THE INTELLIGENT REFINERY LEADERSHIP PROGRAMME
AI earns its place in a refinery one decision at a time.
20-Session Programme · 25 Minutes Per Day · Free · For leaders of Nigeria’s modular and mega refineries
WEEK 2 OF 4 · MARGIN, FEEDSTOCK AND PRODUCTION · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Refinery Economics · Chapter 4SESSION 6
Margin, Yield and the Executive Operating Review
Why This Matters
Week 1 gave you a configuration sheet and a binding constraint with an owner. This week we put money and material through them. Most operating reviews I have sat in argue about margin without agreeing what it covers. A refinery can report more output while earning less contribution, and production, commercial and finance can each be right about yesterday. Today you learn to draw the financial boundary, connect yield to value, build a plan-to-actual bridge and put the cash dates beside the operating dates. If your margin figure cannot be reproduced by someone outside the team, it is an opinion.
25-Minute Module
0:00 – 5:00
More output, less contribution
- Throughput can rise while contribution falls: prices move, crude costs more, the mix deteriorates, or handling and energy consume the gain. Leadership needs a reconciled account of what changed.
- Three measures get confused. Gross product value is what the outputs are worth; contribution deducts crude and other included costs; accounting profit also takes account of tax, financing, depreciation and the rest.
- THE DOCTRINE
Define the financial boundary before discussing a margin. Gross product value, contribution and accounting profit are different measures, and they do not belong on the same chart without a clear basis.
- An AI application should use your company's agreed definitions, not generate its own plausible version.
Before the next margin discussion, ask which of the three numbers is on the screen.
5:00 – 10:00
Yield, value and the bridge
- Yield is output relative to the stated input, on a specified mass or volume basis. Value depends on the price and costs of those particular outputs.
- A higher percentage of one stream is not automatically better if its realisable netback is lower, or if it cannot be sold within the period.
- EIA, as the book cites it, explains that refining can produce a greater volume of products than the volume of crude input because of differences in density. That processing gain is not the creation of mass. Use one consistent basis for yields and balance closure. The book does not apply U.S. average yield figures to a Nigerian refinery.
- For the executive, show a bridge from planned to actual result. Separate crude quantity, crude price, mix, product netbacks, controllable conversion costs and timing. Where the accounting result also reflects inventory valuation or exchange movements, identify them.
- A process team should not be held responsible for a market-price movement it did not cause.
A bridge with a large bar labelled “other” is unfinished.
10:00 – 17:00
A transparent campaign calculation
- The book's teaching example. Every figure is an assumption; I have recomputed each line.
Line Assumed quantity and price Value Crude received and processed 10,000 t × $600/t delivered $6,000,000 cost Product A 4,000 t × $850/t netback $3,400,000 Product B 2,500 t × $720/t $1,800,000 Residual product 2,500 t × $420/t $1,050,000 Refinery fuel use and other 1,000 t, no separate sale revenue assumed $0 Product value $3,400,000 + $1,800,000 + $1,050,000 $6,250,000 Other included campaign costs Assumed $180,000 Contribution within this boundary $6,250,000 − $6,000,000 − $180,000 $70,000 - That $70,000 is not net profit. Tax, financing, depreciation and other excluded items need their own treatment.
- Now an intervention shifts 200 tonnes from residual into product A, total mass unchanged.
Case Calculation Incremental contribution Product A at $850/t 200 × ($850 − $420) = $86,000, less $30,000 added operating cost $56,000 Product A falls to $700/t 200 × ($700 − $420) − $30,000 $26,000 - THE DOCTRINE
The physical yield gain did not change; the financial case did. A process prediction and a financial case need separate but connected evidence.
- Even the $56,000 awaits evidence on quality, engineering feasibility and the effect on other units.
Ask what netback a yield gain was priced at, and what happens if it falls.
17:00 – 21:00
Cash can reverse a profitable choice
- The base campaign needs $6,000,000 for crude and $180,000 for included costs. If those fall before collections, the refinery must fund up to $6,180,000, on these simplified assumptions, to earn $70,000.
- Positive contribution can still be infeasible if the cash is not there when required. Model supplier credit, customer advances and collection timing as actual obligations, not assumed benefits.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. The campaign record links crude receipt, laboratory evidence, production, tank movements and collection, so the cash trough can sit on one page. The risk is a margin sheet and a treasury sheet that never meet.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. Many units and interfaces make the bridge harder. Inventory valuation and exchange movements must be separated before any unit team is judged.
- THE DOCTRINE
A dashboard that shows margin but omits payment dates leaves treasury with the problem after the campaign has already been approved.
- The review page shows planned and actual feed, released product, delivered product, contribution and cash, the main constraint and the decision requested, with reconciled records and definitions behind it.
- AI can help prepare the bridge, detect inconsistencies and explore alternatives; finance and operations jointly verify the inputs. A fluent explanation is not sufficient if the quantities do not reconcile.
Put the payment dates on the same page as the production dates. The conversation changes immediately.
21:00 – 25:00
Choosing the first application, and your wrap
- The Executive Application: recalculate one recent campaign on the agreed basis, explain the plan-to-actual differences, then set the cash dates beside the operating dates.
- If discrepancies dominate, improve reconciliation first. If the information is sound but alternatives are slow to assess, planning assistance may be useful. If the constraint is funding, arrange funding before buying a tool intended to recommend more campaigns.
- Before you close: write your contribution boundary in one sentence, the bar of the bridge your Session 1 problem sits in, and the date your cash need peaks. That is Section 6.
Many first AI use cases turn out to be reconciliation problems. That is a finding, not a disappointment.
Key Terms
Financial boundary
The stated revenues and costs a margin figure includes and excludes, so that gross product value, contribution and profit are not confused.
Contribution
Product value less crude cost and other costs included within the agreed boundary. It is not net profit.
Processing gain
Extra product volume arising from density differences in refining. A volume effect, not created mass.
Plan-to-actual bridge
A reconciliation from planned to actual result separating crude quantity, crude price, mix, netbacks, controllable conversion costs and timing.
Self-Check — Before You Submit
1. In the book's assumed campaign, 200 tonnes move from residual ($420/t) to product A at an added cost of $30,000. If A's netback falls from $850 to $700, what is the incremental contribution?
- A. $56,000
- B. $86,000
- C. $26,000
- D. $30,000
C. $26,000 — 200 × ($700 − $420) − $30,000 = $26,000. $56,000 is the answer at $850: the physical gain is unchanged, its financial value is not.
2. Production reports record monthly throughput; finance reports contribution below plan. Both figures are checked. What should the review do first?
- A. Build a plan-to-actual bridge on an agreed boundary
- B. Accept the finance figure, because finance owns the accounts
- C. Ask production to explain the contribution shortfall
- D. Commission an AI tool to explain the variance
A. Build a plan-to-actual bridge on an agreed boundary — Both can be true; the bridge shows why. Asking production to explain risks blaming them for a market movement, and a fluent AI explanation is not enough if quantities do not reconcile.
3. A campaign shows contribution of $70,000. Crude and included costs of $6,180,000 fall due before collections, and treasury cannot fund that. What is the correct reading?
- A. Approve it, because contribution is positive
- B. It is profitable, so it will finance itself
- C. The boundary is wrong and should include financing
- D. It may be infeasible until funding or terms change
D. It may be infeasible until funding or terms change — Cash can reverse a profitable choice: feasibility needs the money on the payment date. Redrawing the boundary hides the problem rather than solving it.
Today’s Assignment
Recalculate one recent campaign in your company on an agreed boundary, bridge plan to actual, and set the cash dates beside the operating dates.
- State your contribution boundary in one sentence: what it includes, what it excludes and on which mass or volume basis yields are stated.
- Recalculate one recent campaign on that boundary and bridge plan to actual across quantity, price, mix, netbacks, conversion costs and timing, separating inventory valuation or exchange effects. Label any assumed figure.
- Place the payment and collection dates beside the operating dates, and state the peak funding need and when it falls.
- Connect the result to your Session 1 problem, and say which of the three first-application tests applies: reconciliation, planning assistance or funding.
- The honest part: where does your company mix margin measures, or approve a campaign without cash dates?
- Not at a refinery? Use the campaign your business finances, supplies or reviews. Aim for 400 to 900 words; this becomes Section 6 of your 90-Day Intelligent Refinery Plan.
YOUR 90-DAY PLAN
Section 6 — Our campaign bridge: contribution boundary, plan-to-actual and the cash dates.
How this is marked — the rubric
| States a contribution boundary with explicit inclusions, exclusions and a consistent mass or volume basis | 25 |
| Builds a bridge that separates at least five drivers and keeps market movements apart from controllable performance | 30 |
| Places cash dates beside operating dates and quantifies the peak funding need | 25 |
| Names one real place where the company mixes margin measures or approves without cash dates, and selects a first application from the result | 20 |
| Total · pass mark 70 | 100 |
Submitting session 6
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: 6, 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
Financial boundaryYield to valuePlan-to-actual bridgeCash timingExecutive review page
Feedstock · Chapter 5SESSION 7
Feedstock Quality and Crude Selection
Why This Matters
The crude decision is often the largest commitment a refinery makes, and it is too often argued on one number: the price. The cheapest crude at purchase is not necessarily the cheapest feedstock to process. What you are really buying is a campaign: delivered cost, assay evidence, the products your units can actually make from it, the conversion cost and the cash needed to carry it. Today you learn to compare crudes on one consistent basis, to see which parts of a recommendation rest on evidence and which on estimates, and to govern the acceptance decision so one approval does not quietly become a standing authority.
25-Minute Module
0:00 – 5:00
The cheapest crude is not the cheapest feedstock
- The relevant choice includes delivered cost, assay evidence, expected products, unit capability and the cash needed to carry the campaign. A crude recommendation should explain those relationships before it asks for approval.
- EIA, as the book cites it, identifies density and sulphur content as important crude characteristics affecting processing and product quality. They are useful starting points, not your acceptance criteria. EIA is technical context, not Nigerian law.
- THE DOCTRINE
A refinery's actual acceptance criteria require the facility's approved technical basis. A broad crude category is not a complete description of suitability.
- Buying a generator on its sticker price, without asking what fuel and spares it needs, is the same mistake.
When a crude is presented on price, ask what the campaign earns, not what the cargo costs.
5:00 – 11:00
Buy a campaign rather than a price
- Three teams own the evidence; the executive sees all three on one sheet.
Who Prepares Procurement and commercial The delivered basis: crude differential, freight or inland transport, handling, contractual quantity and quality provisions, payment terms, and the consequences of delay Operations The feasible processing route and its restrictions Laboratory and technical Which assay information is available and how well it represents the proposed cargo - An old assay can be useful history without being adequate evidence for a current acceptance decision. Record its origin, sampling basis and date.
- Where cargo-specific information is missing, name the uncertainty and decide whether further testing, contract protection or a smaller commitment is warranted.
- THE DOCTRINE
AI may compare assays and estimate likely outcomes within a validated range. It should not silently turn missing properties into assumed facts.
- If you cannot see which parts rest on evidence and which on estimates, send it back.
Ask the date of the assay before you ask what it says.
11:00 – 18:00
A consistent basis, and the worked choice
- Use the same campaign size, period and boundary. A delivered price cannot be compared with an ex-loading-point price, nor an all-units-available yield with a plan that includes a known outage, unless the difference is explicit.
- Show every mass, volume and currency conversion, with the source of density and reference conditions. A convenient factor can move a decision while staying invisible in a summary.
- The book's worked choice for a 10,000-tonne campaign. All figures are teaching assumptions, not a view on any grade; Crude A is Session 6's campaign.
Assumed line Crude A Crude B Crude B, higher product value Delivered cost per tonne $600 $580 $580 Feed cost (10,000 t) $6,000,000 $5,800,000 $5,800,000 Expected product value $6,250,000 $6,080,000 $6,160,000 Included conversion costs $180,000 $260,000 $260,000 Expected contribution $70,000 $20,000 $100,000 - On the base assumptions Crude A is preferable by $50,000. B's $200,000 feed-cost saving is more than offset by the assumed lower product value and higher conversion cost. If B's product value were $6,160,000, its contribution would be $100,000 and the ranking would reverse.
- The executive question is why the product-value estimate is uncertain, and whether better evidence could resolve it before purchase. Include a conventional reference method: if your engineering model and planners answer adequately, the AI tool must show what it adds.
One product-value estimate reversed this ranking. Find the fragile estimate in yours.
18:00 – 22:00
Cash, and governing the acceptance
- Both cases need a cash schedule: a cheaper cargo payable immediately may need more working capital than a dearer one on credit. Finance separates the effect of terms from the effect of the crude.
- Separate three things: the model recommendation, the technical suitability (the technical team, against the current configuration and operating envelope) and the commercial commitment (commercial and procurement). The authorised executive approves after finance confirms the funding basis.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. With a narrow processing scope, one unsuitable cargo can be the whole campaign. A smaller commitment or contract protection may be worth more than a better forecast, and limited laboratory staffing may need time to test a cargo properly.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. The feasible route runs through many units. A comparison that assumes every unit available, set against one that includes a planned outage, is not a comparison until the difference is made explicit.
- Record what would invalidate the approval: materially different quality evidence, an unavailable unit or a changed delivery window should each trigger a defined review.
- THE DOCTRINE
An approval given for one campaign should not become a standing authority for every later cargo bearing a similar label.
Write the invalidating conditions into the approval itself.
22:00 – 25:00
Your wrap
- The Executive Application: take the next crude purchase to a joint review with one comparison sheet showing delivered feed cost, evidence quality, feasible products, included conversion costs, contribution range and peak cash requirement. Include the option of delaying or reducing the campaign.
- Require it to name its most important unresolved assumption, fund evidence that could change the choice, and after processing compare predicted with actual.
- If your Session 1 problem is not a feedstock decision, use the feedstock decision it depends on.
- Before you close: write the assumption most likely to reverse your next crude choice. That anchors Section 7.
The option to buy less is a real alternative. Put it on the sheet.
Key Terms
Delivered basis
Crude cost at the refinery, including differential, transport, handling, contract provisions, payment terms and the consequences of delay.
Assay
Laboratory evidence of a crude's properties, judged by its origin, sampling basis, date and how well it represents the cargo.
Approved technical basis
The facility's own approved criteria for accepting a crude within its configuration and operating envelope. General characteristics such as density and sulphur are only a starting point.
Consistent basis
Comparing alternatives on the same campaign size, period, economic boundary and price point, with mass, volume and currency conversions shown.
Self-Check — Before You Submit
1. On the book's assumptions, Crude B costs $580/t for 10,000 t, with product value $6,080,000 and conversion costs $260,000. Crude A earns $70,000. Which is correct?
- A. B, because it saves $200,000 on feed
- B. A, by $50,000
- C. They are equivalent after conversion costs
- D. B, by $30,000
B. A, by $50,000 — B earns $6,080,000 − $5,800,000 − $260,000 = $20,000, so A wins by $50,000. The $200,000 saving is the classic trap: buying on price instead of campaign economics.
2. Commercial proposes a cargo whose only assay is old and from another loading point. The model's recommendation is attractive. What do you require?
- A. Approval, since the model accounted for the assay
- B. Rejection, because an old assay is never usable
- C. The assay's origin, sampling basis and date, and a decision on testing, contract protection or a smaller commitment
- D. The model to fill missing properties from similar crudes
C. The assay's origin, sampling basis and date, and a decision on testing, contract protection or a smaller commitment — An old assay is useful history, not adequate evidence. Rejection discards context; model gap-filling turns missing properties into assumed facts.
3. You approved a crude for one campaign. Procurement wants another cargo of the same name on that approval. What is right?
- A. A new acceptance decision, checking quality evidence, unit availability and delivery window
- B. Proceed, because the grade is approved
- C. Proceed if the price is no higher
- D. Ask the AI tool if the cargoes are similar enough
A. A new acceptance decision, checking quality evidence, unit availability and delivery window — One campaign's approval is not standing authority. Price does not address suitability, and a model's similarity judgement is not technical acceptance.
Today’s Assignment
Prepare the one-sheet comparison your company would take to its next crude-purchase review, on a consistent delivered basis.
- Name the next crude decision in your business and at least two alternatives, one of which is delaying or reducing the campaign.
- For each, show delivered feed cost, evidence quality, feasible products, conversion costs, contribution range and peak cash, with conversions shown.
- Name the single most important unresolved property or assumption, and whether further evidence could realistically change the choice.
- State who decides technical suitability, who makes the commercial commitment, who confirms funding, and what would invalidate the approval.
- The honest part: describe how your company actually chose its last cargo, and where that differed from this sheet. If you do not buy crude, use the feedstock decision your work touches most directly.
- Aim for 400 to 900 words. This becomes Section 7 of your 90-Day Intelligent Refinery Plan.
YOUR 90-DAY PLAN
Section 7 — Our crude-campaign comparison on a consistent delivered basis.
How this is marked — the rubric
| Compares at least two alternatives, including delay or reduction, on the same campaign size, period, boundary and delivered basis | 30 |
| Shows conversions and separates actual evidence from model estimates | 20 |
| Identifies the assumption most likely to reverse the choice and says whether more evidence is worth funding | 25 |
| Separates technical suitability, commercial commitment and funding by named role, with stated invalidating conditions | 25 |
| Total · pass mark 70 | 100 |
Submitting session 7
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: 7, 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
Campaign economicsDelivered-basis comparisonAssay evidenceCrude acceptance governanceWorking-capital view
Feedstock · Chapter 6SESSION 8
Blending and Feedstock Scheduling
Why This Matters
A good campaign can fail before a tonne is processed: the component is late, the receiving tank is full, the transfer clashes, and Monday's plan is fiction by Wednesday. Scheduling is where an economic preference becomes a physical sequence, and most of the surprises I have seen in refinery reviews were visible in the schedule days earlier. Today you learn to give each movement an identity, to treat blend properties and inventory status as evidence, and to plan the delay you already suspect is coming. Then Pop Quiz 2 checks the first three sessions of this week.
25-Minute Module
0:00 – 4:00
From preference to sequence
- A blend that looks attractive on a spreadsheet may be impractical because a component is delayed, a tank is occupied or a transfer conflicts with another movement.
- A useful application connects quantities, locations and time. It shows where each component is now, when it can move, what the receiving arrangement permits and which approvals are outstanding.
- THE DOCTRINE
A recommendation that assumes all planned material has already arrived gives management false confidence. Scheduling turns an economic preference into a physical sequence.
Ask where the material is now, not where the plan says it will be.
4:00 – 10:00
Give each movement an identity
- Link shipment, receipt, tank, sample, blend and campaign with controlled identifiers, not similar names in separate files. One barge named three ways in three spreadsheets is three barges to a computer.
- Preserve quantity adjustments and the reasons for them. A changed vessel estimate should not overwrite the record of what the team used when it approved the original plan.
For each movement, record Why it matters Earliest and latest feasible time Shows the window, not one hopeful date Required route Exposes conflicts with other transfers Available receiving capacity A free tank is a dependency, not an assumption Source of the quantity estimate Tells you how far to trust the number Date status: confirmed, contractually expected, or assumed by the planner Tells leadership where contingency is needed - A scheduling assistant flags conflicts and offers alternatives using the operating team's tank and transfer restrictions. It makes feasible choices easier to compare; it does not replace transfer instructions or permits.
Count how many dates in this week's schedule are actually confirmed.
10:00 – 15:00
Blend properties and inventory status are evidence
- Blending calculations need a technical method suited to the property in question. Do not assume every property combines linearly. The technical basis, component evidence and validation against laboratory results belong with the recommendation.
- For an AI estimator, know the component combinations and conditions it was validated on. A new component or materially different ratio may fall outside; the system should say so and withhold unsupported confidence.
- THE DOCTRINE
Forecasting a blend property informs planning. The controlled sampling and technical acceptance process establishes the evidence for the actual material. Keep that distinction in the workflow and the access permissions.
- A tank can hold volume that is physically present but unavailable: under investigation, awaiting a representative sample, or reserved under an approved commitment. Distinguish physical, technically available and commercially allocated inventory.
- That distinction prevents double allocation. Commercial may nominate material operations has reserved for a blend; planning may assume a component procurement has not secured. A common inventory view should preserve those claims and force the right owner to resolve the conflict.
When two departments count the same tank, the problem is the missing owner.
15:00 – 21:00
A delay scenario worth planning
- The book's fictional schedule: a campaign needs two components. One is available; one is due tomorrow. The plan depends on a receiving tank becoming free after a product dispatch, and that dispatch is awaiting release evidence. Arrival, tank availability and product release are one connected decision.
- Prepare at least three scenarios. The book gives no figures for them, so neither do I; the discipline is in what each must show.
Scenario Each must show Planned sequence Feasible feed, production consequences, extra handling, cash effects Late arrival of the second component The same four, and which commitment becomes unacceptable Delayed tank release The same four, and the release condition still outstanding - Do not represent uncertainty through a single optimistic arrival date. The executive needs to know which commitment fails if the schedule changes.
- THE DOCTRINE
If a system proposes accelerating a dispatch, it must reveal the unresolved release condition. Quality approval is not an administrative delay. Some dependencies can be expedited; others must be satisfied.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. Limited tankage means one held product tank can stop the next crude receipt. The campaign record linking receipt, laboratory, tanks and collection is the schedule; keep it in one place with one set of identifiers.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. Components, intermediates and products compete for many tanks and routes across units. The scenario must show which interface owner authorises each response, and when the choice escalates beyond local authority.
When the tank depends on a laboratory result, that result is on the critical path.
21:00 – 25:00
Count it once, and your wrap
- The benefit of integration appears as fewer avoidable replans and earlier recognition of a genuine choice. Count any financial improvement through the resulting campaign or delivery. Do not add the same avoided disruption again as a separate scheduling benefit.
- The Executive Application: choose one week's feedstock schedule and trace its critical movements. For every planned quantity, identify location, quality status, ownership, route and availability date, and mark assumptions that lack confirmation.
- Then run a controlled scenario in which the most important arrival is delayed. Compare the feasible responses and who can authorise each. If your Session 1 problem sits elsewhere, trace the feedstock movement it depends on most.
- Before you close: write the one movement in next week's schedule whose delay would break the plan. That is Section 8.
- Pop Quiz 2 opens now: five questions on Sessions 6 to 8, ten minutes, closed book. Sit it before you start Session 9.
A scheduling tool earns its place when it produces an earlier, clearer decision the refinery can execute.
Key Terms
Controlled identifier
A managed reference linking shipment, receipt, tank, sample, blend and campaign, so records of the same material join reliably.
Date status
Whether a scheduled date is confirmed, contractually expected or assumed by the planner. It shows leadership where contingency is needed.
Technically available inventory
Material that is not only physically present but cleared for the planned use, as distinct from material under investigation, awaiting a representative sample or reserved.
Validation range
The component combinations and conditions an estimator was tested against. A recommendation outside that range should be flagged and its confidence withheld.
Self-Check — Before You Submit
1. A product tank is full. Commercial has nominated it for a customer. Operations has reserved it for a blend. Part of it awaits a representative sample. What does the schedule need?
- A. One inventory figure for the tank
- B. Whichever claim was entered first
- C. Commercial's allocation, because revenue comes first
- D. Physical, technically available and allocated views, with conflicts forced to the right owner
D. Physical, technically available and allocated views, with conflicts forced to the right owner — Present is not available. One figure invites double allocation, and first-come or revenue-first rules decide by default what an owner should resolve.
2. Your scheduling assistant proposes advancing a dispatch to free a tank. The product still awaits release evidence. What is right?
- A. Accept; release is an administrative step
- B. Require the unresolved release condition shown; release is satisfied, not scheduled around
- C. Ask the laboratory to approve sooner
- D. Stop using scheduling assistants
B. Require the unresolved release condition shown; release is satisfied, not scheduled around — Quality approval is a dependency to satisfy, not a delay to remove. Pressing the laboratory turns a schedule problem into a quality risk.
3. A blend estimator was validated on certain components. Planners now add a new component at a new ratio. What should it do?
- A. Flag it is outside its validation range and withhold confidence
- B. Assume linear blending and continue
- C. Use the average of past blends
- D. Proceed; the laboratory will test later
A. Flag it is outside its validation range and withhold confidence — A new component or ratio may fall outside what was tested. Linear assumptions and averages hide that; later testing does not make an unsupported plan sound.
Today’s Assignment
Trace one week's feedstock schedule in your company to its critical dependency, and plan the delay.
- Choose one week's schedule. List each critical movement with its controlled identifier (or note that none exists).
- For every planned quantity, give location, quality status, ownership, route and availability date, and mark each date as confirmed, contractually expected or assumed.
- Run the delay: describe the planned sequence, a late arrival and a delayed tank release, each with feasible feed, production consequence, extra handling and cash effect.
- Name who can authorise each response, and which commitment becomes unacceptable first.
- The honest part: describe one place where your business counts the same material twice, or schedules around a quality release. If you are not at a refinery, use the schedule your work depends on.
- Aim for 400 to 900 words. This becomes Section 8 of your 90-Day Intelligent Refinery Plan.
YOUR 90-DAY PLAN
Section 8 — One week's feedstock schedule traced to its critical dependency.
How this is marked — the rubric
| Traces each critical movement with location, quality status, owner, route and a date marked confirmed, expected or assumed | 30 |
| Presents three scenarios, each showing feed, production, handling and cash effects | 25 |
| Names the authoriser for each response and the first commitment that fails | 25 |
| Identifies a real double-allocation or release-dependency weakness in the company | 20 |
| Total · pass mark 70 | 100 |
Submitting session 8
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: 8, 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 2 — Margin and Feedstock 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 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Movement identityScenario planningInventory statusRelease dependenciesBlend evidence
Production · Chapter 7SESSION 9
Planning Across Process Units
Why This Matters
Most planning failures I have investigated were not bad arithmetic. They were good arithmetic on a plant that did not exist that week: a unit assumed available, a utility assumed spare. Today is about choosing the right planning method, moving from historical averages to the actual campaign, keeping a reference plan so an AI alternative has something honest to beat, and deciding who may trade a shared resource between competing units. The question to carry into your next planning review is simple: can this plan actually be run, with this plant, in this period.
25-Minute Module
0:00 – 4:00
The plan must carry the whole route
- A crude unit can produce an intermediate another unit cannot currently process. A conversion unit may have capacity but lack a utility or feed. A product pool may be economical in theory while storage and dispatch prevent realisation.
- EIA's description of conversion, as the book cites it, distinguishes cracking and other processes that turn fractions into intermediates and finished products. That is useful context. It does not establish which routes or yields are available at your refinery; those come from your engineering and operating evidence.
- THE DOCTRINE
Refinery planning must respect the connected system. The plan needs to carry the entire feasible route, from crude to dispatched product.
Ask the planner to trace one product back to the crude through every unit.
4:00 – 9:00
Use the right planning method
- Recall Session 2: linear programming needs no machine learning, and a simulator is not automatically AI. Ask which uncertainty each addresses.
Method What it does What it needs to be trusted Optimisation model Compares feasible plans under defined constraints and objectives A current, complete representation of the plant Machine learning Improves selected inputs, such as a response relationship or quality prediction Validation in the conditions now proposed Simulator Tests technical implications of a plan Specified engineering relationships that match the units - An optimiser is only as useful as its representation of the plant. If important restrictions are omitted or obsolete, it produces a mathematically attractive plan that operations cannot run.
- Review the constraint set with engineers and planners: each limit's basis and owner, design values separated from current restrictions, and the planned period, since an annual view can conceal a weekly problem.
A constraint without an owner is a constraint nobody will update.
9:00 – 14:00
From average conditions to the campaign
- Historical averages help screen options, but the next campaign may differ. Feedstock, catalyst condition, equipment availability and product commitments should be represented as they apply to that period.
- THE DOCTRINE
A model fitted across several operating modes may predict an average that is valid in none of them.
- For a learned model, compare performance by mode and period, and ask whether validation includes the conditions now proposed. Use records available at decision time. The application should not receive credit for predicting a result after evidence of that result had entered its input.
- Keep a reference plan prepared through your established method. Compare any AI-supported alternative against it on preparation effort, feasibility, realised economics and the need for later correction.
- Without a reference plan, an AI recommendation competes only with a forecast.
Ask the supplier which records the model could see on the day the decision was made.
14:00 – 20:00
Shared resources and a fictional review
- Hydrogen, fuel gas, steam, power, cooling and tank capacity are shared. The executive question is who can trade between competing uses, and on what basis. A local application should flag when its recommendation changes another unit's commitment, never quietly assume the resource is its own.
- Sometimes the best action is to cut a local rate to preserve a more valuable system result. Distinguish that deliberate allocation from avoidable underperformance, or managers will resist a sensible plan because their scorecard penalises it.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. There may be no hydrogen system at all, and that is fine to say. The shared resources are more likely tank capacity, power and the small team itself; the interface owner may well be the plant manager in person.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. Hydrogen, fuel gas and intermediates cross unit boundaries. The planning review needs a person authorised to resolve the interface, with engineering review of feasible alternatives, and a clear line between local authority and escalation.
- The book's fictional review: a conversion unit is down. A learned planning assistant recommends a crude blend using product values from periods when that unit ran. It beats the conventional plan, but the slate cannot be produced through the available route.
The book's response Why Correct the configuration and rerun the comparison The input, not the arithmetic, was wrong Check whether inputs include current availability The outage should have been visible Check the outage was within validation, or whether it should have refused Refusal is correct outside the range Record a boundary issue So the lesson survives the meeting - THE DOCTRINE
The error is not resolved by averaging the two recommendations. A tool that cannot distinguish feasible from infeasible routes should not receive authority in the planning process.
Make configuration change a supplier acceptance test: take a unit out and see whether the tool notices.
20:00 – 25:00
Your wrap
- The Executive Application: ask for a campaign plan with a constraint sheet and a cash schedule. Operations confirms the route is executable, commercial confirms the product assumptions, treasury confirms the funding dates. Record the unresolved conditions.
- During a pilot, compare recommendations with the reference method before acting. After each campaign, reconcile the result and classify differences as input error, model error, execution variance or market change.
- That classification turns planning into a learning process rather than a competition between attractive forecasts.
- Before you close: name the shared resource that decides your Session 1 problem, and who today can trade it. If nobody can, write that down. That is Section 9.
The most useful planning question I know is who confirmed this route is executable this week.
Key Terms
Reference plan
A plan prepared through the established method, kept so an AI-supported alternative is compared with a real decision process.
Constraint set
The limits a planning model represents, each with its basis, owner and period, separating design values from current operating restrictions.
Shared resource
A utility or capacity several units use, such as hydrogen, fuel gas, steam or tank capacity, needing a named person with authority to allocate it.
Boundary issue
A failure recorded because an application was used outside the conditions it was built or validated for, as when a unit outage was not represented in its inputs.
Self-Check — Before You Submit
1. A planning assistant trained on years when every unit ran recommends a blend while a conversion unit is down. It beats the conventional plan, but the slate cannot be produced. What is right?
- A. Average the two recommendations
- B. Adopt the assistant's plan for its higher contribution
- C. Correct the configuration, rerun, check inputs and validation range, record a boundary issue
- D. Retrain on more historical data
C. Correct the configuration, rerun, check inputs and validation range, record a boundary issue — An infeasible plan is not fixed by averaging. More data from running periods would not represent the outage; the fault is configuration and validation range.
2. Which statement about planning methods is correct?
- A. Linear programming is a form of machine learning
- B. An optimiser can compare feasible plans without machine learning
- C. A simulator learns relationships from data
- D. An AI-labelled tool should replace the reference plan
B. An optimiser can compare feasible plans without machine learning — The three methods are complementary. Calling linear programming AI is a naming error that leads to buying and governing the wrong capability.
3. A unit manager objects to a plan that cuts his rate to protect a more valuable system result, because his scorecard will show underperformance. What do you do?
- A. Record it as deliberate allocation, separate from avoidable underperformance
- B. Overrule him; scorecards are a personnel matter
- C. Keep his unit at full rate
- D. Let the model decide the allocation
A. Record it as deliberate allocation, separate from avoidable underperformance — A sensible plan is resisted if the scorecard penalises it. Overruling leaves the incentive intact; a model removes the authorised person who should trade between uses.
Today’s Assignment
Describe how your company plans across units today, the reference plan it keeps, and the shared resource that decides your live problem.
- Name the planning method your business actually uses (spreadsheet, linear programming, simulator, learned model or a mix) and which uncertainty each addresses.
- List the five most important constraints in the current plan, each with its basis, owner and whether it is a design value or a current restriction.
- Say whether a reference plan exists and how an AI alternative would be compared with it.
- Name the shared resource that decides your Session 1 problem, who can trade it today and how a deliberate allocation shows on the unit scorecard.
- The honest part: one plan your company approved that could not be run as written, and what it missed. Not at a refinery? Use the planning decision your work depends on.
- Aim for 400 to 900 words. This becomes Section 9 of your 90-Day Intelligent Refinery Plan.
YOUR 90-DAY PLAN
Section 9 — Our planning method, reference plan and the shared resource that decides it.
How this is marked — the rubric
| Names the actual planning method and correctly distinguishes optimisation, machine learning and simulation | 20 |
| Lists constraints with basis, owner and design-versus-current status | 25 |
| Defines a reference plan and the measures for comparing an alternative against it | 25 |
| Names the deciding shared resource, its authorised trader by role and how deliberate allocation is recognised, plus one real infeasible plan | 30 |
| Total · pass mark 70 | 100 |
Submitting session 9
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: 9, 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
Planning method choiceConstraint-set reviewReference planShared-resource authorityConfiguration testing
Production · Chapter 8SESSION 10
Yield and Product Quality
Why This Matters
Yield is a percentage; value is what you release, deliver and collect. Off-specification material may need correction, segregation, another outlet or a lower price, each costing money and tank space. This is also where AI is most tempting and most dangerous: a quality estimator that is right most of the time invites people to treat its prediction as the release. Today you learn where prediction may advise and where it may not decide, how to investigate a quality event properly, how to price a correction, and how to keep a learning record that does not flatter itself. Then Practice Exam 2 closes the first half.
25-Minute Module
0:00 – 4:00
Measure released value, not isolated yield
- An improvement in yield is useful only within the refinery's product and operating requirements. Leadership should measure the value of saleable, released products, not maximise an isolated yield percentage.
- The product pool has an economic context. More of one grade can displace another stream, consume a constrained component or add handling.
- THE DOCTRINE
The technical team establishes feasibility; commercial establishes realisable value. An executive review should bring those assessments together.
When yield is reported, ask how much of it was released and delivered.
4:00 – 10:00
Predict quality without replacing release evidence
- A quality estimator may flag a likely deviation earlier. It needs a defined product, operating range and validation against reference tests, and should show its uncertainty and when it is unreliable.
- A value inferred from process variables is not automatically a tested property of the dispatched batch. Keep its source, time basis and quality status visible, especially where a summary mixes tested, inferred and manually entered values.
- THE DOCTRINE
A prediction may advise earlier action. It does not replace the laboratory result or the established release approval.
- For the Nigerian refinery, keep the currently applicable regulatory, contractual and internal specifications in a controlled register; grade names alone are insufficient. The book names no specification values, and neither will I.
Register field What it holds Requirement The property and its limit Test method How it is measured Source Regulatory, contractual or internal Revision Which version is current Responsible person Who confirms applicability - Do not let a language model invent a current specification from general memory.
Ask whether the quality number was tested, inferred or typed in. If nobody knows, that is the finding.
10:00 – 15:00
Investigate the full quality chain
- Start with the question that causes work: why was a product held or corrected? Trace the sample, result, operating period, tank and blend components.
- Then ask which it was: a process deviation, a sampling problem, a data mismatch or a genuine change in material quality.
- An assistant can locate records and spot patterns, but must not call a root cause established because a pattern is plausible. An event coinciding with a feed change may also coincide with a sampling or instrument issue.
- Measure the time from evidence becoming available to an appropriate decision, not the time to generate a message. Faster detection has value where it changes an action; poorly targeted alerts create work.
Association is where an investigation starts, not where it ends.
15:00 – 20:00
A worked quality decision
- The book's example. All figures are teaching assumptions: a 2,000-tonne batch needs correction before sale through the intended outlet. I have recomputed each line.
Assumed line Correct and sell Sell to lower-value outlet Quantity sold 2,100 t (2,000 + 100 t component) 2,000 t Product value 2,100 × $760 = $1,596,000 2,000 × $680 = $1,360,000 Correction component 100 × $900 = −$90,000 None Other included costs −$15,000 None in this comparison Result before common original batch cost $1,491,000 $1,360,000 - Correction is preferable by $131,000 within the stated boundary, assuming the corrected batch meets the destination requirements, the component is available and the extra quantity can be delivered. Otherwise the result is conditional. The technical team must validate the correction route; the example gives no operating instruction or blend formulation.
- Cash and timing still matter. Immediate correction spending can delay receipts; a lower-value sale may release tank capacity a larger campaign needs. Where those effects change the choice, show them explicitly, not hidden in a revised netback.
- MODULAR REFINERY · THE WORKED CASE
- Modular refinery. The held batch may occupy the tank the next crude receipt needs, so the tank-release effect can outweigh the $131,000. Limited laboratory staff carry both the correction tests and routine release work.
INTEGRATED (MEGA) REFINERY · THE PARALLEL- Integrated (mega) refinery. The correction component may itself be a constrained intermediate from another unit. Using it here has a cost elsewhere in the pool that the interface owner must confirm.
A correction that wins on paper still has to be technically validated and physically deliverable.
20:00 – 25:00
Protect the learning record, and your wrap
- Preserve original test results, corrections and release decisions. A model later trained on a corrected batch must distinguish the material before and after correction; otherwise a poor campaign looks as if it produced satisfactory quality unaided.
- THE DOCTRINE
The learning record should preserve failed outcomes as well as successful ones, with stable sample and batch identifiers and a reasoned treatment of invalid or superseded measurements.
- The Executive Application: review one repeated quality issue with production, laboratory and commercial; establish correction cost, delayed delivery and lost or recovered contribution, and which earlier evidence could have changed the decision. Pilot that, with release kept in the established approval process and false alerts counted.
- Before you close: for one product, write where prediction may advise and the exact point where it may not release. That is Section 10.
- Practice Exam 2 opens now: fifteen questions covering Sessions 1 to 10, twenty-five minutes, closed book. Sit it before you start Week 3.
A model trained on unrecorded corrections learns that your problems fix themselves.
Key Terms
Release evidence
The tested results and approval, under the established process, that a batch meets its applicable specification. A prediction from process variables does not substitute for it.
Specification register
A controlled record of the applicable regulatory, contractual and internal requirements, each with test method, source, revision and the person responsible for confirming applicability.
Quality chain
The linked sample, result, operating period, tank and blend components behind a product decision, traced to find whether a problem was process, sampling, data or material.
Learning record
Preserved original results, corrections and release decisions, including failures, so that later analysis and model training see what actually happened.
Self-Check — Before You Submit
1. A validated, usually accurate estimator predicts a batch is on specification. The laboratory result is not in, and a truck is waiting. What happens?
- A. Release on the prediction, since the model is validated
- B. Release half now, the rest after testing
- C. Ask the supplier to certify the prediction
- D. Hold release for the laboratory result and established approval
D. Hold release for the laboratory result and established approval — A prediction is not a tested property of the batch, and accuracy is not permission. Partial release and supplier certification both let a prediction stand in for release evidence.
2. On the book's assumptions, correction yields $1,491,000 before common batch cost; the lower-value outlet yields $1,360,000. Which is correct?
- A. Correction wins by $236,000
- B. Correction wins by $131,000, if the batch meets requirements and can be delivered
- C. The outlet wins, having no correction cost
- D. They are equal after the component cost
B. Correction wins by $131,000, if the batch meets requirements and can be delivered — $1,491,000 − $1,360,000 = $131,000, and the result is conditional on technical and delivery conditions. $236,000 ignores the $90,000 component and $15,000 other costs: it compares $1,596,000 with $1,360,000.
3. An assistant notes that every held batch last quarter followed a feed change and labels it the root cause. What is right?
- A. Treat it as association; test sampling, instrument and data explanations first
- B. Accept it and change feed policy
- C. Ignore it; assistants cannot investigate quality
- D. Accept it if the confidence score is high
A. Treat it as association; test sampling, instrument and data explanations first — A plausible pattern is not an established cause; the same events could coincide with sampling or instrument issues. Ignoring the pattern wastes a useful lead, and a confidence score does not turn association into explanation.
Today’s Assignment
Map the quality chain for one product in your company and state where prediction may advise but not release.
- Choose one product, ideally one with a repeated hold or correction, and link it to your Session 1 problem where you can.
- Trace its quality chain: sample, result, operating period, tank and blend components, and say whether recent events were process, sampling, data or material.
- Give its specification register entry: requirement, test method, source, revision, responsible person. If no register exists, say so.
- Price one recent correction or downgrade on a stated boundary, with cash and tank effects shown separately.
- State the exact point where prediction may advise and where release evidence must decide, and who approves release.
- The honest part: describe where your business mixes tested, inferred and manually entered values, or overwrites original results. If you are not at a refinery, use the product decision your work touches.
- Aim for 400 to 900 words. This becomes Section 10 of your 90-Day Intelligent Refinery Plan.
YOUR 90-DAY PLAN
Section 10 — The quality chain for one product and where prediction may advise but not release.
How this is marked — the rubric
| Traces the full quality chain for a named product and classifies the cause type with evidence | 25 |
| Describes a controlled specification register entry, or honestly records its absence | 20 |
| Prices a correction or downgrade on a stated boundary with cash and tank effects separate | 25 |
| Draws a clear line between advisory prediction and release evidence, with the release approver by role | 30 |
| Total · pass mark 70 | 100 |
Submitting session 10
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: 10, 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 2 — Midpoint 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 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Released-value measurementPrediction versus releaseQuality investigationCorrection economicsLearning-record control