THE INTELLIGENT EPC LEADERSHIP PROGRAMME
The contractor that can explain its delivery.
30-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s engineering, procurement and construction sector
WEEK 1 OF 6 · LEADERSHIP AND THE ENTERPRISE FOUNDATION · SESSIONS 1–5 · WORK AT YOUR OWN PACE
Orientation · The EPC Company's Actual BusinessSESSION 1
The Project That Looks Healthy and the Reader's Toolkit
Why This Matters
A project can have activity, records and several definitions of progress, and still lack a dependable connection between a document issued and a document approved, a weld completed and a weld accepted, revenue recognised and cash received. Over thirty sessions you will build a signed 90-Day Intelligent EPC Programme for your own company around one decision chain, which you choose today. The question I will keep asking is the one the book asks of every proposal: what the business is committing to deliver, which evidence establishes readiness, who can authorise a response and how the outcome will be checked. AI becomes part of that discipline. It does not replace the discipline.
25-Minute Module
0:00 – 4:00
The project that looks healthy, and the finish line
- The book's explicitly hypothetical review: engineering reports most documents issued, procurement most orders placed, construction thousands of hours worked, finance substantial revenue recognised. The client asks when a usable system will be handed over, and nobody can answer without another round of spreadsheets. AI can help here, but if the underlying definitions are weak, a fluent summary can present several incompatible positions as a single confident answer.
- THE DOCTRINE
An intelligent EPC company begins with a delivery discipline: what the business is committing to deliver, which evidence establishes readiness, who can authorise a response and how the outcome will be checked.
- Six weeks run from the enterprise foundation through tendering, design, procurement, construction, commissioning, claims, validation and finance to the author's practices and the ninety days. Twenty-nine plan sections are signed as one programme in Session 30. 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. Pass mark 70.
- Said once: this is leadership training, not investment, legal, contractual or engineering advice. The frameworks and records are the author's synthesis, proposed management tools rather than certifications.
Could your own review meeting answer the client's handover question today?
4:00 – 9:00
The EPC company's actual business
- An EPC contractor accepts a particular combination of engineering responsibility, purchasing obligations and execution risk under a particular contract. One delivers a process package with performance guarantees; another constructs to a client design; another subcontracts construction. Neither a reimbursable nor a lump-sum label alone settles liability.
- The Nigerian setting matters through the actual project: a swamp location, a coastal fabrication yard, a refinery modification and a gas-processing installation differ in access, interfaces, records and work windows. Identify these conditions in the project basis rather than apply an unsupported national average.
- The Petroleum Industry Act 2021 establishes separate upstream and midstream and downstream regulatory institutions; the Nigerian Oil and Gas Industry Content Development Act 2010 provides a separate Nigerian content framework that expressly includes contractors and subcontractors. Which obligations apply directly, which belong to the client and which flow through the contract is for the contractor to determine; a generic compliance assistant cannot answer that from a company label.
- Not a main EPC contractor? Apply every session to the same commitment seen from your side of the contract, or your nearest equivalent decision.
Your contract, not your company's description of itself, tells you which risks you are actually carrying.
9:00 – 18:00
The six toolkit frameworks, as questions
Framework In the book's words The question it forces Operating Intelligence Loop Sense, decide, act, learn Which stage breaks for our decision? Data Foundation Precondition Evidence adequate for the intended decision Is this input good enough for this commitment? Autonomy Ladder Retrieval, drafting for review, advisory recommendation, bounded execution What is this service permitted to do? Value versus Risk Map Value of a changed decision beside the consequence of error Which condition cannot be traded away? Sequencing Discipline Bounded workflow, repaired process, comparison, shadow mode, expansion What must be shown before we expand? Operations Excellence Lens Accepted delivery, cost, cash, quality, safety, reliability, capability Does the measure resolve an obligation? - The loop. A forecast completes none of the four stages by itself. Record recommendations that received no action; excluding them makes an application appear more effective than the operating process actually is.
- AT THE WORKFRONT · THE WORKED CASE
- Workfront. The book's example: an application predicts a late control panel. Sensing includes the vendor's actual manufacturing state; deciding includes feasible recovery options and their cost.
IN THE COMMITMENT · THE PARALLEL- Commitment. Acting includes approval, funding and execution; learning asks whether the panel arrived when needed and whether the intervention caused that. The same loop seen twice: the warning does nothing for the schedule promise if nobody funds the response.
- The data floor and the ladder. A preliminary quantity can support an early option comparison but not a firm fixed-price commitment. A missing inspection record is not a failed inspection, nor evidence of a successful one. Permission attaches to a task, population, user and environment.
- THE DOCTRINE
No level of autonomy automatically follows from the one below. An appropriate authority boundary can create more durable value than a premature attempt to remove review.
- The map, the sequence and the lens. Financial benefit cannot offset an unacceptable engineering, legal or safety condition. Compare any AI method with the existing process and a simpler alternative. A better model score is evidence about a model, not of project profit or client acceptance.
Find the broken stage of the loop before you fund another analytical feature.
18:00 – 22:00
Select the method, then record the decision
- Use arithmetic for balances, rules for known eligibility, engineering analysis for physical relationships, optimisation for constrained choices, statistical forecasting for uncertain outcomes and generative AI for suitable content tasks. Their failure modes remain different.
- THE DOCTRINE
A language model should not be the calculator of record for project economics.
- The common decision record keeps recommendation, approval, execution and assurance as separate states. Appendix A, the Executive Decision Record, is its template:
Fields (Appendix A) Required content Decision identity; choice requested Decision identifier; specific action, authority and response deadline Current position; evidence Accepted facts, forecasts and unresolved conditions separated; source, revision, cut-off, owner and location Assumptions; options Basis, uncertainty, approving reviewer; alternatives including the consequence of waiting Technical conditions; contract position Engineering, quality and operating acceptance; obligations, notice, exposure, competent interpretation Economics; recommendation Cost, cash, funding, eligible value, downside; proposed choice and its evidence Resolution; follow-up Approver, conditions, resources, owner, due date; evidence to verify execution and outcome - Record material dissent. If an assumption changes in the meeting, update the calculation rather than approve the old headline. A completed action requires execution evidence; a claimed benefit requires outcome evidence.
A record that cannot be filled in honestly says the decision is not ready.
22:00 – 25:00
Write it down, and Section 1
- Before you close the laptop, write your company's place in EPC delivery and its contract model.
- Choose one decision chain that currently costs time, money or trust between commitment and collected cash, with material value, accessible evidence, an accountable owner and a practical response.
- Section 1 of your programme is Our company, the one decision chain we will carry, and where its loop breaks. The right answer may later prove to be a simpler method, or no model at all.
The chain you choose today is the one you will sign for in Session 30.
Key Terms
Operating Intelligence Loop
Sense, decide, act and learn as four connected responsibilities. A forecast completes none of them by itself.
Acceptance
An authorised determination that specified evidence and conditions meet the applicable requirement.
Autonomy
Permission to act within an accepted use. General model capability does not establish that permission.
Executive Decision Record
One record per material decision (Appendix A), keeping recommendation, approval, execution and assurance as separate states.
Self-Check — Before You Submit
1. A fictional EPC contractor's service correctly warns of a late control panel. A feasible recovery option exists, but no one approves the funding. Where has the Operating Intelligence Loop broken?
- A. At sense: the forecast needed more data
- B. At decide: no options were compared
- C. At act: the choice was never funded or executed
- D. At learn: the outcome was never reviewed
C. At act: the choice was never funded or executed — Acting includes approval, funding and execution, and that is where this chain stopped. Improving the forecast is tempting but does nothing for the failing stage.
2. Under the Autonomy Ladder, an assistant has been permitted to draft technical queries for review. What else does that permission allow?
- A. Nothing further: drafting for review does not include sending it to the client
- B. Sending routine queries without review
- C. Changing the contractual baseline
- D. Any advisory action, because drafting sits above it
A. Nothing further: drafting for review does not include sending it to the client — No level automatically follows from the one below. The tempting answer, sending routine queries, confuses capability with authority.
3. An option with a large financial benefit fails an engineering acceptance condition. A colleague proposes a weighted score in which the benefit wins. What does the Value versus Risk Map require?
- A. Accept the weighted score
- B. Reduce the weight given to engineering
- C. Refer the score to the technology team
- D. Keep the mandatory condition separate; benefit cannot offset it through a weighted average
D. Keep the mandatory condition separate; benefit cannot offset it through a weighted average — The map keeps mandatory conditions separate from tradeable preferences. A weighted average feels balanced but hides exactly the condition that must not be traded.
Today’s Assignment
Name your company and its place in EPC delivery, choose one decision chain that costs time, money or trust between commitment and collected cash, and state where its Operating Intelligence Loop breaks.
- Your company. What your business delivers, its place in EPC delivery and its contract model, seen from your side of the contract.
- The decision chain. One chain with material value, accessible evidence, an accountable owner by role and a practical response, and what it currently costs.
- The loop. How sense, decide, act and learn work for it today; the stage where it breaks, why, and who owns that stage.
- The honest part. What your company currently gets wrong about this chain. No AI in use yet is normal; work from the decision in front of you. Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 1 — Our company, the one decision chain we will carry, and where its loop breaks.
How this is marked — the rubric
| Names the company, its place in EPC delivery and its contract model | 20 |
| States one decision chain with material value, accessible evidence, an owner by role (not a department) and a practical response | 30 |
| Locates the loop break at a named stage, with a reason and an owner | 30 |
| Names a real current weakness, separating method from process repair | 20 |
| 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
Delivery discipline before technologyThe six toolkit frameworksLocating the broken loop stageDeliberate method selectionWriting an executive decision record
Part One · Leadership and the Enterprise FoundationSESSION 2
The Leadership Mandate
Why This Matters
Engineering, procurement and finance can each be working to a plan. The difficulty arises when they are different plans: engineering promising one sequence, procurement funding another and finance assuming a third. AI can reveal those differences quickly; it cannot resolve them. That is the managing director's work, and the work of every executive who signs a commitment that crosses departments. Today we turn the decision chain you chose in Session 1 into a mandate: who owns it, what question it answers, what behaviour lets people challenge it and what the board should see. No material AI investment without a named decision. If you hold one discipline from this week, hold that one.
25-Minute Module
0:00 – 6:00
Own the commitments that cross departments
- The managing director's first responsibility is to establish which commitments the enterprise is willing and able to carry. A project director cannot create a reliable delivery model while engineering promises one sequence, procurement funds another and finance assumes a third.
- THE DOCTRINE
AI can reveal the differences between what engineering promises, procurement funds and finance assumes, but executives must resolve them.
- Create a project mandate at award. It states scope boundaries, performance commitments, contract model, accepted baseline, funding basis, key interfaces and retained risks, and names the people who can approve changes to them. A useful mandate is short enough to read and specific enough to constrain decisions.
- The AI programme needs the same discipline: name the delivery decisions it will improve, the owner of each workflow and the authority that remains outside the application.
Role What the book says it owns Chief executive The portfolio and the resolution of conflicting incentives Technical authority Technical acceptance Commercial leadership Contractual positions Finance The treatment of cost and cash Information technology The supported service environment within agreed governance
If three departments can each change the sequence, nobody owns it.
6:00 – 12:00
A decision portfolio rather than an application catalogue
- Start with questions that have an answer the business can use. The book's examples: Which packages threaten the next system-completion milestone? Which bid assumptions have changed enough to require an executive decision? Which accepted work cannot yet be invoiced, and why? These questions connect data to responsibility.
- For each question identify the latest useful response date: the last moment at which an answer can still change what you do. Think of it as the cut-off for a lifting plan; after it, the information is still true but the crane has gone.
- AT THE WORKFRONT · THE WORKED CASE
- Workfront. A fictional EPC contractor receives a warning about a long-lead package while expediting, substitution or resequencing remain possible. On the workfront the warning has its full value: there is still a physical response.
IN THE COMMITMENT · THE PARALLEL- Commitment. The same warning after those options close may still support commercial notice or funding decisions, but its value claim must change accordingly. The data is identical; what has changed is the decision it can still serve, so the benefit cannot be claimed as if the package had been recovered.
- Do not insist that every business unit produce the same number of AI applications. One controlled engineering revision service may matter more than several polished office assistants. Allocate attention by consequence and evidence readiness.
- This is the Sequencing Discipline from Session 1 applied at portfolio level: choose a bounded workflow with an owner, usable evidence and a feasible response before counting applications.
Count decisions changed, not applications launched.
12:00 – 17:00
The behaviour that makes challenge possible
- Executives should ask where the number came from and what would make the recommendation wrong. They should also avoid punishing the person who identifies an unsupported completion claim. Otherwise staff learn to make the dashboard greener while unresolved work migrates into the final months.
- THE DOCTRINE
Reward an early, evidenced exception that permits a response. Distinguish it from repeated failure to execute an agreed action.
- A culture that treats both as bad news encourages concealment rather than learning. The first is a gift to the project; the second is a performance issue. Your executive team needs to be seen to treat them differently.
- The executive meeting should record a choice, its conditions and its owner. A list of concerns without decisions consumes attention and leaves the delivery model unchanged. AI can prepare the agenda and retrieve evidence, but the meeting's usefulness depends on management acting within its authority.
Watch how your team reacts the next time someone reports bad news early; that reaction is your real AI governance.
17:00 – 22:00
The board's evidence and the sponsor's job
- The board should receive changes in exposure, decisions required and evidence of programme outcomes, with unresolved mandatory conditions reported separately from financial expectations. Show the original baseline and approved changes, so a result cannot be created by silently moving the comparison. Report full programme cost, reviewer effort and operating response cost, including applications withdrawn or retained at limited authority.
- THE DOCTRINE
No material AI investment without a named decision; no consequential deployment without an authority boundary; no benefit claim without an accepted comparison.
- I expect an AI sponsor to understand the commitment being improved before selecting the technology. If the opportunity is quote completeness, the sponsor should know who accepts the scope, who owns the rates and who can approve the final offer. My portal-monitoring, PSR and quote practices reflect that approach: they place intelligence in recurring work that management already needs to perform. Their actual financial value must be demonstrated through records rather than attached to the story after the event.
- The sponsor must provide a decision owner and capacity to act. A daily tender service with no preparation resources cannot turn every detected opportunity into a bid; a quote service cannot make an unverified rate reliable through more careful formatting. And ask what the business will stop doing if the service is withdrawn; that tests whether it understands the dependency it is creating.
A sponsor who cannot name the approver of the final offer is sponsoring a demonstration, not a decision.
22:00 – 25:00
Write it down, and Section 2
- Before you close the laptop, write the cross-department commitment behind your decision chain and the departments whose plans currently disagree about it.
- Turn your chain into a portfolio entry: the question it answers, the latest useful response date, the owner by role and the authority that stays outside any application.
- Section 2 of your programme is The commitment, the portfolio entry and the board's evidence.
If your entry reads like a product description, rewrite it as a question someone must answer by a date.
Key Terms
Project mandate
A short record at award stating scope boundaries, performance commitments, contract model, accepted baseline, funding basis, key interfaces and retained risks, and naming who can approve changes to them.
Decision portfolio
A set of business questions with usable answers, each with an owner and a latest useful response date, in place of a catalogue of applications.
Latest useful response date
The last date at which an answer can still change the response. After it, a warning may support other decisions, but its value claim must change.
Baseline
A retained reference for scope, schedule, cost or performance. Approved changes should not erase the original comparison.
Self-Check — Before You Submit
1. A fictional EPC contractor's site manager reports that a claimed system completion is not supported by inspection records. The site manager is criticised in the executive meeting for the bad news. What does the book warn will follow?
- A. Staff will improve the inspection records
- B. Staff learn to make the dashboard greener while unresolved work migrates into the final months
- C. The board will receive better evidence
- D. The AI service will learn to detect the gap
B. Staff learn to make the dashboard greener while unresolved work migrates into the final months — Punishing the person who identifies an unsupported completion claim teaches concealment. Expecting better records to follow is the tempting answer, but the incentive created points the other way.
2. Which item does the book say the board should receive?
- A. The number of AI applications each business unit has launched
- B. Model accuracy scores in place of outcomes
- C. A single combined figure of financial benefit and risk
- D. Full programme cost, reviewer effort and operating response cost, including applications withdrawn or retained at limited authority
D. Full programme cost, reviewer effort and operating response cost, including applications withdrawn or retained at limited authority — The board needs exposure changes, decisions required and outcome evidence, with full cost and withdrawn applications. Counting applications is the catalogue mistake; a combined figure hides unresolved mandatory conditions, which the book says to report separately.
3. Your group chief executive asks each of five business units to deliver three AI applications this year. What is the better leadership response?
- A. Replace the quota with a decision portfolio, allocating attention by consequence and evidence readiness
- B. Accept the quota, because equal effort is fair
- C. Ask the IT function to choose the fifteen applications
- D. Raise the quota so more is learned
A. Replace the quota with a decision portfolio, allocating attention by consequence and evidence readiness — The book says not to insist on equal numbers of applications, since one controlled engineering revision service may matter more than several polished office assistants. Fairness of effort is the tempting reason, but it ignores consequence.
Today’s Assignment
Write the cross-department commitment behind your decision chain as a decision portfolio entry, with the behaviour that makes challenge possible and the evidence your board should see.
- The commitment. The commitment in your company that crosses departments, and where engineering, procurement, construction and finance currently assume different things about it.
- The portfolio entry. The question it answers, the latest useful response date, the decision owner by role and the authority that remains outside any application. Write it as a question, not a product.
- Challenge. How your business currently treats an early, evidenced exception, and how it treats repeated failure to execute an agreed action. Be honest if the two are treated the same.
- The board's evidence. What the board should see: change in exposure, decisions required, the original baseline and approved changes, full cost including reviewer effort, and what you would stop doing if the service were withdrawn.
- No AI in use yet is normal; work from the decision or proposal in front of you. Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 2 — The commitment, the portfolio entry and the board's evidence.
How this is marked — the rubric
| Names a commitment that crosses departments and the specific disagreement between them | 25 |
| Writes a portfolio entry as a question with a latest useful response date and an owner by role, not a department | 30 |
| Describes how the business currently treats an early exception, honestly and specifically | 20 |
| Specifies board evidence including the original baseline, full cost and a withdrawal fallback | 25 |
| 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
Owning cross-department commitmentsBuilding a decision portfolioRewarding early, evidenced exceptionsBoard evidence without moved baselinesSponsoring a decision, not a tool
Part One · Leadership and the Enterprise FoundationSESSION 3
The Nigerian EPC Operating Model
Why This Matters
Two companies can both call themselves EPC and carry very different risks. That is why the book insists that you specify the business before you specify the technology. Today you map how your company actually earns money and where it can lose it, model the project conditions that genuinely affect your decision and trace the six states of the commitment chain from bid promise to collected cash. This is where an AI proposal can go quietly wrong: it describes the software in detail and the business in a sentence. Leadership needs to see exactly where the chain breaks. Pop Quiz 1 opens at the end of the session.
25-Minute Module
0:00 – 6:00
Specify the business before specifying the technology
- Map how the contractor earns money and where it can lose it. The book contrasts four arrangements:
Arrangement Where exposure sits, per the book Lump sum Quantity growth and rework may sit with the contractor, subject to actual contract provisions Reimbursable Recoverability depends on the contract's rules and evidence Unit rate Depends on measurement and accepted quantities EPC management services May expose the business differently from direct procurement and construction obligations - THE DOCTRINE
Do not infer risk allocation from the contract title. Build a responsibility matrix from the executed terms, schedules and authorised clarifications.
- The matrix identifies design responsibility, procurement ownership, client free-issued material, site access, utilities, permits, performance testing, delay treatment, insurance and acceptance. Connect those responsibilities with actual capabilities: if a proposal depends on specialist engineering, establish the arrangement and its limits; if on imported equipment, identify technical data, funding and logistics dependencies; if on a subcontractor, define what evidence you will accept before certifying the subcontractor's work.
Read the executed terms, not the cover page, before anyone builds anything.
6:00 – 11:00
Model the Nigerian project conditions explicitly
- Describe access routes, work calendars, seasonal assumptions, client operating windows, site accommodation, power needs, communications and emergency arrangements for the project, using verified site information and current commitments. A map location does not establish the bearing capacity of a route or permission to use a landing point.
- Each setting changes the question. For a brownfield modification, distinguish contractor-controlled work from operator-controlled isolation and shutdown windows. For a fabrication-yard scope, connect workshop capacity, lifting facilities and inspection access to the delivery sequence. For dispersed pipeline work, model how crews, materials and accepted access reach each workfront.
- These differences shape data collection. An offline workfront needs a controlled synchronisation process. A vendor package may need a secure exchange of drawings rather than live sensor integration. A project with several contractual parties needs clear information ownership and restrictions.
- As in Session 1, the point is the actual project, not a national average. If your company is a client, vendor or subcontractor, model the conditions that govern your side of the same commitment.
Ask who has verified the route, not who has seen it on a map.
11:00 – 18:00
The commitment chain
- The book uses six connected states: bid promise, contractual commitment, planned delivery, executed work, accepted evidence and collected cash. Each transition needs an owner and a record. Think of it as a relay: the baton is the commitment, and a dropped handover is invisible in any one runner's lap time.
Transition What the book says if it fails Bid promise to contractual commitment A promise that disappears at award is a mobilisation risk Contractual commitment to planned delivery to executed work The book names no separate risk here; each transition still needs an owner and a record Executed work to accepted evidence Work that cannot become accepted evidence is a completion risk Accepted evidence to collected cash Acceptance that cannot become a valid invoice is a commercial risk - THE DOCTRINE
An AI service can compare the six states and highlight discrepancies. It should not merge them into one ambiguous progress figure.
- AT THE WORKFRONT · THE WORKED CASE
- Workfront. The book's hypothetical: a tender promises a locally available specialist team. After award, the team is not secured. On site, the workfront cannot start the scope that depended on it.
IN THE COMMITMENT · THE PARALLEL- Commitment. A generated mobilisation narrative describing the team as ready would worsen the problem. The correct output is an exception showing the promise, the current evidence, the required decision and the feasible correction. Here the two sides differ: the workfront needs a team, the commitment needs an honest record that it is missing.
- Students who are not the main contractor: trace the same six states for the same commitment seen from your side of the contract, and find the transition that breaks.
Never let a fluent narrative cover a missing state in the chain.
18:00 – 21:00
The enterprise view without erasing project differences
- Use common identities for projects, packages, systems, suppliers, obligations and cost accounts. Allow project-specific acceptance rules and interfaces. Standardise the meaning of a status rather than forcing every contract into an identical workflow.
- The enterprise review should distinguish profitable backlog from executable backlog and funded backlog. The same award can appear in all three, but the criteria differ. A signed project without mobilisable resources or available funding is a commitment requiring action, not a dependable delivery forecast.
- The intelligent operating model is therefore an architecture of responsibility. Technology supports it by keeping the commitments and evidence connected, and becomes useful when leadership can identify the next responsible decision without reconstructing the project from informal messages.
Ask which of your awards are executable and funded, not only profitable.
21:00 – 25:00
Write it down, Section 3, and Pop Quiz 1
- Before you close the laptop, write the contract arrangement under which your decision chain sits and the three responsibility-matrix items that most affect it.
- List the project conditions that genuinely change your decision, and mark which are verified and which are assumed. Then trace your chain through the six states and mark the transition that breaks.
- Section 3 of your programme is The business specified, the Nigerian conditions modelled and the commitment chain traced.
- Pop Quiz 1 opens now: five questions, closed book, covering Sessions 1 to 3. Sit it before Session 4.
The transition you mark today is where your programme will do its first useful work.
Key Terms
EPC
Engineering, procurement and construction, with actual responsibilities determined by the agreement rather than the acronym alone.
Responsibility matrix
A map built from the executed terms, schedules and authorised clarifications showing who holds design, procurement, free-issued material, access, utilities, permits, testing, delay, insurance and acceptance.
Commitment chain
Six connected states: bid promise, contractual commitment, planned delivery, executed work, accepted evidence and collected cash, each transition with an owner and a record.
Executable and funded backlog
Awards that can actually be mobilised with available resources, and those with available funding, distinguished from backlog that is merely profitable on paper.
Self-Check — Before You Submit
1. After award, a fictional EPC contractor finds that the specialist team promised in its tender is not secured. A draft mobilisation report generated by an AI service describes the team as ready. What should the output be?
- A. The generated report, with a note to confirm later
- B. A single progress percentage for mobilisation
- C. A revised report omitting the team
- D. An exception showing the promise, the current evidence, the required decision and the feasible correction
D. An exception showing the promise, the current evidence, the required decision and the feasible correction — The book says a narrative describing the team as ready worsens the problem. Omitting the team is tempting because it avoids a false claim, but it hides the broken transition instead of routing it to a decision.
2. Which three views of backlog does the book say the enterprise review should distinguish?
- A. Signed, unsigned and expected
- B. Profitable, executable and funded
- C. Upstream, midstream and downstream
- D. Lump sum, reimbursable and unit rate
B. Profitable, executable and funded — The same award can be profitable without being executable or funded. Contract types are a tempting distractor, but they describe risk allocation, not whether a commitment can actually be delivered.
3. A fictional fabrication subcontractor plans pipeline material deliveries from a map showing a route to a landing point near the workfront. What does the book require before relying on it?
- A. Nothing further; the map is sufficient for planning
- B. An AI route optimisation
- C. Verified site information and current commitments, since a map location does not establish the bearing capacity of a route or permission to use a landing point
- D. A national average of delivery times for the region
C. Verified site information and current commitments, since a map location does not establish the bearing capacity of a route or permission to use a landing point — Conditions must be modelled from verified information. A route optimiser is tempting, but it would optimise over unverified constraints; a national average is exactly what the book says not to apply.
Today’s Assignment
Specify your company's business before any technology, model the Nigerian project conditions that affect your decision and trace your decision chain through the commitment chain.
- The business. The contract arrangement your chain sits under and where your business earns money and can lose it. Name the responsibility-matrix items that matter, drawn from the executed terms rather than the contract title.
- The conditions. The project conditions that change your decision (access, calendars, windows, power, communications, offline workfronts, information ownership), each marked verified or assumed.
- The chain. Your decision traced through bid promise, contractual commitment, planned delivery, executed work, accepted evidence and collected cash, with the owner and record at each transition and the one that breaks.
- The honest part. Where your company currently merges states into one progress figure, or counts profitable backlog as executable. No AI in use yet is normal. Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 3 — The business specified, the Nigerian conditions modelled and the commitment chain traced.
How this is marked — the rubric
| Specifies the contract arrangement and responsibility-matrix items from executed terms, not the title | 25 |
| Models the project conditions relevant to the decision, separating verified from assumed | 25 |
| Traces all six states with an owner and record per transition, and names the break | 30 |
| Names a real current weakness in how states are merged or backlog is classified | 20 |
| 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 — The Mandate and the Operating Model 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
Specifying the business firstReading risk from executed termsModelling actual project conditionsTracing the commitment chainDistinguishing executable and funded backlog
Part One · Leadership and the Enterprise FoundationSESSION 4
Project Evidence and Data Architecture
Why This Matters
An argument about progress in an EPC company can quickly become an argument about records: which tag, which revision, which receipt, which date. AI does not settle those arguments; it inherits them. Today we cover the foundations that let a reviewer trace a recommendation back to accepted evidence: stable identities, the state of knowledge at the time of decision, reconciliation before modelling and a data service that someone actually operates. None of this is glamorous, and all of it decides whether your decision chain can be trusted. A link to the latest file alone cannot establish what management knew when it approved an action.
25-Minute Module
0:00 – 5:00
Identity comes before integration
- In plain language, identity means being sure that two records refer to the same physical or documentary thing. It is the site practice of checking the stamped tag on a spool before installing it, applied to data. Without it, joining two systems only joins two sets of guesses.
- Define a project identifier, work breakdown structure, cost breakdown structure, engineering document identifier, equipment tag, material identifier, purchase order line, work package and system boundary. Specify where each identity originates and which relationships are authoritative.
- THE DOCTRINE
Do not assume that equal text strings mean equal objects. Maintain controlled mapping tables with source, reviewer and effective date.
- A vendor's equipment tag can resemble a client tag while referring to a different assembly; different descriptions can refer to the same approved object.
- The essential structure is a set of relationships: a drawing applies to tags; a requisition requests material; a purchase line supplies it; a receipt establishes quantity and condition; a work package consumes it; an inspection supports acceptance; a system record establishes readiness. These links can live in relational tables. A graph database is optional, not a prerequisite.
Ask who owns the mapping table before you ask who owns the dashboard.
5:00 – 10:00
Preserve the state at the time of decision
- In plain language, the state at the time of decision is what could actually have been known when the choice was made. Store event time and recording time separately where they differ.
- AT THE WORKFRONT · THE WORKED CASE
- Workfront. The book's example: a receipt occurs on Monday and is entered on Wednesday. On site, the material was there from Monday.
IN THE COMMITMENT · THE PARALLEL- Commitment. A forecast made on Tuesday must not use the Wednesday entry as though it had been available. Using it is leakage: information that was not available at the time flatters the evaluation. Keeping the two times apart is what makes a retrospective review of the Tuesday decision fair.
- Retain revisions and effective periods. Superseded documents must remain traceable while being excluded from current execution according to the project's controls. Deletion can destroy claim evidence and make a past decision impossible to reconstruct.
- Maintain a decision snapshot containing the relevant input versions, calculation version and permissions. A link to the latest file alone cannot establish what management knew when it approved an action.
If you cannot reconstruct last month's decision, you cannot defend it next year.
10:00 – 15:00
A minimal schema, missingness and reconciliation
Record Fields the book specifies Evidence object object_id, object_type, project_id, source_system, source_reference, revision, status, observed_at, recorded_at, effective_from, effective_to, unit, owner, access_class Relationship from_id, relationship_type, to_id, source_reference, status, approving_owner - Keep machine-inferred relationships separate from approved ones. A proposed match can assist review; it must not silently become the basis for an inspection certificate or a payment quantity. Preserve the reason for rejecting a suggestion when it will improve later matching.
- Missingness deserves its own meaning. Unknown delivery date, confirmed no date, date pending client approval and date not applicable are distinct states. Converting all of them to a blank field loses the reason for the exception.
- In plain language, reconciliation means checking that numbers which must agree do agree, before anyone models them. Ordered quantity should reconcile with receipts, returns and remaining quantity on a consistent unit basis. Accepted hours and quantities should reconcile with the progress rules. Currency totals should not add naira and dollars without an explicit conversion basis.
- THE DOCTRINE
A tolerance is not permission to hide an unexplained difference. Automated cleaning should propose corrections and preserve the original record.
A blank field is an answer to nothing; ask which of the four states it is hiding.
15:00 – 21:00
Operate the data service, and the source register
- Assign owners for definitions, interfaces, quality incidents and access. Monitor freshness at the level of the actual source: a dashboard refreshed every hour is still stale if the vendor's manufacturing report is two weeks old.
- Plan exports and recovery. The business should be able to reconstruct relationships and decisions if a provider changes. A folder of PDFs without identifiers, revisions and links may be inadequate even though all the documents appear present.
- Appendix D, the Source and Configuration Register, records what your evidence and your AI services actually rest on:
Register element What to retain Every source Identity, title, issuer, revision, effective date, retrieval date, scope, owner, access and authoritative location Controlled project document Approval state and its relationship to the relevant configuration External reference The actual publication and the claim it supports Interpretation (separate field) Competent interpretation, reviewer, date and affected decisions, without altering the source text AI service Model or service version, prompt configuration, retrieval index version, tool permissions, evaluation set, release evidence and change history Register governance Who updates it and how affected uses are reviewed - Where a provider does not expose every internal version, document the available identifiers and the limitation. Do not invent a version for audit neatness. A new publication is not automatically an applicable requirement, and an obsolete source should remain traceable while excluded from current use where the responsible owner decides it no longer applies.
- THE DOCTRINE
The architectural goal is modest but consequential: a reviewer can trace a recommendation back to accepted evidence and determine what remains inferred.
Ask when the underlying source last changed, not when the screen last refreshed.
21:00 – 25:00
Write it down, and Section 4
- Before you close the laptop, list the identities your decision chain depends on and mark where each originates and who owns its mapping.
- Note how your company would reconstruct what was known on the day of the last decision in this chain, and which reconciliation would fail if you ran it today.
- Section 4 of your programme is Identities, the state at decision time, reconciliation and the source register.
The reconciliation you are least keen to run is usually the one that matters.
Key Terms
Controlled mapping table
A maintained record linking identities across systems, such as vendor and client tags, with the source, reviewer and effective date of each mapping.
Leakage
Information improperly available during evaluation, such as future status used to predict an earlier outcome.
Decision snapshot
The retained input versions, calculation version and permissions behind a decision, so that what management knew can be reconstructed.
Evidence freshness
Whether the underlying source is sufficiently current for the decision. Dashboard refresh frequency alone does not establish it.
Self-Check — Before You Submit
1. A fictional EPC contractor joins its vendor register and its client equipment list wherever the tag text matches exactly. What does the book say about this?
- A. It is acceptable once a graph database is used
- B. Equal text strings do not mean equal objects; maintain controlled mapping tables with source, reviewer and effective date
- C. It is acceptable if the dashboard refreshes hourly
- D. It is the correct first step before defining identities
B. Equal text strings do not mean equal objects; maintain controlled mapping tables with source, reviewer and effective date — A vendor tag can resemble a client tag while referring to a different assembly. The graph database answer is tempting, but the book calls it optional; the problem is identity, not storage.
2. In a retrospective test, a delivery forecast dated Tuesday is recomputed using a receipt that occurred on Monday but was entered on Wednesday. What is the problem?
- A. None, because the receipt did occur on Monday
- B. The receipt should be deleted
- C. Leakage: the Wednesday entry was not available on Tuesday, so the recomputed forecast uses information it could not have had
- D. The unit basis is inconsistent
C. Leakage: the Wednesday entry was not available on Tuesday, so the recomputed forecast uses information it could not have had — Event time and recording time must be kept separately; a Tuesday forecast must not use the Wednesday entry. The tempting answer treats event time as decisive, which flatters the forecast.
3. A procurement dashboard shows blank delivery dates for twenty lines. One is pending client approval, some are unknown and some are not applicable. What should the data service do?
- A. Record each as a distinct state, since converting them to blanks loses the reason for the exception
- B. Fill blanks with the average lead time
- C. Leave the blanks, since a blank is neutral
- D. Remove the lines from the report
A. Record each as a distinct state, since converting them to blanks loses the reason for the exception — Unknown, confirmed no date, pending client approval and not applicable are distinct states. Filling blanks with an average looks helpful, but it converts missing evidence into an invented estimate.
Today’s Assignment
Set out the identities your decision chain depends on, how your company preserves the state at the time of decision, the reconciliation required before modelling and who operates the data service.
- Identities. The identifiers your chain depends on (tags, document numbers, purchase order lines, work packages, systems), where each originates, which relationships are authoritative and who owns the mapping tables.
- State at decision time. How your business separates event time from recording time, retains revisions and keeps a decision snapshot, or honestly, that it does not.
- Reconciliation. The deterministic checks required before any model: quantities against receipts, returns and remaining; hours and quantities against progress rules; currency with a conversion basis. State the tolerance and what happens to an unexplained difference.
- The data service and register. Owners for definitions, interfaces, quality incidents and access; how freshness is measured at the source; and the entries your source and configuration register needs for this chain.
- No AI in use yet is normal; these foundations come first anyway. Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 4 — Identities, the state at decision time, reconciliation and the source register.
How this is marked — the rubric
| Lists the identities the decision depends on with origin and a named mapping owner by role | 25 |
| Explains how the state at decision time is or is not preserved, including event and recording time | 25 |
| Specifies deterministic reconciliations, tolerances and the treatment of unexplained differences | 25 |
| Names owners for the data service and the source register entries, with freshness measured at the source | 25 |
| 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
Identity before integrationPreserving decision-time evidenceReconciling before modellingOperating a data serviceKeeping a source and configuration register
Part One · Leadership and the Enterprise FoundationSESSION 5
AI Methods That Fit EPC Decisions
Why This Matters
Executives are seldom the people who build models, but they are the people asked to approve them. The question that protects your company is not which algorithm is best. It is whether the method fits the decision, whether it was compared fairly with something simpler and whether its limits are controlled. Today you learn what a forecast, a retrieval and drafting service, an image interpreter and an optimiser each can and cannot tell you, and you work through the book's Case 9 to see how a fair evaluation separates measures that a single percentage would blur. A prompt demanding confidence cannot create evidence.
25-Minute Module
0:00 – 6:00
Match the model to the question, and forecast with uncertainty
- A useful technical brief states input, output, decision horizon, uncertainty and failure consequence before naming an algorithm. Predicting a vendor delivery date, extracting a specification clause and selecting a feasible installation sequence are different tasks.
- For delivery prediction, a baseline might use accepted lead time and current milestones. For clause retrieval, a controlled search may outperform a generative summary. For installation sequencing, a solver may be appropriate once precedence and resource constraints are represented.
- THE DOCTRINE
Compare against baselines using the same information cut-off. A model evaluated with richer inputs than the comparator does not isolate the value of its algorithm.
- Forecasting with uncertainty, in plain language, means giving a range of likely outcomes rather than one date, because mobilisation depends on the downside. Calibration asks whether outcomes the service calls likely actually happen about as often as it says. Evaluate prediction intervals and calibration alongside point error; an interval so wide it never excludes the observed result can be well covered but operationally unhelpful.
- Distinguish variation in the process (contingency) from uncertainty about the model, for example a new package with little comparable history (more evidence or limited use). Do not pool all projects without examining scope, contract, geography and data collection: a model may learn a project's reporting style rather than the physical behaviour it should predict.
Ask what the forecast is being compared with before you ask how accurate it is.
6:00 – 13:00
Retrieval, images and solvers
- Retrieval and generation. A retrieval service selects permitted source passages; a language model produces an answer using them. Evaluate the two separately. The relevant passage may be absent, incorrectly retrieved, or accurately retrieved but misrepresented. A displayed citation can point to a real document without supporting the claim.
- Require source identity, revision and section for factual answers. When evidence conflicts, show the conflict and route it to authority. Use deterministic calculations for quantities and money, and let the assistant explain a reviewed calculation. The book cites the NIST Generative AI Profile (NIST AI 600-1, July 2024), which identifies risks including confabulation and information-security concerns; the controls proposed are not a claim that a framework certifies an application.
- THE DOCTRINE
When support is absent, abstention is an acceptable output. A prompt demanding confidence cannot create evidence.
- AT THE WORKFRONT · THE WORKED CASE
- Workfront. A fictional EPC contractor's site team photographs installed equipment and a vision service flags an anomaly. Camera angle, lighting and occlusion can change performance, so the flag should lead to competent assessment, with the original image, capture time and permissions preserved.
IN THE COMMITMENT · THE PARALLEL- Commitment. The same photograph offered as support for an installed-quantity claim proves less than it appears to. It may not establish serial number, hidden connections, inspection status or approved installation. An image can support an observation without establishing acceptance, and a machine finding must not become an undisclosed inspection result.
- Optimisation searches the represented problem. If isolation, access or inspection constraints are missing, the answer can be mathematically optimal and operationally unacceptable. Separate mandatory constraints from preferences and costs, test feasibility independently and report the objective, constraint set, solver status and unresolved assumptions. If no feasible plan exists, the useful answer is the conflict requiring a decision; do not relax a mandatory condition merely to return a schedule.
A citation shows where an answer looked, not that the answer is right.
13:00 – 21:00
Case 9: evaluating a retrieval and drafting service
- The book's casebook uses explicitly assumed inputs; this is a teaching calculation, not reported performance. Assume a controlled evaluation with forty answerable questions and ten for which the approved source set contains no answer.
Measure (assumed inputs) Count Result Evidence retrieved, answerable questions 36 of 40 90% Answer materially correct and supported 33 of 40 82.5% Evidence retrieved but material condition omitted 3 of 40 Separate error record Remaining answerable cases (40 − 33 − 3) 4 of 40 Need examination; not assumed acceptable Correct abstention, unanswerable questions 8 of 10 80% Invented answers, unanswerable questions 2 of 10 Must stay visible - Combining all measures into one percentage can obscure the two fabricated answers. Define the reference through qualified review; some questions may have conflicting evidence rather than one accepted answer, and the correct response may be to report the conflict and route it to authority.
Effort (assumed) Per case Fifty comparable cases Current method 30 minutes 1,500 minutes Service: 2 minutes preparation + 20 minutes review and correction 22 minutes 1,100 minutes Simple difference 8 minutes 400 minutes, or 6 hours 40 minutes - The book calls the 400 minutes a teaching estimate, not demonstrated productivity. Freed time that is not redeployed or removed is not automatically a cash saving, and greater cognitive load or concentrated risk for reviewers makes the time measure incomplete. Observe actual work and accepted outcomes in shadow operation before making a business claim.
- The decision may be to narrow the service to retrieval while improving drafting: useful capability retained without authority the evaluation does not support. That is the Autonomy Ladder from Session 1 applied to evidence.
Never let one headline percentage stand in for a fabricated answer.
21:00 – 25:00
Write it down, Section 5, and Practice Exam 1
- Before you close the laptop, write your decision chain's question as input, output, decision horizon, uncertainty and failure consequence, then the simplest credible baseline it must beat.
- Note which method fits: forecasting with uncertainty, retrieval and generation, document interpretation, optimisation, or arithmetic and rules alone. Then note how you would evaluate it with the same information cut-off and which measures you would keep separate.
- Section 5 of your programme is The method that fits our decision and how it would be evaluated.
- Practice Exam 1 opens now: fifteen questions, closed book, covering Sessions 1 to 5 and closing Week 1.
The best method for your decision may be the one that needs no model at all.
Key Terms
Prediction interval
A range intended to represent uncertainty about a predicted outcome under stated modelling assumptions.
Calibration
Agreement between predicted probabilities and observed frequencies across an appropriate population, assessed with uncertainty and sufficient data.
Retrieval
Selection of source material relevant to a query within the permitted information set, evaluated separately from the answer a language model writes from it.
Shadow operation
Use alongside the existing process without granting the proposed service its intended consequential authority.
Self-Check — Before You Submit
1. In the book's Case 9, why should the three answers that retrieved the evidence but omitted a material condition receive a separate error record?
- A. Because retrieval failed in those cases
- B. Because they are counted as abstentions
- C. Because they reduce the retrieval rate below 90%
- D. Because retrieval succeeded while the answer still failed, so a combined score would hide the drafting error
D. Because retrieval succeeded while the answer still failed, so a combined score would hide the drafting error — Retrieval succeeded in 36 of 40 cases including these three, while supported answers were 33 of 40. Assuming retrieval failed is the tempting error; the failure is in the generated answer, which is why the two are evaluated separately.
2. A fictional EPC contractor's scheduling optimiser returns no feasible plan because a shutdown window cannot accommodate the hook-up. A planner proposes relaxing the isolation constraint so the solver returns a schedule. What should leadership do?
- A. Approve the relaxation to obtain a schedule
- B. Report the conflict requiring a decision; do not relax a mandatory condition merely to return a schedule
- C. Replace the optimiser with a generative model
- D. Average the infeasible and feasible plans
B. Report the conflict requiring a decision; do not relax a mandatory condition merely to return a schedule — If no feasible plan exists, the useful answer is the conflict. Relaxing the constraint is tempting because it produces an answer, but it can be mathematically optimal and operationally unacceptable.
3. A vendor shows that its delivery model beats your current forecasting method, but its model used manufacturing status updates your planners did not have. What is the right conclusion?
- A. The algorithm is proven superior
- B. The comparison is invalid only if the vendor is untrustworthy
- C. The comparison does not isolate the algorithm's value; compare at the same information cut-off, and recognise that better information sharing may be the main improvement
- D. Pool all projects to increase the sample
C. The comparison does not isolate the algorithm's value; compare at the same information cut-off, and recognise that better information sharing may be the main improvement — A model evaluated with richer inputs than the comparator does not isolate its algorithm. Pooling projects is a tempting fix, but the book warns it may teach the model a project's reporting style.
Today’s Assignment
Match a method to your decision chain's question and state how your company would evaluate it fairly before granting it any authority.
- The question. Your decision written as input, output, decision horizon, uncertainty and failure consequence, before any algorithm is named.
- The method. Which fits: forecasting with uncertainty, retrieval and generation, computer vision or document interpretation, optimisation, or a simpler method of arithmetic and rules. Say why, and what the method must never be the calculator of record for.
- The baseline. The existing process and the simplest credible alternative, compared at the same information cut-off.
- The evaluation. The separate measures you would keep (for example retrieval, supported answers, abstention, calibration or feasibility), the qualified reference, the effort counted in full and the shadow operation before any business claim.
- The honest part. Where your business has been, or could be, persuaded by a single headline score. No AI in use yet is normal; if the right answer is no model, say so. Four hundred to nine hundred words.
YOUR 90-DAY PLAN
Section 5 — The method that fits our decision and how it would be evaluated.
How this is marked — the rubric
| States the question as input, output, horizon, uncertainty and failure consequence before naming a method | 25 |
| Chooses a method with reasons, including a simpler alternative, and names what it must not decide | 25 |
| Specifies a baseline compared at the same information cut-off | 20 |
| Defines separate evaluation measures, full effort and shadow operation before any value claim | 30 |
| 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 — The Enterprise Foundation 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
Matching method to questionReading forecasts with uncertaintyEvaluating retrieval and drafting separatelyTesting optimiser feasibilityFair baseline comparison