.
3. A fictional distribution company's payment-risk model was trained on accounts that include many disputed bills. The commercial director wants to escalate collections on its top-scored customers. What should the executive require first?
- A. Nothing; the model has learned payment behaviour
- B. Transaction evidence and a challenge process before consequential action
- C. A higher confidence label on the scores
- D. A switch to a larger model
B. Transaction evidence and a challenge process before consequential action — A model trained on disputed accounts may learn the consequences of billing defects and present them as customer risk. A high confidence label does not establish calibration, so evidence and a challenge route come before action.
Today’s Assignment
State the kind of output your decision would use, how it would be tested against the current method, what generated text would need before anyone relied on it, and how you would explain the limitation to your board.
- The output. Replace the word AI in your company's decision with the function: forecast, classifier, anomaly detector, language model or optimisation method. Show input, output and proposed operating action separately. If none applies, say so.
- Conditional evidence. What the output would learn from, what those records omit, and why its pattern may not be causal.
- The test. How it would be evaluated against your current rule, a simpler method and the manual process, using only information available at the time of the decision. Name the errors that matter and the workload they create.
- Generated text. If any language output is involved, what source, version and reviewer it needs before anyone relies on it. If none, say what would stop one creeping in.
- The board paragraph. What the service would do, its limitations and prohibited uses, and what the user does when it is wrong. Be honest about where your business has trusted a figure because it looked precise.
- Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 4 — The output we would use, its test against the current method, and the limitation in board language.
How this is marked — the rubric
| Names the output by function and separates input, output and operating action | 20 |
| Explains the data's omissions and the limits of prediction without overstating causation | 20 |
| Designs a test against the current method using only information available at decision time, with error types and workload | 30 |
| Writes a board-ready limitation statement covering prohibited uses and the consequence of a wrong output | 30 |
| Total · pass mark 70 | 100 |
Submitting session 4
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: 4, 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
Naming the output, not the labelPrediction versus explanationReviewing generated languageTesting against the current methodExplaining limits to the board
Chapters 1 to 4 · Choose the OpportunitySESSION 5
Where AI Actually Creates Value in Power
Why This Matters
The same analytical method can have different economics in different parts of the sector. A generator, a distribution company, a transmission asset owner, a system operator, a supplier, a mini-grid developer and a financier each receive a recommendation with different authority and different cash. Value belongs to the business that can act. This session closes Week 1: you will follow the recommendation to the business that receives it, test whether a technical improvement converts into financial value, and select one use your company can execute. Sometimes the honest answer is a foundation repair or a simpler method.
25-Minute Module
0:00 – 5:00
Follow the business receiving the recommendation
- A generator settles energy under its applicable arrangements; a distribution company runs a commercial chain from received electricity to customer cash; a mini-grid developer makes site-level decisions; a transmission asset owner and a system operator have distinct responsibilities. The value model follows the business that can act.
- Begin with a controllable operating loss. Does it recur often enough to evaluate? Do usable records exist? Can a better decision change the outcome?
- THE DOCTRINE
A model can identify exposure the company cannot resolve during the intervention. That information may be useful, but it should not be presented as recoverable financial benefit.
- A modest improvement in a frequent decision may justify a small service the company can maintain; a larger theoretical benefit may depend on unavailable engineering capacity or costly integration. Show the delivery path beside the value path.
Value you cannot act on is information, not benefit.
5:00 – 11:00
Generation, transmission and the conversion test
- The conversion test, in plain words: does a technical improvement turn into money under the conditions the business really faces? More technically available capacity does not necessarily create more saleable output.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. For a generator, prioritise investigation of a defined failure mode, with the lead time a useful intervention needs. The recommendation may justify closer review or a diagnostic test; it does not acquire authority to alter protection settings or operating limits. A rise in discovered defects may reflect better investigation, not deterioration.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. For the same generator, fuel, dispatch, network access and settlement conditions can restrict production or revenue. Finance counts the additional contribution attributable to the intervention under those conditions, and does not count avoided failure costs again if they are already in it.
- Transmission and system operation. NERC's Who Does What Under the Electricity Act 2023, published on 30 April 2026, distinguishes TCN's transmission infrastructure role from NISO's system-operation and market-administration role. Identify the decision owner from the task.
- Work affecting protection, switching, dispatch or system security needs the relevant engineering authority. A planning study, a day-ahead forecast and a real-time response have different evidence needs, and value here should not be forced into a retail collection formula.
Count each benefit once, and only where it reaches your cash.
11:00 – 16:00
Distribution and the commercial chain
- Distribution offers several candidates: meter-data defects, inspection priority, account reconciliation, demand estimates, customer-service work. One broad loss-reduction project makes attribution harder; choose the specific mechanism.
- THE DOCTRINE
An anomaly warrants investigation. It does not prove theft or misconduct. Record the verified explanation rather than force every unusual case into a fraud label.
- Meter replacement, occupancy changes, outages, incorrect account mappings and ordinary demand changes can all explain an anomaly.
- Figure 4.1 compares reported ATC&C losses with MYTO targets from NERC's Second Quarter 2026 Report, Table 8; actuals are for Q2, targets are for 2026. In percent, target then Q2 actual: Abuja 16.97 and 32.31; Benin 17.23 and 33.84; Eko 14.19 and 21.00; Enugu 18.08 and 36.80; Ibadan 17.50 and 40.03; Ikeja 13.54 and 25.12; Jos 20.80 and 61.30; Kaduna 18.18 and 67.70; Kano 17.45 and 60.43; Port Harcourt 16.67 and 29.60; Yola 29.00 and 54.04.
- These differences invite diagnosis. They do not rank AI readiness or measure a loss that AI can necessarily recover. Use them to frame a discussion, then examine your own feeders, accounts and collection cohorts. Promising to close the entire published gap would assume technical, commercial and institutional effects the model has not demonstrated.
A published gap is a question about your records, not a promise about a model.
16:00 – 21:00
Supply, mini-grids, projects, and a use you can execute
- Electricity supply. State the unit, settlement interval, agreement and counterparty. Keep a forecast separate from the contractual quantity and a reported delivery separate from an accepted settlement. A model should not silently amend prices, volumes or obligations.
- Mini-grids. Stated interest differs from observed use, and consumption under limited supply may understate unconstrained demand. The NERC Mini-Grid Regulations 2026, section 4, set scope thresholds of up to 5 MW per site for isolated mini-grids and 10 MW per site for interconnected mini-grids, and address jurisdiction where a state regulator has assumed intrastate oversight. Verify your project's actual jurisdiction rather than apply an older threshold from memory.
- Development projects. Document analysis cannot establish physical completion merely because a report says work is complete; approved drawings, inspection records and engineering acceptance remain necessary.
Segment Candidate decision Evidence to establish Value to test Generation Prioritise a defined equipment inspection Condition records and verified failure history Additional contribution or defensible avoided cost Distribution Select meter or account records for investigation Account, meter, billing and field findings Verified corrections and attributable net cash Transmission assets Prioritise an asset-risk review Inspection findings and asset history Better risk treatment and work priority System operation Evaluate a forecasting aid Time-aligned operating and forecast records Useful forecasting and decision support Mini-grid Assess a site or operating schedule Demand, resource and service evidence Feasible service at a justified lifecycle cost Project delivery Escalate a package dependency Approved documents and verified progress Reduced avoidable delay or rework - These are candidate uses, not documented deployments. State who pays, who acts and who benefits; exclude benefits your company cannot receive.
Keep a simpler alternative and a foundation repair on the list until the evidence removes them.
21:00 – 25:00
Write it down, Section 5, and Practice Exam 1
- Before you close the laptop, write the row of the table nearest your business, and the one thing that could prevent your company executing it.
- Section 5 of your plan is The business that receives the recommendation, the conversion test and the use we can execute. With it, Week 1 of your plan is complete.
- Practice Exam 1 opens now: fifteen questions, closed book, on Sessions 1 to 5, closing Week 1. Sit it before you start Session 6.
Choose the use your company can carry, not the one that reads best in a brochure.
Key Terms
Conversion test
The check that a technical improvement becomes financial value under the actual fuel, dispatch, network access and settlement conditions, counting each benefit once.
Controllable operating loss
A recurring loss the company has records for and can change through a better decision within the intervention.
MYTO target
The 2026 ATC&C loss target shown beside each DisCo's reported Q2 actual in NERC's Second Quarter 2026 Report, Table 8.
Attributable contribution
The additional contribution that can be credited to the intervention itself, after the conditions that restrict output or revenue are taken into account.
Self-Check — Before You Submit
1. A fictional generating company forecasts that a maintenance model will add technically available capacity and values all of it at the full tariff. What is wrong?
- A. Nothing, if the model is accurate
- B. Available capacity should be valued at a higher rate
- C. Maintenance models cannot create value
- D. Fuel, dispatch, network access and settlement can restrict saleable output; only attributable contribution under those conditions counts
D. Fuel, dispatch, network access and settlement can restrict saleable output; only attributable contribution under those conditions counts — More available capacity does not necessarily create saleable output. Accuracy is the tempting answer, but an accurate model still fails the conversion test if fuel or dispatch prevent sale.
2. How does the book say Figure 4.1's gaps between ATC&C targets and Q2 2026 actuals should be used?
- A. As a ranking of each DisCo's AI readiness
- B. To frame a diagnosis, then examine the company's own feeders, accounts and collection cohorts
- C. As the loss an AI programme can recover
- D. As evidence that targets are unrealistic
B. To frame a diagnosis, then examine the company's own feeders, accounts and collection cohorts — The gaps invite diagnosis; they do not rank AI readiness or measure a loss AI can necessarily recover. Promising to close the whole gap assumes effects the model has not demonstrated.
3. A fictional distribution company's anomaly model flags a customer whose consumption fell sharply. What should happen?
- A. Investigate, considering meter replacement, occupancy, outages and account mapping, and record the verified explanation
- B. Disconnect the customer as a suspected thief
- C. Ignore the flag, since anomalies are usually wrong
- D. Add the customer to a fraud list pending review
A. Investigate, considering meter replacement, occupancy, outages and account mapping, and record the verified explanation — An anomaly warrants investigation but is not proof. A fraud list pending review is the subtler error: it labels the case before any explanation is verified.
Today’s Assignment
Follow the recommendation to the business in your company that receives it, apply the conversion test to the value claimed, and select one use your company can execute.
- The receiving business. Who receives the recommendation, who can act and who receives the benefit. Name any counterpart your company depends on, by role.
- The controllable loss. Whether your decision addresses a loss that recurs often enough to evaluate, has usable records and can be changed by a better decision.
- The conversion test. The value claimed, the conditions that could stop it reaching your company's cash, and what must be excluded or not counted twice.
- The use you can execute. One use as a row of the segment table, with what could prevent execution. Keep the simpler alternative and any foundation repair visible.
- The honest part. Where your business has counted a benefit it could not receive. No AI in use yet is normal: apply this to the proposal in front of you.
- Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 5 — The business that receives the recommendation, the conversion test and the use we can execute.
How this is marked — the rubric
| Identifies who receives, who acts and who benefits, including any counterpart dependency | 25 |
| Applies the conversion test with the specific conditions that restrict value and avoids double counting | 30 |
| Selects one executable use with evidence, value measure and named obstacles to execution | 30 |
| Keeps a simpler alternative or foundation repair visible and names an honest past error | 15 |
| Total · pass mark 70 | 100 |
Submitting session 5
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: 5, 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 1 — Define the Decision and Choose the Opportunity 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 1 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Following value to the business that actsThe conversion testReading published loss figuresAnomaly is not proofSelecting an executable use