THE INTELLIGENT PETROCHEMICAL COMPANY LEADERSHIP PROGRAMME
The decision behind the chemical.
20-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s petrochemical and specialty chemical sector
WEEK 2 OF 4 · PRODUCTION, QUALITY AND THE CUSTOMER · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Part Two · Feedstock and ProductionSESSION 6
Production Planning and Operating Constraints
Why This Matters
In Week 1 you named your decision, wrote your business-model record, learned the correct denominator and defined the material before comparing offers. This week the decision meets the plant. A production plan is a set of commitments against a physical system: material, equipment, utilities, competent people, laboratory capacity and a route to customer acceptance. A spreadsheet of desired quantities is useful, but it is not yet a feasible plan. Today you find the constraint that controls your objective, separate what cannot be traded from what is merely preferred, and work through Worked Case One, in which the feed with more premium output earns less unless the hours it frees are actually used.
25-Minute Module
0:00 – 5:00
Find the binding constraint
- The plain idea first: one stage limits what the whole sequence can deliver. Widening every pipe except the narrowest moves no more product between two tanks. That limiting stage is the binding constraint: a vessel, a utility, a raw material, a release process or a customer acceptance window. It can change across products and periods.
- Map the sequence from material availability to accepted dispatch. A batch business includes preparation, production, cleaning, testing, disposition, packaging and storage. A continuous business includes the downstream capacity for the product mix. A supplier replaces the manufacturing steps with import readiness, receiving, release and delivery.
- THE DOCTRINE
Do not assume that reducing time at one stage releases the same capacity for the whole system. The planning model should show the bottleneck that controls the objective.
- Faster batching may simply fill storage with material awaiting testing. Faster laboratory work may not change dispatch if packaging or transport remains constrained.
When someone promises freed hours, ask which stage those hours flow into next.
5:00 – 10:00
Hard constraints, preferences and time
- A constraint can be a technical limit, a contractual commitment, a funding restriction or an operating rule. Minimising changeovers or smoothing workload are preferences. An algorithm should not trade away a hard technical constraint to improve a weighted score.
Rule Where it comes from Plant condition Approved engineering or operating documentation Delivery obligation The contract Credit limit The commercial and finance arrangement Uncertain delivery date An expectation: marked uncertain, never a guaranteed input - When no feasible plan exists, the system should say so and show the limiting conditions, not quietly relax a rule. Management then decides whether to change a commitment, obtain resources or approve a technically justified alternative.
- Keep planned start, actual start, processing complete, sample received, release granted and dispatch complete as separate events, with cause codes for any changed sequence. In continuous processes the competent technical team, not the model, establishes how a sample lags the process.
A plan that never reports infeasibility is relaxing a rule somewhere you cannot see.
10:00 – 14:00
Plan against uncertainty and formulate before buying
- Present a base plan and its credible disruptions without invented precision, and include cash: advancing production may build inventory before collection.
- Write the problem down before choosing software. Decision variables are what the planner can choose: quantities, assignments or start times. The objective might be contribution from accepted dispatch within a period. Constraints cover material balances, equipment occupancy, approved routes, commitments and resource limits. A sophisticated label, such as mixed-integer optimisation, does not prove the model holds every relevant constraint.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional batch formulator schedules with a fixed average changeover. Cleaning time really depends on the preceding and following products, so the average hides a sequence that cannot be run.
IN THE MARGIN · THE PARALLEL- Margin. A fictional continuous producer's model proposes a higher upstream rate. Finance tests the whole path to accepted output: extra intermediate inventory consumes working capital and storage without creating a sale. Different plants, one question: what reaches the customer.
- A simpler model with a correct occupancy record can be more useful than a sophisticated solver built on nominal capacity.
If the vendor cannot show you the constraints in words, the solver is not your first problem.
14:00 – 21:00
Worked Case One: feedstock choice in a constrained campaign
- The book labels this case hypothetical: every figure is assumed, both feeds are assumed qualified for the same process, and the response estimates come from a hypothetical engineering review, not a validated AI model. The window is sixty hours. Feed F1 occupies all sixty; F2 occupies fifty-two. Each consumes 100 tonnes. Premium product sells at an assumed ₦1.10 million per tonne, the alternative grade at ₦0.60 million.
Main campaign (assumed) Feed F1 Feed F2 Premium / alternative / loss, tonnes 90 / 7 / 3 94 / 4 / 2 Revenue, ₦ million 99.0 + 4.2 = 103.2 103.4 + 2.4 = 105.8 Feed + conversion + disposal, ₦ million 73.0 + 12.0 + 0.3 = 85.3 76.0 + 12.4 + 0.2 = 88.6 Contribution, ₦ million 17.9 17.2 - On the main campaign alone, F1 is better by ₦700,000 despite four fewer premium tonnes. Now assume a supplementary campaign delivers 32 accepted tonnes in the freed eight hours at ₦50,000 each, or ₦1.6 million, and the scheduling programme costs ₦400,000. F2 then gives ₦18.4 million against ₦17.9 million: an assumed ₦500,000 gain, conditional on using the time. With no supplementary order F2 gives ₦16.8 million and loses ₦1.1 million.
- Break-even: ₦1.1 million ÷ ₦50,000 = 22 tonnes, so 32 gives ten tonnes of headroom; at 20 tonnes the decision is ₦100,000 worse than F1. In the downside case F2 yields 92 premium and six alternative tonnes: contribution ₦16.2 million, ₦17.4 million after supplement and programme cost, ₦500,000 below F1. These are sensitivities, not probabilities.
The question is never which feed is better; it is which complete plan is better, and under what conditions.
21:00 – 25:00
Close the campaign decision, and your wrap
- The evidence separates the estimate available at the planning meeting from later measurements, and tests against an engineering baseline on a later period and unfamiliar qualified lots.
- THE DOCTRINE
Choose F2 only when the response evidence supports it, the supplementary campaign is confirmed and funded, and the schedule fits the window. No software may alter the operating envelope or release product through this decision.
- Afterwards, reconcile the actuals. A correct recommendation that did not lead to the supplementary campaign has not earned the assumed ₦500,000; record the cause, whether order, material, release, funding or a wrong estimate.
- Before you close: write the binding constraint, the hard constraints with their owners and the problem as variables, objective and constraints for your Session 1 decision. That is Section 6 of your 90-Day AI Leadership Plan.
The cause record tells your next investment review whether to fix the model, the workflow or the commercial arrangement.
Key Terms
Binding constraint
The vessel, utility, material, release step or acceptance window that controls what the whole system can deliver. It can change with product mix and period.
Constraint register
Each planning rule with its source and owner, keeping hard constraints apart from preferences and marking uncertain inputs as uncertain.
Decision variables
The quantities, assignments or start times a planner can choose. With an objective and constraints, they define the problem before software is selected.
Robust plan
A plan designed to remain feasible under a stated set of variations; not proof that every disruption is covered.
Self-Check — Before You Submit
1. A fictional formulator's planning tool cannot fit all confirmed orders, so it shortens the cleaning time between two products. What should the head of operations require?
- A. Accept the plan, because it scored highest
- B. The tool reports infeasibility and the limiting conditions, so management can decide
- C. More computing power
- D. Drop two products without telling sales
B. The tool reports infeasibility and the limiting conditions, so management can decide — The system must say no feasible plan exists, not quietly relax a hard technical constraint. Accepting the higher score is the trap: a weighted score must never buy feasibility by trading away a technical limit.
2. In Worked Case One, with no supplementary order, how does F2 compare with F1 after the ₦400,000 programme cost?
- A. F2 gains ₦500,000
- B. F2 loses ₦700,000
- C. F2 loses ₦1.1 million
- D. They are equal
C. F2 loses ₦1.1 million — ₦17.2 million less ₦400,000 is ₦16.8 million against ₦17.9 million. The ₦700,000 answer forgets the programme cost; the ₦500,000 gain exists only if the freed hours are used.
3. How many supplementary tonnes at ₦50,000 each does F2 need to break even against F1?
- A. 22 tonnes
- B. 20 tonnes
- C. 32 tonnes
- D. 10 tonnes
A. 22 tonnes — The supplement must cover ₦700,000 plus ₦400,000: ₦1.1 million ÷ ₦50,000 = 22. Thirty-two is the assumed volume, ten tonnes of headroom above break-even.
Today’s Assignment
Write Section 6 of your plan: the binding constraint, hard constraints versus preferences and the problem formulated.
- Map your company's sequence from material to accepted dispatch for your Session 1 decision and name the binding constraint. A business with no production uses import readiness, receiving, release and delivery.
- Write a short constraint register: hard constraints with source and owner by role, preferences listed separately, and what happens today when no feasible plan exists.
- List which time events your records keep separate and which they collapse.
- Formulate the problem as decision variables, objective and constraints, with the credible disruptions and their cash effect, before naming software.
- The honest part: where your company plans on nominal capacity, an average changeover or an unconfirmed input. If no AI is in use yet, that is normal; the right answer may be a simpler method. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 6 — The binding constraint, hard constraints versus preferences and the problem formulated.
How this is marked — the rubric
| Names the binding constraint from a mapped sequence and whether it moves | 25 |
| Separates hard constraints, each with source and owner by role, from preferences | 25 |
| Formulates variables, objective and constraints, including cash, before software | 30 |
| Gives an honest example of nominal capacity or an uncertain input treated as fact | 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
Binding-constraint diagnosisConstraint registersProblem formulationBreak-even readingConditional approval
Part Two · Feedstock and ProductionSESSION 7
Yield, Formulation and Process Performance
Why This Matters
Session 6 placed your decision inside a constrained plan. Today the question is whether a claimed improvement in yield, conformity or process performance is real. A model may find a relationship between feed properties and product quality, or between batch history and rework. That relationship is evidence for investigation; it does not establish that changing the variable will cause the outcome. You will preserve the physical accounting, keep the approved recipe visible, price the whole improvement, and read soft sensing, probability and thresholds in plain language. Worked Case Two then shows a warning model whose accuracy barely beats doing nothing and whose financial case reverses under one changed assumption.
25-Minute Module
0:00 – 5:00
Preserve the physical accounting and the recipe
- Record input, output, inventory changes and identifiable losses on a consistent basis within a defined balance boundary. In a reaction, the technical team accounts for the actual process and participating materials. A formulator records every addition and withdrawal, including approved rework and retained material; an apparent loss may be measurement, transfer residue, packaging difference or real loss.
- THE DOCTRINE
A model should not repair the balance by inventing an unmeasured quantity. The approved reconciliation preserves the measured values, the assumptions and the unresolved difference.
- A formulation needs an identifiable version: what was authorised, what was made and any approved deviation. Predicting a need for investigation differs from recommending a recipe change, and searching validated formulations differs from generating a new one; each needs its own acceptance criteria and authority.
- Protect proprietary formulation data. Technical service may need a performance description without the full manufacturing formula.
If the batch record cannot say what was actually made, no model can say why it turned out as it did.
5:00 – 10:00
Evaluate beyond the fit and price the whole improvement
- A model predicting final quality must not use a post-release status or a later rework instruction as an input. That is target leakage: it can produce impressive evaluation and poor operating use. Time-based evaluation is often relevant; examine performance across material sources, recipe versions and conditions; a model that succeeds on familiar batches may fail on a new lot.
- Both error directions cost something: an unnecessary warning consumes investigation and may delay a valid batch; a missed warning may let a developing issue through. The threshold depends on the decision and existing safeguards.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional formulator recovers a batch after investigation. The technical owner first checks which recipe version was used and whether the apparent loss was transfer residue.
IN THE MARGIN · THE PARALLEL- Margin. Finance asks what the batch would have earned under the baseline. If it would have sold at a lower permitted grade, the benefit is the difference between the feasible alternatives after their costs, not the batch's full value.
- Freed vessel time earns contribution only if it is used for feasible, profitable work. A specialist supplier uses procurement suitability and accepted application performance rather than manufacturing yield.
Unused capacity is an operating improvement, but not automatically an earnings gain.
10:00 – 14:00
Soft sensing, optimisation and explanations
- A soft sensor estimates a property that is unmeasured or measured late, from inputs available now; its initial executive purpose in the book is earlier investigation or planning. Process optimisation chooses an operating alternative under a model and constraints. A prediction of poor quality does not identify a valid adjustment.
- Fix the target property, method, unit, sample identity and decision time first. Acceptance shows error in the property's unit, bias, behaviour near the limits and the share of cases in scope; a stale input should produce an understood abstention. Lawrence and colleagues' survey of machine learning for industrial sensing and control (Control Engineering Practice, 2024) distinguishes work on simulation or industrial data from reported industrial implementation. A research result is not demonstrated operation at your site.
- THE DOCTRINE
A variable ranking describes the fitted model. It does not establish that changing the highest-ranked variable will produce the desired outcome; that needs an engineering explanation and an approved trial.
- If a relationship disappears once recipe revisions are separated, the programme has found a record problem, not a process improvement.
Ask whether the team is predicting a property or proposing to change the process. They are two projects.
14:00 – 21:00
Worked Case Two: early review of a quality exception
- The book states this case is entirely hypothetical; no model was trained. It recommends earlier technical review; testing and authorised disposition are unchanged. Of 600 assumed records, 360 develop the model, 120 select the threshold and 120 stay locked; 600 is not a recommended minimum.
- Precision asks: of the warnings raised, how many were real? Recall asks: of the real exceptions, how many were warned? In the final 120 batches, 18 later need review.
Assumed, same 120 batches Model Existing rule Reviews recommended 27 30 Exceptions warned / missed 15 / 3 12 / 6 Precision 15 ÷ 27 ≈ 55.6% 12 ÷ 30 = 40% Recall 15 ÷ 18 ≈ 83.3% 12 ÷ 18 ≈ 66.7% ₦10,000 per review, ₦350,000 per miss ₦1.32 million ₦2.4 million - Accuracy is (15 + 90) ÷ 120 = 87.5 percent; never warning scores 102 ÷ 120 = 85 percent while missing all 18. The ₦1.08 million difference less ₦600,000 support gives an assumed ₦480,000. At ₦100,000 per miss the costs are ₦900,000 and ₦570,000; the ₦330,000 saving loses ₦270,000 after support. Safety consequences are excluded and no test is bypassed.
- A probability score is calibrated when, among batches given a similar score, roughly that share prove to be exceptions. The threshold is where a score becomes a warning. Guo and colleagues (On Calibration of Modern Neural Networks, 2017) showed strong classification need not mean calibrated confidence, on image and document data only. Under the simplified costs, if review reliably avoids the stated consequence, ₦10,000 ÷ ₦350,000 suggests reviewing above about 2.86 percent: an arithmetic illustration, not an approved threshold.
Eighteen assumed exceptions cannot establish coverage of every material condition. Ask for the uncertainty, not just the rate.
21:00 – 25:00
What the pilot has earned, and your wrap
- The case supports a shadow evaluation and, if its gates are met, controlled earlier review; nothing for automated recipe changes or release. Never tune the threshold repeatedly on the final test and then call it independent. If reviewers lack capacity, a warning can join a queue and leave the late cost unchanged.
- THE DOCTRINE
A stopped pilot can still produce a valuable finding about records or capacity. It should not be labelled a financial success merely because the software identified a plausible pattern.
- Before you close: write the accounting boundary, recipe identity, whole improvement priced and what a pilot must earn for your Session 1 decision. That is Section 7 of your 90-Day AI Leadership Plan.
The investment paper shows the sensitivity, never the most favourable consequence presented as fact.
Key Terms
Soft sensor
An estimate of an unmeasured or delayed property from other available inputs, with an explicit purpose and relationship to the required measurement.
Precision and recall
Precision: the share of warnings that concerned a confirmed event. Recall: the share of confirmed events that were warned.
Calibration of probability
Agreement, over suitable cases, between predicted probabilities and observed frequencies; separate from classification accuracy.
Target leakage
Using information unavailable at the decision time, such as a post-release status, which can inflate evaluation results.
Self-Check — Before You Submit
1. A vendor's quality-warning model scores 87.5 percent accuracy against 85 percent for never warning, and calls the gain marginal. What should the quality director ask for?
- A. Higher aggregate accuracy
- B. Exceptions detected, review workload and warning window, against the existing rule
- C. Permission for the model to release batches
- D. A larger training set first
B. Exceptions detected, review workload and warning window, against the existing rule — Never warning scores 85 percent while missing all 18 exceptions; the model catches 15. Aggregate accuracy hides that difference, and the case supplies no evidence for product release.
2. The model recommends 27 reviews, 15 of which concern a confirmed exception. What is its precision?
- A. About 83.3 percent
- B. 87.5 percent
- C. 40 percent
- D. About 55.6 percent
D. About 55.6 percent — Precision is 15 ÷ 27 ≈ 55.6 percent. The tempting 83.3 percent is recall, 15 of 18 exceptions; 87.5 percent is accuracy.
3. A recovered batch would have sold at a lower permitted grade under the baseline. The team claims its full sales value. What is the benefit?
- A. The difference between the feasible alternatives, after costs
- B. The full sales value
- C. Nothing
- D. Full value less software cost
A. The difference between the feasible alternatives, after costs — The avoided loss is the difference between feasible alternatives after their costs. Full value ignores what the baseline earned; nothing ignores a real, smaller gain.
Today’s Assignment
Write Section 7 of your plan: physical accounting, the approved recipe, the whole improvement priced and what a pilot must earn.
- State the balance boundary for your Session 1 decision in your company and where quantities fail to reconcile.
- Explain how your company separates authorised, made and deviated recipes, and which formulation data must stay controlled.
- Price the whole improvement against the feasible baseline, with all costs, and say whether freed time would really be used. A specialist supplier uses suitability and accepted application performance.
- State what a pilot must earn: both error directions, reviewer capacity and the gate to controlled review.
- The honest part: where your company has claimed a gain without its baseline or treated a correlation as a cause. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 7 — Physical accounting, the approved recipe, the whole improvement priced and what a pilot must earn.
How this is marked — the rubric
| Defines the balance boundary and preserves unreconciled differences | 25 |
| Separates authorised, made and deviated recipe identity | 20 |
| Prices the improvement against the feasible baseline with all costs | 30 |
| States error directions, capacity and pilot gates, with an honest example | 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
Material balance disciplineRecipe version controlWhole-improvement pricingPrecision and recallPilot gates
Part Three · Quality and Customer ApplicationSESSION 8
Laboratory Data and Product Release
Why This Matters
Release is where technical evidence becomes a commitment to a customer. A laboratory result supports a specific conclusion about a specific sample under a stated method; it carries no meaning outside that context. AI can identify trends, retrieve approved methods or assemble an exception history. It can also make weak records look coherent. Today you will identify the sample and the question, keep prediction apart from measurement, read measurement uncertainty and process capability in plain language, investigate an exception before disposition, and find where your release queue actually waits. Reduce delay without weakening release.
25-Minute Module
0:00 – 5:00
The sample, the question and the specification
- Record the sampled material, lot or batch, sampling point, time, person or authorised method, and property. Incoming, in-process, final-product and retained samples may serve different decisions.
- The specification carries its version and customer or product basis. A general product range may differ from a contractual requirement; comparing with the wrong one gives a technically neat but commercially incorrect conclusion.
- THE DOCTRINE
Preserve the original result and any authorised correction. If a test is repeated, the first result remains part of the investigation. Deleting inconvenient results weakens the operating record.
- A business without its own laboratory applies the same identity rules to the release evidence it depends on.
Ask of any result on a dashboard: which sample, which method, which specification revision.
5:00 – 10:00
Keep prediction and measurement distinct
- A statistical estimate is not the same event as the required laboratory determination. Any permitted use of a prediction must be explicit; its purpose may be early investigation or planning, not release.
- ISO/IEC 17025:2017 is the international standard addressing the competence of testing and calibration laboratories. It supports the need to evaluate laboratory competence; it does not certify an AI model or show that a laboratory is accredited for every test.
- Ask which measurement remains necessary, which uncertainty matters and who authorises disposition. Overall accuracy may look strong because most batches are comfortably within specification, so judge a model near the limit. A warning issued after the release meeting is not early warning. Never round a borderline value so that it silently changes the disposition.
Comfortable batches can flatter a model. Judge it on the ones near the limit.
10:00 – 16:00
Process capability and measurement uncertainty
- Process capability, plainly: does the process's normal spread fit inside the specification band, and is it centred? NIST's engineering handbook describes it as comparing a stable process's variation with its specification limits, under distributional assumptions and with adequate independent observations.
Assumed illustration, generic units Value Limits; mean; standard deviation 90 to 110; 104; 2 Cp (110 − 90) ÷ (6 × 2) = 1.67 Cpk, upper and lower sides (110 − 104) ÷ (3 × 2) = 1.00; (104 − 90) ÷ (3 × 2) ≈ 2.33 Cpk, the smaller 1.00: the process sits nearer the upper limit - THE DOCTRINE
A capability index is a statement under a statistical model, not a product-release rule. A product can require individual disposition even when the process-level index looks favourable.
- Measurement uncertainty, plainly: how far a reported result can be relied on. NIST's uncertainty guidance treats it as necessary context for judging a result's fitness, including bias, variability and propagation through calculations. In the book's illustration, 9.8 against an upper limit of 10.0 with expanded uncertainty 0.4 gives an interval reaching 10.2, above the limit. That does not decide acceptance; the contractual and technical decision rule does.
- A new instrument, sampling point or method can shift the measured distribution and a model's target. Investigate before declaring a production improvement.
When a result sits near a limit, ask for the decision rule before you ask for the decision.
16:00 – 21:00
Investigate the exception, then find where the queue waits
- An exception triggers an investigation of sample integrity, method, equipment, material history and processing record. Disposition may be authorised release, further evaluation, approved rework, a permitted alternative grade or rejection. A customer concession is recorded with scope and authority, not as a silent change to the specification. An assistant drafts the summary; it does not select the disposition.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional specialty formulator breaks release time into collection, transport, receipt, analysis, investigation, review and disposition, and finds two hours lost locating a missing dossier. A retrieval assistant helps there; it does nothing for an overloaded instrument.
IN THE MARGIN · THE PARALLEL- Margin. Finance asks what the saving changes. An assumed cut from 48 hours to 24 does not release a day's cash if delivery and payment schedules stay the same; with weekly acceptance it depends on crossing a dispatch cutoff.
- THE DOCTRINE
Begin by examining where the delay occurs. If the laboratory is waiting for a missing batch identifier, another predictive model may add little value.
- More warnings can lengthen the laboratory queue. State the benefit in the category actually established: faster reporting can improve service without earning contribution.
Reporting every saved hour as additional production exaggerates the result.
21:00 – 25:00
Your wrap, and Pop Quiz 2
- Before you close: write the sample and question, where prediction stays apart from measurement, how exceptions are investigated and where your release queue waits. That is Section 8 of your 90-Day AI Leadership Plan.
- Without your own laboratory, use the supplier certificates or customer acceptance testing you depend on.
- Pop Quiz 2 opens now: five questions, closed book, on Sessions 6 to 8. Sit it before Session 9, then write your assignment.
Release belongs to a named, authorised person. Everything here gives that person better evidence sooner.
Key Terms
Measurement uncertainty
Necessary context for judging a result's fitness, stated under the laboratory's declared method and coverage basis; it includes bias, variability and propagation through calculations.
Capability index
A statistical comparison of process spread and centring with specification limits under stated assumptions. Not a release authorisation.
Disposition
The authorised decision on material or product after the relevant evidence and review.
Decision rule
The contractual and technical rule for judging a result and its uncertainty against a limit.
Self-Check — Before You Submit
1. A result of 9.8 against an upper limit of 10.0 carries expanded uncertainty of 0.4. Production wants to ship today. What should the quality director do?
- A. Release, because 9.8 is below the limit
- B. Reject, because the interval reaches 10.2
- C. Apply the agreed decision rule for uncertainty, or seek competent clarification if none exists
- D. Use a model prediction to break the tie
C. Apply the agreed decision rule for uncertainty, or seek competent clarification if none exists — The interval crossing the limit does not itself decide acceptance; the decision rule does. Releasing or rejecting on the number alone is premature, and a model estimate never substitutes for measurement.
2. In the assumed illustration Cp is 1.67 but Cpk is 1.00. What does the gap show?
- A. The process sits nearer the upper limit
- B. The batch must be rejected
- C. The laboratory is unaccredited
- D. The process is proven stable
A. The process sits nearer the upper limit — Cpk accounts for centring; with a mean of 104 in a 90 to 110 band, the upper side gives 1.00. Neither index proves stability or decides an individual batch.
3. A tool halves release time from an assumed 48 hours to 24, and finance books a day of released cash. Customers accept deliveries weekly. What is the right challenge?
- A. Accept it
- B. Does the reduction cross a dispatch cutoff and change delivery or payment timing
- C. Ask the laboratory to work faster
- D. Reject it, release cannot be accelerated
B. Does the reduction cross a dispatch cutoff and change delivery or payment timing — If delivery and payment schedules are unchanged, the reduction does not automatically release cash. Rejecting outright is also wrong: complete records, sample logistics and a dependable approval path can shorten release.
Today’s Assignment
Write Section 8 of your plan: prediction kept apart from measurement, exceptions investigated and release protected.
- Identify the sample, property and specification revision behind your Session 1 decision in your company, and how original results are preserved.
- State where a prediction could help, which measurement must remain, and who authorises disposition by role.
- Describe how exceptions are investigated before disposition and how concessions are recorded.
- Break release time into stages, name where it waits, and state what faster release would and would not change in dispatch and cash.
- The honest part: where your company has rounded a borderline result, overwritten a first result or treated capability as release. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 8 — Prediction kept apart from measurement, exceptions investigated and release protected.
How this is marked — the rubric
| Identifies sample, property and specification revision, with original results preserved | 20 |
| Keeps prediction and measurement distinct and names disposition authority by role | 25 |
| Sets an investigation route before disposition, including concessions | 20 |
| Locates the real release wait and its honest effect on dispatch and cash | 35 |
| 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 — Planning, Yield and Release 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
Sample and specification identityPrediction versus measurementUncertainty and capabilityException investigationRelease queue diagnosis
Part Three · Quality and Customer ApplicationSESSION 9
Technical Service and Customer Applications
Why This Matters
A product can meet its manufacturing specification and still be unsuitable for a particular customer application. For a supplier with technical service, leadership should treat that understanding as part of the product offer. For a producer or formulator without a service team, it is the counterpart you depend on. AI may retrieve technical records, compare trial histories and identify missing questions; it cannot establish suitability from a trade name and an attractive sales description. Today you will define the application, agree the trial and baseline, protect the customer's information and measure service effectiveness. Worked Case Three shows a trial whose average looks good and whose adoption test fails, and an assistant judged on its own contribution.
25-Minute Module
0:00 – 5:00
Define the application and agree the trial
- Begin with the customer's objective, such as improved separation, reduced fouling or reliable water treatment, how it is measured and which conditions affect it. Do not fill missing information with a generic industry profile; that a product worked elsewhere does not prove it will work here.
- A trial needs a written purpose, scope, baseline, technical authorisations, measurements and stop conditions, and names who may change or stop it. Align the commercial terms: who pays for trial material and analysis, what constitutes acceptance, and whether this is a sale, a service trial or a performance commitment.
- THE DOCTRINE
An inconclusive result is a legitimate outcome. It should not be rewritten as success to protect the sales forecast.
A broad promise in a quotation can create an obligation the technical team cannot establish.
5:00 – 9:00
Baseline, information and service effectiveness
- Record existing performance and its conditions, and capture changes during the trial with timestamps rather than averaging them away. An observed outcome is not a conclusion about its cause.
- Agree permitted data use, access, sharing and retention. A supplier should not assume customer operating information can train a service other customers can reach. A customer-facing assistant uses approved documentation, keeps the version and date behind each answer, and escalates unqualified conditions; it never offers unsupported guarantees.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional supplier's engineers log many visits and reports at a customer's site. The technical review asks which trials concluded and which recommendations were accepted within scope.
IN THE MARGIN · THE PARALLEL- Margin. The commercial review asks which results were actually obtained and whether the customer's own maintenance explains part of them. A performance obligation must be priced with its service resources and exposure.
Count trials concluded and recommendations accepted, not visits made.
9:00 – 17:00
Worked Case Three: a trial with an unresolved result
- The book labels this case hypothetical; no formulation, dose or setting is given. A supplier proposes a candidate programme for a customer's oil-and-water separation; the customer wants a lower oil-in-water result at an agreed sampling point. An assistant retrieves evidence and prepares reports; specialists approve the programme. Sales first cite a before-and-after average from a week of customer maintenance, which cannot isolate the programme's effect.
- Blocking, plainly: compare like with like, each existing and candidate result taken under a declared group of comparable conditions. NIST's discussion of randomised block designs describes this; statistical design does not supply permission for an intervention. The assumed acceptance plan requires all six candidate results at or below 20 mg/L, a criterion the book says is invented for the case and is not an environmental discharge limit.
Assumed block Existing, mg/L Candidate, mg/L Meets 20 or below A 18 14 Yes B 22 16 Yes C 30 29 No D 20 15 Yes E 26 21 No F 24 19 Yes - Existing values total 140, average about 23.33; candidate values total 114, average 19.00. The mean paired reduction is 26 ÷ 6 ≈ 4.33 mg/L, about 18.6 percent. These are descriptive, not statistical significance. Two results, 29 and 21, fail; the average of 19 would conceal them. The adoption test is not met. Block E fails despite a larger reduction, 26 to 21 mg/L, so magnitude and absolute outcome both matter.
- THE DOCTRINE
Selecting the favourable blocks after seeing the results and calling them the intended scope would misrepresent the trial. A narrower scope is a new question.
The average is the number most likely to be quoted, and the least likely to answer the acceptance question.
17:00 – 21:00
The customer's economics and the AI component
Assumed month, 5,000 m³ Existing Candidate Material + analysis + service ₦4.0m + ₦0.1m + ₦0.1m ₦3.7m + ₦0.3m + ₦0.25m Total ₦4.2 million ₦4.25 million Per cubic metre ₦840 ₦850 - Material alone shows a ₦300,000 saving; the full comparison costs ₦50,000 more, and ₦650,000 more in the first period with a ₦600,000 setup. Unestablished benefits cannot be added to rescue the case.
- Assume report preparation falls from four staff-hours to two and checking adds half an hour: 1.5 hours saved, ₦22,500 per report at an assumed ₦15,000 per hour, ₦450,000 over twenty reports. At ₦600,000 to operate, it does not pay on preparation time, and the saving is capacity, not cash, unless spending changes.
- THE DOCTRINE
The chemical trial and the document assistant must not borrow each other's success. Each needs its own attribution and cost evidence.
Ask for the full included cost per cubic metre, never the material line alone.
21:00 – 25:00
Make the closing decision explicit, and your wrap
- Full-scope adoption is not approved. The parties choose further investigation, a newly defined narrower trial or closure, stating the question the six comparisons failed to resolve.
- The assistant may stay in a bounded reporting role; its permission does not expand because the programme's mean improved.
- Before you close: write the application, the agreed trial and how any AI component is judged for your Session 1 decision. That is Section 9 of your 90-Day AI Leadership Plan.
An inconclusive or unattractive trial is an outcome you should be able to report without distortion.
Key Terms
Blocking
Grouping trial comparisons within declared sets of comparable conditions to account for an important nuisance factor.
Acceptance criterion
The agreed test a trial must meet for commercial adoption, separate from its technical safeguards.
Benefit attribution
How an intervention contributed to an outcome, including other changes, such as maintenance, that may explain it.
Source-bounded assistance
Assistance limited to approved documents that makes missing or conflicting evidence visible.
Self-Check — Before You Submit
1. The candidate averages 19 mg/L against a criterion requiring every result at or below 20. Sales want to report a pass. What should the managing director decide?
- A. Pass, the average is below 20
- B. Pass for blocks A, B, D and F as the intended scope
- C. Full-scope adoption not approved; choose investigation, a newly defined trial or closure
- D. Extend until the average improves
C. Full-scope adoption not approved; choose investigation, a newly defined trial or closure — Results of 29 and 21 fail a rule covering all six. Re-labelling the favourable blocks as the intended scope after the fact misrepresents the trial.
2. What are the existing and candidate costs per cubic metre in the assumed month?
- A. ₦800 and ₦740
- B. ₦840 and ₦850
- C. ₦840 and ₦740
- D. ₦850 and ₦840
B. ₦840 and ₦850 — ₦4.2 million and ₦4.25 million over 5,000 m³. The ₦800 and ₦740 figures use material alone, the comparison that hides the extra analysis and service cost.
3. The assistant yields ₦450,000 of capacity value and costs ₦600,000. What follows?
- A. It pays, because the chemistry improved
- B. It pays, because time saved is cash
- C. Its permission should expand
- D. It does not pay on preparation time alone
D. It does not pay on preparation time alone — ₦450,000 is below ₦600,000, and capacity becomes cash only if spending changes. Borrowing the chemical result to justify the assistant is the trap the book names.
Today’s Assignment
Write Section 9 of your plan: the application defined, the trial agreed and the AI component judged on its own contribution.
- Define the customer application behind your Session 1 decision in your company: objective, measure, conditions and missing information. Without a service team, use the counterpart you depend on.
- Set out the trial: purpose, baseline, authorisations, stop conditions, acceptance criterion and who pays.
- State how your company protects customer information and what a customer-facing assistant may use.
- Say how service effectiveness and any AI component are judged, each on its own evidence.
- The honest part: where your company has quoted an average or a material-only saving as proof. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 9 — The application defined, the trial agreed and the AI component judged on its own contribution.
How this is marked — the rubric
| Defines the application and its measure before naming a product | 25 |
| Specifies baseline, authorisations, stop conditions and acceptance criterion | 30 |
| Sets controls on customer information and assistant sources | 20 |
| Judges service and AI value separately, with an honest example | 25 |
| 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
Application definitionTrial designPaired comparison readingCustomer data protectionSeparate attribution
Part Three · Quality and Customer ApplicationSESSION 10
Chemical Treatment and Controlled Recommendations
Why This Matters
Chemical treatment joins product selection, delivery equipment, operating conditions and measurement. A recommendation can affect a customer's process beyond the product itself, so management needs a defined technical service boundary and a controlled path from analysis to action. The book supplies no dosing recipe or treatment specification, and neither does this session; those belong to a qualified, application-specific programme with the relevant approvals. You will lead the path: name the objective, make units and basis explicit, separate selection, adjustment and execution, control change, evaluate outcomes with the customer and define the evidence needed before permission expands. This session closes Week 2.
25-Minute Module
0:00 – 5:00
Name the objective and the authority
- Define the problem before discussing a product or rate: an agreed water-quality property, a corrosion-management programme or another established need, with how it is measured and which conditions affect it.
- THE DOCTRINE
Selling more product is not evidence of better treatment, and reducing consumption is not evidence that the customer objective is met. Both quantity and outcome belong in the review.
- A supplier's advisory model should not acquire control of the customer's dosing equipment by default. The commercial relationship does not establish permission to make an operating change.
- If your business gives no treatment programmes, apply this to the recommendation you give or receive that changes how someone else operates.
Ask whose objective the recommendation serves before you ask how good it is.
5:00 – 11:00
Units, basis and the product actually delivered
- Basis, plainly: what a number is a quantity of. Supplied formulation, active component, mass or volume must be stated, and a concentration needs its basis and denominator. A percentage by mass should not silently become one by volume.
- An assistant can check dimensional consistency and flag gaps; it must not infer density, active fraction or compatibility from a similarly named product. In the book's illustration, litres in one report and kilograms in another make a combined trend unreliable; the first action is a data clarification, not a programme adjustment.
Record Describes Does not establish Commanded pump setting Intended delivery Product actually delivered Opening + receipts − closing Apparent consumption before documented transfers, returns and losses Delivery into the process Accepted delivery record Delivery as the programme defines it Effectiveness or suitability - Converting volume to mass needs the actual product's density on a compatible basis. There is no treatment-rate recommendation in these accounting relationships.
A calculation can be arithmetically correct and still technically inappropriate for the application.
11:00 – 16:00
Separate selection, adjustment and execution; control change
Action May require Held by Select a candidate product Laboratory and field evaluation Competent specialists Recommend an adjustment Review of current operating measurements A competent person Execute Customer authorisation and equipment readiness The operator, under the customer's procedure - The application may retrieve the approved programme, identify a change in conditions or draft a review request. A change of product, concentration, location, equipment or condition needs a record of reason, evaluation, approval, revised documents and follow-up evidence.
- THE DOCTRINE
Do not allow a conversational assistant to update the treatment programme simply because a user described a problem urgently.
- The United Nations GHS provides a framework for hazard classification and communication, including safety data sheets. GHS Revision 11 was published in 2025; that is not, by itself, proof of a Nigerian implementation date. A safety data sheet informs hazard communication; it is not application validation or permission to combine materials.
Urgency is a reason to convene the competent person faster, not to bypass them.
16:00 – 20:00
Evaluate outcomes with the customer
- When a result deteriorates, the product may be unsuitable, execution may have differed, conditions may have changed or the measurement may need investigation. Track the throughput denominator and lag: a sample can describe material that entered before the latest intervention.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional water-treatment supplier's model links high consumption to poor results and suggests reducing it. The intervention history shows operators raised treatment because conditions worsened; cutting it could reverse the intended response.
IN THE MARGIN · THE PARALLEL- Margin. Priced per delivered product, the supplier's revenue rises with consumption; a contract priced around a verified outcome may create different incentives. Visits, analysis and equipment support are programme costs. Same loop, read from two sides.
- AI can highlight a pattern. Require it to keep observation and explanation distinct.
Before any model reads consumption, record why each change in consumption was made.
20:00 – 25:00
First application, expansion evidence, and close Week 2
- An appropriate first application can be a controlled review assistant that retrieves the approved programme, finds missing measurements and assembles evidence, with no execution. Test it on incomplete records and deliberate unit mismatches; it must abstain rather than invent a property.
- THE DOCTRINE
A reliable summary does not establish readiness for automated adjustment. The approval record should show which assessment earned each action permission.
- Feedback control needs a further assessment of dynamics, equipment behaviour, constraints, measurement failure, cybersecurity and fallback. Before a service proposal goes out, explain how the technical commitment and pricing fit.
- Before you close: write the objective, units and basis, separated authority and expansion evidence for your Session 1 decision. That is Section 10 of your 90-Day AI Leadership Plan, and it completes Week 2.
- Practice Exam 2 opens now: fifteen questions, closed book, on Sessions 1 to 10, closing Week 2. Sit it before Session 11, then write your assignment.
Validation of a reporting assistant provides no evidence about control. Each permission is earned on its own.
Key Terms
Active-content basis
A quantity comparison on the stated active component and denominator; not proof of technical equivalence.
Apparent consumption
Opening quantity plus receipts less closing, before documented movements. Not proof of delivery into the process.
Abstention
An application's declared refusal to recommend when required evidence is absent; the workflow still needs a response.
Change record
For a programme change: reason, evaluation, approval, revised documents and follow-up evidence.
Self-Check — Before You Submit
1. A customer's operator urgently asks a supplier's assistant to raise the treatment. What may the assistant do?
- A. Raise the dosing setting
- B. Update the programme and tell the customer later
- C. Draft a review request with the approved programme and current measurements for a competent person
- D. Ignore the request
C. Draft a review request with the approved programme and current measurements for a competent person — It may retrieve, identify and draft; a competent person evaluates and the operator executes under the customer's procedure. Ignoring the request wastes the useful part of its role.
2. One report gives consumption in litres, another delivered mass in kilograms. What comes first?
- A. A data clarification of product, density basis, temperature relevance and method
- B. Convert with a similar product's density
- C. Average the series
- D. Adjust the programme
A. A data clarification of product, density basis, temperature relevance and method — A combined trend is unreliable until the basis is established. Borrowing a similar product's density is the trap: density must not be inferred from a name.
3. A reliable reporting assistant is proposed for feedback control of dosing. What is required first?
- A. Nothing, it is validated
- B. A satisfaction survey
- C. More training data
- D. Assessment of dynamics, equipment, constraints, measurement failure, cybersecurity and fallback
D. Assessment of dynamics, equipment, constraints, measurement failure, cybersecurity and fallback — Validating a reporting assistant says nothing about control functions. The approval record must show which assessment earned each permission.
Today’s Assignment
Write Section 10 of your plan: the objective, units and basis, separated authority and the evidence for expansion.
- Name the objective behind your Session 1 decision in your company, how it is measured and whose objective it is.
- Make units and basis explicit, and state which records show intended, withdrawn and delivered product.
- Separate selection, adjustment and execution with owners by role, and describe change control.
- Define the evidence an assistant needs before proposing an adjustment.
- The honest part: where your company read volume sold as performance or let urgency bypass change control. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 10 — The objective, units and basis, separated authority and the evidence for expansion.
How this is marked — the rubric
| Names the objective and its measure, distinct from sales volume | 20 |
| Makes units and basis explicit and separates delivery records | 25 |
| Separates selection, adjustment and execution with owners and change control | 30 |
| Defines expansion evidence, with an honest current weakness | 25 |
| 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 — Production, Quality and the Customer 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
Treatment objectivesUnits and basisSeparated authorityChange controlExpansion evidence