30-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s oil and gas sector
WEEK 2 OF 6 · THE FOUNDATIONS YOU CANNOT SKIP · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Part Two · FoundationsSESSION 6
The AI-Ready Oilfield: Garbage In, Confident Garbage Out
Why This Matters
Everything in Week 1 quietly assumed that the numbers coming out of your field can be trusted. This week tests that assumption. Chapter 3 is the Data-Foundation Precondition in practice, and its warning is sharper than the old one: garbage in, confident garbage out. A model fed poor data does not hesitate; it hands your executive committee a clean, fluent answer that is wrong. Today you find where the data behind your first decision leaks, and you score it honestly. Nothing you build later in this programme is stronger than this.
25-Minute Module
0:00 – 5:00
Readiness is an operating property
- The AI-ready oilfield is not the most instrumented field. It is the field whose data is captured with discipline, validated before use, traceable to its source, consistently defined and connected to a decision owner.
- Readiness lives in calibration routines, shift-log discipline, naming standards and management behaviour, not in the data centre. It is an operating property, not an IT property, so it sits on your desk.
- THE DOCTRINE
A model trained on unvalidated data does not become wrong quietly; it becomes wrong confidently.
- The national targets you met in Session 2 are, in the book's words, execution tests, and Chapter 3 calls meeting them a data-readiness question before a technology question. Its test for any AI effort: does it change field decisions, reduce delay, protect production and strengthen accountability? If not, it is a decorative digital project.
When someone tells you the field is well instrumented, ask how much of it is trusted.
5:00 – 11:00
The leak, not the shortage
- The core economic problem is not a lack of data; most companies, the book says, have more than they can use. The data is scattered, late, unvalidated, or disconnected from the decision, and that gap becomes cost, downtime, deferment, compliance exposure and lower confidence in the asset plan.
- Symptoms: uncalibrated sensors, inconsistent logs, missing metadata, spreadsheet dependence and poor master data, managed through meetings, email and personal experience. Weakness accumulates: a sensor nobody trusts, a procurement delay treated as administration rather than a production threat.
- The book's signal-to-trust funnel, an illustrative figure, shows operating intelligence leaking away at every stage between what the field produces and what changes a decision. Think of a gathering line weeping at every flange: leadership acts only on what arrives.
- IN THE FIELD · THE WORKED CASE
- Field. At a fictional onshore flow station, three shifts mention the same valve in passing in their logs. Each note is filed and never read again; the instruments show nothing unusual.
IN THE BOARDROOM · THE PARALLEL- Boardroom. The company's monthly deck, built from instrumented data alone, shows the area green. One loop seen from both ends: it breaks between sense and decide.
- THE DOCTRINE
Measure AI by how effectively it reduces the distance between signal, decision and action.
Ask your team where the last avoidable loss was visible before anyone acted on it.
11:00 – 19:00
Six disciplines that turn data into evidence
- The chapter names six methods. None is glamorous; together they make every downstream model trustworthy. You need to know what to ask of them.
Discipline What it does, in plain terms Your question Data quality scoring A running score for each critical data set: how complete, current, plausible and consistent Does every output we see carry the quality score of its inputs? Sensor validation Cross-checks each instrument against physics, redundant measurements and its own history Which of our readings should not be trusted today? Automated field-report classification Reads shift logs and rounds, routes urgent observations, spots recurring themes; tuned to local terminology Who reads our field notes after they are written? Data lineage mapping Traces every reported figure back to its raw source Where did this number come from? Master data management One authoritative registry of wells, equipment, tags and definitions across systems Do our systems agree on what exists? Exception detection Pushes each missing, stale or implausible reading to a named owner with a deadline Who owns the report that did not arrive? - Carry one line into your next meeting: anomaly detection on an uncalibrated sensor does not detect anomalies; it manufactures them.
- THE DOCTRINE
A model that predicts a failure without generating a maintenance action is incomplete. A dashboard that identifies underperformance without assigning ownership is weak.
- Then the decision rights: who receives, validates, owns and escalates each alert, and how leadership will know it worked. These are operating discipline questions, not technology questions.
Exception detection lets you manage data the way you already manage safety: continuously, by exception, with accountability.
19:00 – 23:00
Build the foundation around the decision
- An AI-ready field needs not perfect data on day one but a disciplined path: ownership, standard definitions, validation rules, quality thresholds and practical governance.
- A deferred barrel, a downtime event, a critical spare or a production constraint must mean the same thing in operations, maintenance, finance, commercial and executive reporting. Otherwise, the book warns, AI will amplify confusion.
- The key principle is operational traceability: where the data came from, who owns it, when it was captured, whether it was validated and how it links to a decision. It also serves compliance, as NUPRC moves upstream operators toward measurement-based reporting of methane and greenhouse gas emissions.
Dimension (the book's Figure 4) The question to ask of each critical data set Accuracy Is it right? Calibration decay, drift and readings that look plausible while wrong are the threat. Timeliness How current is it at the moment the decision is made? Completeness Are the required readings, logs and reports actually captured? Consistency Does it agree with related data, and does it mean the same thing in every function? Traceability Can we say where it came from, when it was captured and whether it was validated? Ownership Who owns it, and which decision does it serve? - These six names are the book's own, from Figure 4 in Chapter 3, which it labels an illustrative profile: a typical brownfield baseline against an AI-ready threshold. Each operator must score its own data sets, and the shape of the gap shows where the readiness programme should start. Score each red, amber or green with a reason.
Start with the decision that matters most, not the data set that annoys you most.
23:00 – 25:00
Your wrap
- THE DOCTRINE
Data readiness is a field performance programme, not a back-office data-cleansing exercise. The objective is not a perfect data lake; it is making the most critical decisions more reliable, visible and measurable.
- Before you close: write down the first decision your data foundation must serve, following from your Session 1 loss, and the data sets it depends on. That starts Section 6 of your 90-Day AI Survival Game Plan: data readiness for your first decision.
- Then write your assignment.
An honest amber is worth more than a confident green.
Key Terms
AI-ready oilfield
A field whose data is captured, validated, traceable, consistently defined and connected to a decision owner; an operating property, not a count of instruments.
Confident garbage out
A model fed unvalidated data returns a fluent, certain-sounding answer that is wrong.
Data lineage
The traceable path of a reported figure back through every transformation to its raw source.
Master data management
One authoritative registry of wells, equipment, tags and definitions, so that every system agrees on what exists.
Self-Check — Before You Submit
1. A vendor offers a fictional operator an anomaly-detection model for its compressors. Nobody has checked whether the pressure transmitters feeding it are calibrated. What should the executive require first?
- A. Run the model and let the alerts reveal which transmitters are faulty
- B. Sensor validation on the transmitters the model will use
- C. A larger historian to store more of the readings
- D. A dashboard that shows the alerts to the executive committee
B. Sensor validation on the transmitters the model will use — Anomaly detection on an uncalibrated sensor does not detect anomalies, it manufactures them. Running the model first is tempting because it looks like progress, but it builds confident answers on unvalidated readings.
2. Which statement matches the book's definition of the AI-ready oilfield?
- A. The field with the most sensors per well
- B. The field whose data has all been moved into one enterprise data lake
- C. The field whose data is captured, validated, traceable and acted upon
- D. The field that has deployed the most models
C. The field whose data is captured, validated, traceable and acted upon — Readiness is an operating property, not an IT property. Heavy instrumentation can coexist with untrusted data disconnected from any decision owner, which is why the tempting first option is wrong.
3. According to Chapter 3, what is the core economic problem in field data?
- A. Data is scattered, late, unvalidated or disconnected from the decision
- B. Companies do not collect enough data
- C. Field staff refuse to report
- D. Storage is too expensive
A. Data is scattered, late, unvalidated or disconnected from the decision — Most companies have more data than they can use. The shortage answer is tempting because it points toward buying sensors, but the loss sits between what the field knows and what leadership can act upon.
Today’s Assignment
Write Section 6 of your plan: score the data behind your company's first decision and name its readiness gap.
- Name the first decision your data foundation must serve, tied to the one loss you chose in Session 1, and the role that owns that decision.
- List the three to five data sets that decision depends on, and score each against the six dimensions of the book's Figure 4 (accuracy, timeliness, completeness, consistency, traceability, ownership), with a short reason for each score, not just a colour.
- Name one definition your company does not use consistently across functions, and what that inconsistency costs.
- Name the readiness gap: the weakest link between signal and decision, and which discipline closes it first.
- The honest part: where does your company manage data quality through meetings, email and personal experience rather than owned exceptions? No AI in use yet is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 6 — Data readiness for our first decision.
How this is marked — the rubric
| Names one specific decision tied to the Session 1 loss, with an owner by role | 20 |
| Scores each critical data set against all six Figure 4 dimensions (accuracy, timeliness, completeness, consistency, traceability, ownership) with evidence-based reasons | 30 |
| Identifies one inconsistent definition and states its consequence in operating or commercial terms | 20 |
| Names the weakest link and the discipline that closes it first, and admits one current weakness honestly | 30 |
| 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
Data-Foundation PreconditionData quality scoringSensor validationData lineageDecision rights
Part Two · FoundationsSESSION 7
From Field Signal to Executive Decision
Why This Matters
Yesterday you scored the data behind your first decision. Today you follow one signal from the point where physical reality becomes recorded reality to the executive who must act. Chapter 3 calls that route a chain of trust: if any link is weak, your company may see a dashboard and still fail to act in time. You will meet the eight data classes, the five tiers, the board's scoreboard and the data-readiness pilot, then trace one real signal in your own company. Be specific. A vague path cannot be piloted.
25-Minute Module
0:00 – 6:00
Where readiness is built
- Chapter 3 names where physical reality becomes recorded reality:
Data class How it goes wrong What readiness looks like SCADA points Thousands of tags of uneven trust A tag audit, the most common first step Flow meters Drift between provings: revenue loss that never appears as an incident Continuous revenue assurance Pressure transmitters Plausible while wrong, the most dangerous failure mode Validation that tells models which inputs to distrust Tank gauging Phantom gains and losses Levels reconciled against recorded movements Field reports Prose that disappears into files Digital capture that makes reporting easier, not heavier Maintenance records Failure coded as general fault Every closed work order teaches the system something HSE observations Filed as paperwork Trended as leading indicators Production allocations Opaque logic, chronic dispute Followed from sensor to statement - On tanks, the book ties measurement credibility to the national loss-reduction story: crude losses are falling to multi-year lows, which the NUPRC report it cites describes as a 16-year low.
- Three patterns recur: early warning, prioritisation by decision impact, and executive visibility of how much of the operation is current and trusted, which is the ceiling on every other intelligence ambition.
The board does not need more dashboards. It needs better decision intelligence.
6:00 – 11:00
The chain of trust, in five tiers
- The chapter's architecture has five tiers: data capture; data quality and validation; contextual analytics using the six disciplines from Session 6; workflow automation that assigns actions and escalations; and executive visibility that translates technical events into production, cost, safety, compliance or risk consequences. Each layer earns the one above it.
- THE DOCTRINE
An AI alert without a responsible owner is operational noise. A forecast without an intervention path is a document. Build a system of action, not only a system of insight.
- Be pragmatic: instrumented data where available, structured manual reporting where required, field validation where necessary. Do not wait for a fully automated oilfield.
- IN THE FIELD · THE WORKED CASE
- Field. At a fictional swamp asset, an allocation meter drifts between provings. Validation flags it, and an exception goes to the metering supervisor with a deadline.
IN THE BOARDROOM · THE PARALLEL- Boardroom. The same company reconciles allocation with partners. Because the figure can be followed from sensor to statement, it argues from evidence. Same chain; the boardroom stands at the far end.
Ask who owns the action at the end of your favourite dashboard.
11:00 – 16:00
People, rhythm and trust
- Human-in-the-loop: AI observes, detects, ranks, recommends and tracks; humans validate, decide, approve and execute. Leadership sets the rules: which alerts the field team may close, and which must go to the asset manager or the executive committee.
- Adoption fails, the book says, when people see AI as surveillance without value. Introduce it as an operations-support capability.
- To the rhythm you met in Session 5, Chapter 3 adds data-health questions: which alerts need field validation today, which were false positives this week, which models need adjustment this month.
- Inaction now carries compliance and talent risk too: as NUPRC moves reporting toward measurement, regulators, lenders, partners and investors will expect credible, traceable, defensible data.
- Ask the chapter's trust questions: what data did the model use, how current is it, who can override it, and what happens when it is wrong? Under Local Content 2.0, it puts field-reporting automation and local data engineering in the intelligence layer Nigerian companies should own.
Answer the question of what happens when the model is wrong before anyone has to ask it.
16:00 – 21:00
The pilot and the scoreboard
- Chapter 3 sharpens the 90-day pilot from Session 5. The first 30 days end with a pilot charter: scope, baseline, success metrics, data sources, stakeholders, governance controls and a decision workflow. Many pilots fail in the last 30, by not forcing use inside existing routines.
- The strongest first pilots are narrow and consequential: a tag audit on one critical asset, sensor validation on the meters that drive allocation, digital capture of one field's operator rounds. It must end with a decision that visibly improved because the data beneath it improved, then continue, expand, adjust or stop. Lead with the value case: fix the data can absorb unlimited budget without touching a decision.
Data health: can the field be trusted? Decision velocity: is trust becoming speed? Data completeness Reporting cycle time Data timeliness Exception closure rate Sensor uptime Decision latency Measurement accuracy Field-to-management visibility - THE DOCTRINE
A high detection rate with a low closure rate means the organisation has bought visibility without accountability, the most common failure mode of data-quality initiatives.
- Scale only after adoption is proven, with five named owners per use case: executive sponsor, process owner, data owner, field owner and governance owner.
Every KPI needs a baseline; without one you can describe activity but you cannot claim improvement.
21:00 – 25:00
Your wrap
- The chapter closes nationally: Nigeria's 3 million barrel per day ambition also requires a new standard of operational intelligence, because value is already being lost inside existing operations.
- THE DOCTRINE
Refuse the next dashboard until the numbers beneath it can be defended.
- Before you close: write down one field signal in your company, every link it passes through to reach an executive decision, and the link you trust least. That is Section 7 of your 90-Day AI Survival Game Plan: your signal-to-decision path and the pilot outline that tests it.
- Then write your assignment.
Name the data owner on Monday. Most of this chapter waits for that name.
Key Terms
Chain of trust
The governed path a field signal follows from capture to executive reporting; one weak link and the business sees a dashboard but fails to act.
Decision latency
The time from a validated signal to field action; the metric the book says everything else exists to compress.
Exception closure rate
The share of flagged exceptions, such as missing data, suspect readings or reconciliation breaks, resolved on time by their named owner.
Field-to-management visibility
The share of critical assets whose current, validated status leadership can see today rather than in next month's deck.
Self-Check — Before You Submit
1. A fictional operator's data-quality system flags many exceptions weekly, but most stay open past their deadline. The vendor proposes more detection rules. What should the executive do?
- A. Approve the extra rules, since more detection means more visibility
- B. Pause the programme until the data is perfect
- C. Fix ownership and the response workflow first
- D. Move the exception report into the quarterly board pack
C. Fix ownership and the response workflow first — A high detection rate with a low closure rate is the book's most common failure mode of data-quality initiatives. More rules are tempting but add visibility without accountability.
2. Which is the correct order of the chapter's five architecture tiers, from the bottom?
- A. Executive visibility, analytics, capture, validation, workflow
- B. Data capture, quality and validation, contextual analytics, workflow automation, executive visibility
- C. Data capture, contextual analytics, executive visibility, validation, workflow automation
- D. Quality and validation, data capture, workflow automation, analytics, executive visibility
B. Data capture, quality and validation, contextual analytics, workflow automation, executive visibility — Each layer earns the one above it, so validation sits directly on capture. Putting analytics before validation is the tempting error, and it is how AI inherits every weakness beneath it.
3. Which is the strongest first data-readiness pilot by the chapter's standard?
- A. Sensor validation on the meters that drive allocation, with a charter, a baseline and a senior sponsor
- B. An enterprise data lake covering every data set the company holds
- C. A new executive dashboard showing all field data
- D. Company-wide data literacy training before any pilot begins
A. Sensor validation on the meters that drive allocation, with a charter, a baseline and a senior sponsor — The chapter wants narrow, consequential pilots that end with a visibly improved decision. The data lake feels thorough, but it validates every data set rather than the data the decision requires.
Today’s Assignment
Write Section 7 of your plan: trace one field signal in your company to an executive decision and outline the 90-day pilot that tests the path.
- Choose one signal from the eight data classes that feeds your Session 6 decision.
- Trace it through the five tiers, naming the owner by role at each.
- Name the weakest link honestly, and what your company does instead today, whether that is email, a meeting or one experienced person.
- Outline the 90-day pilot: the charter contents for the first 30 days, what is built in the second, how adoption is forced in the third, and the decision that must visibly improve.
- Choose one data-health and one decision-velocity KPI, each with a baseline or a plan to set one within 30 days.
- No AI in use yet is normal; the signal and its path exist anyway. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 7 — Our signal-to-decision path and the pilot outline that tests it.
How this is marked — the rubric
| Traces one named signal through all five tiers with an owner by role at each | 30 |
| Identifies the weakest link and what the company currently does instead | 20 |
| Outlines a 90-day pilot with charter contents and the specific decision that must improve | 30 |
| Selects one data-health and one decision-velocity KPI, each with a baseline or a dated plan to set one | 20 |
| 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
Chain of trustSignal-to-decision tracingPilot charteringData-health KPIsDecision velocity
Part Two · FoundationsSESSION 8
Building the Engine
Why This Matters
Every use case you have met so far runs on something the demo never shows. Chapter 4 calls it the engine: the data, the pipelines, the platform, the connectivity and the lifecycle machinery that keep a model honest. When AI programmes fail, the book says, it is rarely the model; the engine was never built. Today you learn enough of the engine to question an architect, a vendor and your own finance director: where intelligence must run on a Nigerian asset, what the data platform must contain, why models decay, and how foreign exchange and data sovereignty change the design.
25-Minute Module
0:00 – 5:00
The part that does not demo well
- Beneath every use case sits an engine: data, pipelines, platform, connectivity and the lifecycle machinery that keeps models honest.
- The engine is precisely the part that does not demo well, which is why leaders underinvest in it.
- THE DOCTRINE
You cannot run intelligence on infrastructure built to run paperwork.
- The engine is durable operating infrastructure, not a one-time IT project to be delegated and forgotten. Build it well and every use case becomes achievable; neglect it and they remain slides.
If the engine is not on your agenda, it is on nobody's.
5:00 – 11:00
Connectivity is not a given
- Connectivity cannot be assumed. Offshore platforms, swamp and creek operations and remote onshore assets sit at the low, intermittent, expensive end of the spectrum; only terminals, plants and corporate sites have reliable high-bandwidth links.
- A model that needs a live cloud connection is useless when the link drops, which may be exactly when an anomaly is developing. The answer is edge AI: running the model on or near the asset, buffering results and forwarding them when the link allows, which is called store-and-forward.
- THE DOCTRINE
Assume intermittent connectivity as the design case for offshore, swamp and remote assets, and make edge inference and store-and-forward buffering the default there.
- The book's first instruction: map your operating estate against the connectivity spectrum before designing anything, and let the map drive the architecture.
- IN THE FIELD · THE WORKED CASE
- Field. On a fictional offshore platform, a compressor-anomaly model depends on the cloud. The link drops overnight and the model goes blind. At the edge, it keeps watching.
IN THE BOARDROOM · THE PARALLEL- Boardroom. The same company's executive committee sees an all-cloud architecture. The right question is not which vendor, but where each workload must run when the link is down.
Ask your team to show you what happens when the link drops. Do not accept a slide.
11:00 – 16:00
Edge, core and cloud
- The right design, the book argues, is neither purely cloud nor purely on-premise. It is hybrid, and the art is workload placement.
Tier What runs there Why there Edge, on or near the asset Real-time inference, local anomaly detection, store-and-forward buffering Latency-sensitive work exposed to connectivity loss Core, regional or on-premise Aggregation, validation, historian and plant-system integration, sovereign data hosting Critical data stays under national or corporate control Cloud Model training and re-training, large-scale analytics, simulation, enterprise applications Elastic compute for heavy work that is not latency-critical - Four questions decide placement, not a vendor's preference: how latency-sensitive, how exposed to connectivity loss, how data-hungry, how sensitive? Get it wrong one way and the engine is fragile; the other way, expensive.
- The book's own placements: a real-time safety-related inference belongs at the edge; model training belongs in the cloud; heavy reservoir simulation belongs where the compute is; sovereign or sensitive data belongs at the core.
A single-vendor default is a placement decision someone else made for you.
16:00 – 21:00
The platform, MLOps and the Nigerian realities
- The data platform is built bottom-up: sources; ingestion, batch and streaming; a governed store, increasingly a lakehouse; the governance and quality layer of ownership, definitions, lineage, validation and access control; a serving layer; and consumption. Governance sits in the middle by design. Do not let a vendor sell you the top of the stack on a foundation that does not exist.
- THE DOCTRINE
Treat every deployed model as a living asset with an owner, a monitoring regime and a retraining plan, never as a finished deliverable.
- MLOps makes that real: build, validate, deploy, monitor and retrain, under governance. Monitoring watches for drift: conditions change and answers degrade unnoticed, like a meter factor drifting between provings.
- Two Nigerian realities. Foreign exchange: cloud is typically priced in foreign currency, and unmanaged egress, idle compute and over-provisioned services produce bills that escalate against a weakening naira. Data sovereignty: where production, reserves, commercial and emissions data lives, and who controls it, is a strategic choice; the book reads the National AI Strategy's framing of AI as a driver of national competitiveness as partly a sovereignty argument.
Ask who owns each model in production, and when it was last checked against reality.
21:00 – 25:00
Six requirements, and your wrap
- The chapter lists what AI actually requires of the engine: reliable data pipelines, the most underinvested and most decisive part; edge compute; a governed data store; model lifecycle infrastructure; security throughout the stack, with OT/IT segmentation where AI touches operational technology; and indigenous engineering capability, the scarcest ingredient.
- THE DOCTRINE
An engine the organisation cannot operate itself is a dependency, not a capability.
- Before you close: for the first use case behind your Session 1 loss, write down where it must run and why, which platform layer is weakest, and who would notice if its model drifted. That is Section 8 of your 90-Day AI Survival Game Plan: what your first use case requires of the engine.
- Pop Quiz 2 opens now: five questions on Sessions 6 to 8, data readiness and the engine. Sit it before Session 9, then write your assignment.
Carry one question out of this session: can my data reach a decision while that decision still matters?
Key Terms
Edge AI
Running a trained model on or near the asset so that it keeps working when connectivity drops.
Workload placement
Deciding where each piece of AI work runs, at the edge, the core or the cloud, by its latency-sensitivity, connectivity exposure, data volume and data sensitivity.
MLOps
The governed practice of managing models across their lifecycle: build, validate, deploy, monitor for drift and retrain.
Data sovereignty
Deliberate control over where critical data physically lives and who controls it, treated as a strategic rather than merely technical choice.
Self-Check — Before You Submit
1. A fictional swamp-asset operator wants a real-time, safety-related anomaly model. Links from the asset drop for hours at a time. Where should the model's inference run?
- A. In the cloud, where compute is elastic
- B. At the core, so the data stays sovereign
- C. At the edge, with store-and-forward buffering
- D. Wherever the vendor's standard design places it
C. At the edge, with store-and-forward buffering — Latency-critical, connectivity-exposed work belongs at the edge. The cloud answer is tempting on cost, but a model that needs the link fails exactly when the link drops.
2. Why does the book place governance in the middle of the data platform stack?
- A. Because it is the layer that turns raw sources into trustworthy, decision-ready data
- B. Because it should be added after the dashboards are live
- C. Because it is the cheapest layer to build
- D. Because consumption depends on it least
A. Because it is the layer that turns raw sources into trustworthy, decision-ready data — Governance is the spine, not an afterthought; skip it and everything above inherits the weakness below. Adding it after the dashboards is the tempting error the book warns against.
3. What does the book say happens to a model that is deployed and then left alone?
- A. It keeps its original accuracy until it is replaced
- B. It quietly decays and can become confidently wrong without anyone noticing
- C. It improves automatically as it sees more data
- D. It stops producing outputs, which alerts the team
B. It quietly decays and can become confidently wrong without anyone noticing — Conditions drift and performance degrades silently, which is why monitoring and retraining are part of the engine. Assuming it improves on its own is tempting, but nothing retrains a model unless the lifecycle machinery does.
Today’s Assignment
Write Section 8 of your plan: what your company's first use case requires of the engine.
- Name the first use case, tied to the signal you traced in Session 7, and the asset or site where it must operate.
- Place that site on the connectivity spectrum and say what happens to the use case today when the link drops.
- Decide edge, core or cloud for each workload in the use case, using the four placement questions, and justify each placement in one sentence.
- Name the weakest data platform layer in your company and what is being built on top of it regardless.
- Describe the MLOps you need: who owns the model, how drift would be noticed and what triggers retraining.
- State which costs would be FX-exposed and how spend would be monitored, and where the critical data would live and who controls it.
- The honest part: which of the six engine requirements your company has least of today. No AI in use yet is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 8 — What our first use case requires of the engine.
How this is marked — the rubric
| Places every workload at edge, core or cloud with a reason drawn from the four placement questions | 30 |
| Identifies the weakest platform layer, the model owner and the drift trigger | 25 |
| States the FX exposure and a data-sovereignty decision specific to the use case | 25 |
| Names the scarcest engine requirement honestly and its consequence for the use case | 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 — Data Readiness and the Engine 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
Connectivity mappingWorkload placementData platformMLOpsFX disciplineData sovereignty
Part Two · FoundationsSESSION 9
Build, Buy or Partner — and Funding the Engine
Why This Matters
Yesterday you worked out what your first use case needs from the engine. Today you decide how to get it and how to pay for it. The second half of Chapter 4 turns the engine into executive decisions: what to buy, build or partner for, what to watch, in what order to build, and how to make a cost case for something that shows little for months. Make the cost case honestly, or your engine will be starved and your use cases will stay one-off pilots.
25-Minute Module
0:00 – 5:00
Where the engine pays
- The engine pays directly in six places: resilient field intelligence through connectivity loss; a reusable data foundation; lossless data capture, including MRV-grade emissions data; cost-disciplined compute; a trustworthy model estate; and sovereign, resilient operation.
- Invest in the engine and every later capability is faster, cheaper and more reliable; skip it and every use case is a fragile, expensive one-off.
- Like the data path in Session 7, the engine is a chain of trust: a failed link surfaces far downstream as a wrong decision. It must be engineered, not assembled.
The board need not understand every layer, only that the stack is where the ambition lives or dies.
5:00 – 11:00
Build, buy or partner
- THE DOCTRINE
Buy the commodity, build the differentiator, partner for the gap, and structure every partnership to transfer capability to local engineers rather than entrench dependence.
Component Decision The trap Commodity: cloud compute, standard storage, established tooling Buy Building it wastes scarce engineering talent Differentiator: the data foundation tuned to your assets, edge intelligence for your connectivity, the operating model Build and own A black-box platform that surrenders the most strategic part Gap: scarce engineering skills Partner Permanent dependence instead of local capability - Decide component by component, with sovereignty, FX and capability-building as explicit criteria, never by defaulting to one vendor's preferred model.
- The book's reference build: an edge gateway at each exposed asset; a regional core under national and corporate control; the cloud for heavy work with FX discipline; bound by MLOps, observability and security, and run by the company's own engineers.
- IN THE FIELD · THE WORKED CASE
- Field. A fictional marginal field operator buys commodity cloud compute and standard storage for a remote flow station, and builds and owns the edge intelligence designed for its own connectivity, where its advantage lives.
IN THE BOARDROOM · THE PARALLEL- Boardroom. Its board reviews the partner contract for the engineering gap and asks whether, at the end of the term, its own engineers can run the engine alone.
Ask every partner what your people will be able to do at the end of the contract that they cannot do today.
11:00 – 16:00
Edge, links and observability in practice
- Edge AI in practice: train centrally, infer locally. The edge also validates and compresses data so scarce bandwidth carries clean, prioritised data, and must be manageable from the core without a site visit.
- Engineering for uneven links means store-and-forward, data prioritisation so the critical alert travels first, and local autonomy. Satellite, cellular and private LTE, microwave and mesh are improving, but none is universal or free.
- THE DOCTRINE
An architecture that only works on a perfect link is not built for the Nigerian field; it is built for a brochure. Test it by deliberately cutting the link.
- Observability instruments the engine end to end, from edge devices and pipelines to models and costs. Without it, an edge device that has quietly stopped reporting stays invisible until it becomes a failed decision.
You cannot fix what you cannot see, and in a distributed engine you will not see it by accident.
16:00 – 21:00
Sequence, rhythm and the ways it fails
- Build in sequence: fragmented, centralisation, hybrid maturity, sovereign-intelligent, each stage depending on the one before. Build only the slice the first use case needs: the 90-day sprint diagnoses, then builds, then runs the use case on it, including a deliberate connectivity drop. An investment that does not enable a measured business outcome is a cost, not an engine.
Failure mode The discipline that prevents it Cloud-or-nothing thinking Design for the connectivity you have Skipping governance Govern from the foundation Shipping models once Run MLOps Unmanaged cloud spend Manage cost against FX Treating the engine as a finished IT project Operate it continuously An engine the company cannot operate itself Build it sovereignly and locally The boil-the-ocean platform Grow it around real use cases - Review health daily and weekly, maturity and cost against FX monthly. The board tracks eight readiness dimensions against a baseline: connectivity coverage, platform maturity, edge capability, MLOps maturity, sovereignty and hosting control, stack cybersecurity, cost and FX efficiency, and indigenous engineering capability.
- Data, platform, MLOps and OT security engineers are scarce. An organisation that can build models but not the engine to run them has a hobby, not a capability; the book ties this to NCDMB's capacity-development mandate.
Review the engine's health with the seriousness you give any critical asset, because it is one.
21:00 – 25:00
Funding the engine, and your wrap
- The engine shows little for months and competes with use cases that demo well. The cost case: it is not an expense to minimise but the investment that sets the return on every later capability. Fund use cases on a weak engine and you pay repeatedly; fund the engine first and you pay once.
- Two Nigerian dimensions: the shift from capital to operating expenditure demands FX discipline, and the case must fund capability alongside infrastructure.
- THE DOCTRINE
The cost case for the engine is, in the end, the cost case for the entire AI agenda.
- Before you close: note one component you would buy, one you would build, one you would partner for, and your first observability gap. That is Section 9 of your 90-Day AI Survival Game Plan. Then write your assignment.
Fund the unglamorous engine today; it is the one whose models actually run tomorrow.
Key Terms
Buy, build, partner
Buy commodity infrastructure, build and own the differentiators, and partner for capability gaps in ways that transfer skill rather than entrench dependence.
Store-and-forward
Capturing and buffering data locally through a connectivity outage and forwarding it when the link returns, so no field data is lost.
Observability
End-to-end instrumentation of edge devices, pipelines, platform, models and costs, so failures are seen before they surface as wrong decisions.
Infrastructure maturity
The book's sequence for building the engine, from fragmented through centralisation and hybrid maturity to sovereign-intelligent, each stage depending on the one before.
Self-Check — Before You Submit
1. A vendor offers a fictional operator a turnkey platform holding its data foundation and edge models, run entirely by the vendor's staff. What does the book's principle say?
- A. Accept it, because buying is always faster than building
- B. Buy the commodity parts, but build and own the differentiating foundation
- C. Reject all vendors and build every layer in-house
- D. Accept it, provided the price is fixed in naira
B. Buy the commodity parts, but build and own the differentiating foundation — The differentiating foundation should be built and owned; buying it as a black box surrenders the most strategic part. Building every layer is the opposite trap, wasting scarce talent on commodity infrastructure.
2. Which test does the book insist the 90-day engine pilot includes?
- A. A deliberate loss of connectivity, to confirm the field stays intelligent
- B. A comparison of three cloud vendors' prices
- C. A full migration of every data set into the new platform
- D. A board demonstration of the dashboard
A. A deliberate loss of connectivity, to confirm the field stays intelligent — The pilot must prove the engine survives Nigerian realities, not demo conditions. A board demonstration is tempting because it is visible, but an architecture that only works on a perfect link is built for a brochure.
3. Why does the book say the cost case must fund indigenous engineering capability alongside infrastructure?
- A. Because it removes the need for any partner
- B. Because it allows the cloud tier to be skipped
- C. Because engineers cost less than hardware
- D. Because the engine is only as valuable as the engineers who build and run it
D. Because the engine is only as valuable as the engineers who build and run it — Capability is both a cost and a strategic investment that reduces dependence. Removing the need for any partner is the tempting misreading: the principle is to partner for the gap in a way that transfers skill.
Today’s Assignment
Write Section 9 of your plan: the build, buy or partner decision for your company's engine, the observability it needs and the cost case you will put to your board.
- Classify each component your first use case needs as commodity, differentiator or gap, and say buy, build or partner, with sovereignty, FX and capability-building as criteria.
- For any partnership, state how it will transfer capability to your own engineers, and what would happen to the use case if the partner left.
- Name what you must be able to observe across edge devices, pipelines, platform, models and costs, and the one blind spot your company has today.
- Place your company on the infrastructure maturity sequence and name which of the seven failure modes it is closest to.
- Write the cost case as one paragraph your finance director would accept: cost of the engine slice, FX-exposed costs, capability funded alongside it, and the measured business outcome it enables. Label estimates as estimates.
- The honest part: where has your company funded use cases on a weak engine? No AI in use yet is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 9 — Build, buy or partner, observability and the funding case for our engine.
How this is marked — the rubric
| Classifies each component as commodity, differentiator or gap and applies buy, build or partner consistently | 30 |
| Specifies how any partnership transfers capability and names the dependence risk | 20 |
| Names a concrete observability blind spot and the nearest failure mode honestly | 20 |
| Makes a cost case linking the engine slice to a measured business outcome, with FX exposure and capability funding stated | 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
Build, buy or partnerCapability transferObservabilityInfrastructure sequencingThe engine cost case
Part Two · FoundationsSESSION 10
The Subsurface Advantage
Why This Matters
So far this programme has lived at or above the wellhead. Chapter 5 goes below it, to the seismic, the reservoir and the rock, where Nigeria's reserves and production growth are ultimately decided. It is also where the Value-versus-Risk Map bites hardest: the value is high, and so is the cost of being wrong. The chapter's rule is short. Accelerate the expert, never replace the judgement. If your company does no subsurface work, stay with me. Your clients' subsurface decisions drive demand for your work, and your company has its own highest-consequence technical judgement. The same rule governs both.
25-Minute Module
0:00 – 5:00
Below the wellhead
- The subsurface holds the largest technical uncertainties; small improvements in understanding become large changes in value. The book's subsurface intelligence stack runs seismic, interpretation, static model, dynamic simulation, development decision, and AI compresses time and uncertainty at every layer.
- It is an operating-intelligence problem, not a separate specialty, losing value to familiar frictions: scattered data, models drifting from production reality, unquantified uncertainty, slow workflows.
- THE DOCTRINE
A reservoir insight that takes a year to produce, cannot be reconciled with production reality, or that nobody acts on, is as much a loss as a flared molecule or a deferred barrel.
- Be honest about the limit from the start: AI does not see the rock. It works on the data we have about the rock, which is sparse, indirect and costly.
If you do not operate fields, ask which of your clients' subsurface decisions fills your order book.
5:00 – 11:00
The economics and the methods
- The core economic problem: data is scarce relative to the complexity it must describe, expensive and disconnected. A reservoir is seen through a handful of wells, an indirect seismic image and production data, and decisions worth hundreds of millions rest on that. Judge AI by whether it compresses the distance between data, understanding and decision, not by how elaborate the model is.
Method In plain terms Deep-learning seismic interpretation Maps horizons, faults and facies across the whole volume Physics-informed machine learning Builds the physics of flow and rock into learning, so it needs less data Reservoir property modelling Several plausible pictures between wells, not one guess Simulation acceleration and proxy models Fast approximations, so hundreds of options can be explored Automated history matching An ensemble of models that reproduce actual production Uncertainty quantification Ranges with explicit confidence, not single numbers - Physics-informed learning matters because the subsurface rarely offers the abundant, clean data pure machine learning wants. For data-scarce Nigerian reservoirs it is often the difference between a model engineers trust and a black box they ignore.
Ask your technical team which of these six they already use, and which decision each one changed.
11:00 – 17:00
Seeing the whole volume, managing the range
- Seismic is the most expensive image in the business. Deep learning interprets the whole volume in hours rather than weeks, surfacing features a sampled workflow misses; in a mature basin like the Niger Delta, much of the remaining prize is subtle. Re-examine existing seismic over mature areas first. The model proposes; the expert validates.
- History matching tunes a reservoir model until it reproduces actual production. It is non-unique: many models match the same history yet forecast very different futures, so one matched model understates the uncertainty.
- THE DOCTRINE
Make capital and reserves decisions against the range rather than a point estimate. The goal is not a confident number; it is a well-characterised one.
- IN THE FIELD · THE WORKED CASE
- Field. A fictional mature-field operator holds one matched model and one forecast for an infill well. AI-assisted matching finds several models that fit the history equally well and disagree about the future.
IN THE BOARDROOM · THE PARALLEL- Boardroom. Its capital committee now sees a range with probabilities, decides against it, and asks what data would narrow it. The book's figure here is conceptual, not field data.
When a forecast arrives as a single line, ask how many other models matched the same history.
17:00 – 22:00
The recovery prize and the Value-versus-Risk Map
- The recovery factor is the share of the in-place resource actually produced. On Nigeria's mature base the gap is barrels already found, often reachable through existing infrastructure, in bypassed and uncertain zones: among the cheapest barrels available. The book's figure of this split is illustrative, not field data.
- The book names eight paying uses, with mature-field recovery the highest-return for Nigeria's producing base; the others are prospect evaluation, well placement, reservoir surveillance, fault and compartment analysis, seismic reprocessing, reserves estimation and enhanced-recovery decisions.
- On the Value-versus-Risk Map from Session 1, these uses sit where value is high and the damage from a wrong answer is also high. That asymmetry of consequence is why the control must be asymmetric too.
- THE DOCTRINE
Accelerate the expert, never replace the judgement. No confident interpretation acts on a well without an expert's judgement behind it.
- AI amplifies the geoscientist and reservoir engineer; it does not replace them. For EPC, services, marine, fabrication and downstream companies, the same rule governs your highest-consequence technical judgement.
The barrels between produced and unrecoverable are the cheapest you will find, if the analysis reaches a decision and an owner.
22:00 – 25:00
Your wrap
- Before you close: name the one decision where expert judgement must be accelerated and never replaced, and its owner. Operators will often choose a recovery or development decision; others, a client subsurface decision or their own highest-consequence technical call.
- That is Section 10 of your 90-Day AI Survival Game Plan, and it closes Week 2, the foundations you cannot skip.
- Practice Exam 2 opens now: fifteen questions across Sessions 1 to 10. Sit it before Week 3, then write your assignment.
Demand the loop. Fund the data. Keep the human in the chair.
Key Terms
Recovery factor
The share of the in-place resource actually produced; on mature fields the gap holds some of the cheapest barrels available.
History matching
Tuning a reservoir model to reproduce actual production; non-unique, because many models can match the same history.
Physics-informed machine learning
Machine learning that embeds the physics of flow, material balance and rock, so it learns from sparse data and extrapolates more safely.
Uncertainty quantification
Stating forecasts and volumes as ranges with explicit confidence rather than single numbers, so capital and reserves decisions are made against the range.
Self-Check — Before You Submit
1. A fictional operator's AI seismic tool flags a possible bypassed compartment. The development manager wants to commit a well on the model's output alone. What should the executive do?
- A. Approve the well, because the model has seen the whole volume
- B. Retire the tool, because AI cannot see the rock
- C. Ask the vendor to certify the interpretation
- D. Have the geoscientist validate it, then connect it to a development decision with an owner
D. Have the geoscientist validate it, then connect it to a development decision with an owner — Accelerate the expert, never replace the judgement: the model proposes, the expert validates. Retiring the tool is the opposite error; the book treats AI as a major subsurface lever with the expert in the loop.
2. What is the main value of AI-assisted history matching, according to the book?
- A. A single, more confident forecast
- B. An ensemble that honestly represents the range of futures consistent with the data
- C. The elimination of subsurface uncertainty
- D. The removal of any need for production data
B. An ensemble that honestly represents the range of futures consistent with the data — The result is not a more confident forecast but a more honest one. Eliminating uncertainty is tempting but impossible: many models can fit the same data, and AI can only characterise that ambiguity better.
3. Why does the book call barrels from a higher recovery factor on mature fields among the cheapest available?
- A. They are already discovered and often reachable through existing infrastructure
- B. They need no data work before analysis
- C. They can be decided without expert review
- D. They can be booked as reserves without modelling
A. They are already discovered and often reachable through existing infrastructure — These barrels are already found and often accessible through existing infrastructure. The no-data-work option is the tempting error: the book still requires trusted data and an expert's judgement behind every decision.
Today’s Assignment
Write Section 10 of your plan: where expert judgement must be accelerated and never replaced in your company.
- Name the highest-consequence technical decision in your business: for operators, a recovery, development, well-placement or reserves decision; for other company types, the client subsurface decision that drives your work or your own highest-consequence judgement.
- Place it on the Value-versus-Risk Map: what value faster, fuller analysis would create, and what it costs your company when the judgement is wrong.
- Say which methods would accelerate the expert, and what the expert must still decide.
- State how the output would be expressed as a range rather than a single number, and who would decide against that range.
- Name the expert who owns the judgement and the decision owner who acts on it, both by role, and the point at which the machine stops and the expert decides.
- The honest part: where does your company currently treat a single confident number as if it were certain? No AI in use yet is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 10 — Where expert judgement must be accelerated and never replaced in our business.
How this is marked — the rubric
| Names one specific high-consequence decision appropriate to the company type | 20 |
| Places it on the Value-versus-Risk Map with both the value and the cost of being wrong | 25 |
| Separates what the machine accelerates from what the expert must decide, with the handover point stated | 30 |
| Expresses the output as a range and names the expert and the decision owner by role | 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 — Foundations 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
Subsurface intelligenceValue-versus-Risk MapEnsemble thinkingRecovery-factor focusExpert in the loop