15-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s power and energy sector
WEEK 2 OF 3 · FOUNDATIONS, PROCUREMENT AND THE FIRST INTERVENTION · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Chapters 5 to 7 · Company FoundationsSESSION 6
The Data Readiness Doctrine
Why This Matters
In Session 5 you chose a use your company can execute. Today you test whether the evidence behind it can carry the weight. A company can own large databases and still lack the evidence a particular decision needs. This is the Data-Foundation Precondition from Session 1 made practical. You do not need perfect corporate data; you need evidence fit for one decision, with its weaknesses visible. By the end you will know which identities must be resolved, which time and unit conventions must be fixed, what your target label really means, where the operational boundary sits and who releases and maintains the dataset.
25-Minute Module
0:00 – 5:00
Usable evidence for one decision
- Database size is a poor substitute for three plain tests: do the records describe the relevant population, arrive in time and connect reliably?
- Start with a data contract: required fields, source systems, owners, meaning, availability, and when a recommendation should be withheld.
- THE DOCTRINE
A missing critical signal should not silently become an ordinary value to keep the model running. State in advance when a recommendation is withheld.
- Keep measured values, estimates and corrections distinct. An instrument reading differs from estimated consumption; an engineer's finding differs from a fault code selected before inspection.
- Those distinctions must survive extraction, transformation and presentation. Figure 5.1 is the author's proposed data-control framework: each stage needs an owner and a response to failure.
Before you ask how much data the company has, ask which records this one decision cannot do without.
5:00 – 11:00
Resolve identity before analysing patterns
- Identity means knowing that two records refer to the same thing. The same transformer can appear under different names in a geographic information system, a maintenance database and a spreadsheet. A replacement asset can inherit an earlier identifier unless the change is recorded, and a model may then combine the histories of different equipment.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. A fictional distribution company wants to rank transformers for inspection, but its maintenance records and network map name units differently and replaced units carry their predecessors' identifiers. Until a reviewed crosswalk exists, the ranking mixes different equipment.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. The same company's commercial team reviewing unusual consumption must know which meter served which account during the period analysed. A present-day mapping can make a meter exchange look like an unexplained consumption change, and, as Session 5 established, an anomaly is not proof of theft. Same loop, seen twice.
- Build the minimum crosswalk, a reviewed mapping between identifiers, and retain uncertain matches for review rather than forcing a confident join. The acceptance standard reflects the consequence: an aggregate planning error differs from an incorrect customer investigation.
- THE DOCTRINE
The crosswalk needs a maintenance process: who records changes, who approves them and which systems receive the correction. One-time cleanup that ignores ordinary equipment and account changes will deteriorate after deployment.
Ask the team how they know which meter served this account last year, not which meter serves it today.
11:00 – 16:00
Make time and units explicit
- Start with the plain distinction. Power is a rate, like the speed on a car's speedometer; energy is the amount delivered over an interval, like the distance on its odometer. kW and MW describe a rate; kWh and MWh describe energy over an interval. Currency per kWh is a different unit again. Preserve instrument multipliers: a model can fit records containing a unit error while producing an answer unusable in engineering or finance.
- A reading can represent an interval, an event time, a reporting time or a correction time. Record the timezone, interval boundaries and whether a timestamp marks the start or end of measurement; joining conventions without reconciliation can create a pattern that exists only in the data preparation.
- Keep missingness visible: a communications failure, a confirmed outage and an absent meter reading may need different treatment. Retain the original status of any imputed value.
- THE DOCTRINE
Record when each value actually became available. Later corrections are not information known at the decision time, and the deployed service must work with evidence the user could obtain then.
When a chart shows a striking pattern, ask whether the timestamps on both sides of the join mean the same thing.
16:00 – 21:00
What the target means, and the boundary you protect
- The target is the outcome a model learns to predict, recorded as a label such as failed, fraudulent, unpaid or complete. Each needs an agreed definition. A suspected irregularity is not a confirmed finding; a payment received is not necessarily a payment allocated correctly.
- Have the process owner and record creators review a sample of labels; keeping uncertainty is preferable to creating false ground truth. Ask how cases entered the dataset: if only inspected accounts have findings, a model can inherit the earlier inspection choices. None of this justifies collecting every available field.
- Protect the operational boundary. A read-only analytical feed can reduce the need for an experimental service to interact with control systems. NIST's Guide to Operational Technology Security, SP 800-82 Revision 3, published in 2023, addresses OT's distinct safety, reliability and performance concerns. Applied to AI: an experimental analytical service should not receive write access to control equipment simply to make integration easier.
- Removing a name does not necessarily make a customer anonymous if account, location or consumption information still identifies them. The privacy owner approves access; an informal claim of anonymisation is not an adequate control.
Data test Acceptance evidence Response to failure Identity Reviewed equipment, account or meter mapping Hold affected cases for reconciliation Time Agreed availability and timestamp rules Exclude or flag incompatible intervals Units Verified definitions and transformations Correct the source or transformation Completeness Missingness assessed against the decision Withhold or qualify the output Labels Reviewed outcomes and uncertainty Revise the task or seek verified findings Access Approved purpose, permissions and retention Stop unauthorised extraction or use
A label is a claim someone made about a case. Ask who made it, and on what evidence.
21:00 – 25:00
Release, maintain, and your wrap
- The intervention does not need every corporate dataset perfected, only the evidence for its decision to meet a defensible standard. Choose the minimum integration and document what it excludes.
- THE DOCTRINE
The data owner issues a readiness statement on evidence; the operating owner confirms the remaining limitations suit the action. A critical weakness is controlled, repaired or the use changed, never averaged away among stronger scores.
- Keep a release record: extraction date, source versions where available, included population, transformations and exclusions. Never exclude a difficult record merely because it reduces performance. The live feed checks for unresolved mappings, stale inputs and missing critical fields, against limits the owner supplies.
- Before you close: list the identities, conventions, target meaning, boundary and release owner for your Session 1 decision. That is Section 6 of your 90-Day AI Readiness Plan. Write your assignment now.
If the same defect is fixed in the analyst's extract every month, it has not been fixed.
Key Terms
Data contract
For one proposed use: the required fields, sources, owners, meaning, availability and the conditions for withholding a recommendation.
Identity crosswalk
A reviewed, maintained mapping of which equipment, meter or account records refer to the same thing in the period analysed.
Power and energy
Power (kW, MW) is a rate; energy (kWh, MWh) is the amount delivered over an interval. Currency per kWh is a different unit again.
Release record
How a dataset was assembled: extraction date, source versions where available, population, transformations and exclusions.
Self-Check — Before You Submit
1. A fictional distribution company's model flags an account whose recorded consumption fell sharply. The records show a meter exchange during the period, and the meter-to-account mapping used is today's. What should the commercial manager do?
- A. Open a theft investigation, because the fall is large
- B. Hold the case until the meter-to-account history is reconciled
- C. Remove the account from the dataset
- D. Impute the missing consumption
B. Hold the case until the meter-to-account history is reconciled — A present-day mapping can make a meter exchange look like an unexplained consumption change, and the response to an identity failure is to hold affected cases. Opening an investigation is the tempting error: an anomaly is not proof of misconduct.
2. Which statement correctly separates power from energy?
- A. kWh measures a rate and kW measures energy over an interval
- B. Power and energy are interchangeable once multipliers are applied
- C. kW and MW describe a rate; kWh and MWh describe energy over an interval
- D. Currency per kWh is a measure of power
C. kW and MW describe a rate; kWh and MWh describe energy over an interval — Power is a rate; energy is the amount over an interval. The trap is thinking multipliers make them interchangeable: multipliers must be preserved, but they do not turn a rate into an amount.
3. A supplier reports strong early-warning performance on a clean historical dataset that includes later corrections. What is the right challenge?
- A. Ask for a larger dataset
- B. Accept the result, because corrected data is more accurate
- C. Ask for a higher accuracy figure before proceeding
- D. Ask whether the evidence was available at the decision time
D. Ask whether the evidence was available at the decision time — Later corrections were not known at the decision time, and the deployed service must work with evidence the user could obtain then. A clean retrospective dataset is useful for analysis, which is exactly why it can flatter a retrospective early-warning claim.
Today’s Assignment
Write Section 6 of your plan: the evidence picture for your Session 1 decision.
- Write the data contract for the decision in your company: required fields, sources, owners, meaning, availability and when a recommendation is withheld.
- Name the identities to resolve, the crosswalk, how uncertain matches are held, and who maintains it.
- State the time and unit conventions, including power or energy, multipliers, missingness and later corrections.
- Define what the target label means, how cases entered the records, and what remains uncertain.
- Name the boundary to protect, who approves access, and who issues the readiness statement and release record.
- The honest part: where your company treats an estimate or suspicion as established fact. If no AI is in use yet, that is normal; work from the decision or proposal in front of you. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 6 — The evidence for our decision: identity, time, units, target meaning and the protected boundary.
How this is marked — the rubric
| Names identities, the crosswalk, uncertain-match treatment and a maintenance owner by role | 25 |
| States explicit time and unit conventions, including power versus energy and availability | 25 |
| Defines the target label and how cases entered the records | 20 |
| Names the boundary, access approver and release owner, with an honest example | 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 contractsIdentity resolutionPower and energy disciplineLabel integrityOperational boundary protection
Chapters 5 to 7 · Company FoundationsSESSION 7
The Organisational Readiness Doctrine
Why This Matters
In Session 6 you tested whether the evidence can carry your decision. Today you test whether the organisation can. An AI programme can have a sponsor, a steering committee and a delivery plan while lacking anyone able to change the operating decision. Leaving that gap to be negotiated later creates an implementation risk the committee could have identified earlier. The duties matter more than the number of committees or job titles. This session puts authority beside responsibility, designs the work with the people who will do it, trains for actual duties, aligns measures with the result you intend, and prepares the manager who inherits the service when the project team leaves. It is the receiving-organisation half of the Sequencing Discipline.
25-Minute Module
0:00 – 6:00
Put authority beside responsibility
- Think of a permit to work: whoever signs it must have authority over the work it releases. The sponsor may secure funding; the process owner needs authority over work; the data owner needs authority to correct records; the model owner carries responsibility for the technical service. A small company may combine roles, but the duties remain distinct.
Duty Accountable contribution Evidence during the intervention Executive sponsor Scope, funding and cross-functional decisions Approved charter and gate decisions Process owner Action policy, capacity and operating result Workflow, resources and outcome record Data owner Meaning, quality and lawful access Data contract and defect register Model owner Evaluation, deployment and monitoring Version record and performance report Control owners Engineering, security, privacy and legal boundaries Approved controls and unresolved issues Finance owner Cost, attribution and recognised value Reconciled value assessment - Include the operating owner before procurement. That manager agrees the baseline, response and capacity required. If the proposal depends on crews or engineers whose priorities are controlled elsewhere, resolve that in the approval, not later.
- THE DOCTRINE
These are duties, not compulsory new positions. Ask whether each contribution has a named person with time and authority to perform it. An impressive organisation chart is not evidence of operating capacity.
- A delivery manager can coordinate finance, engineering, safety, privacy, security and legal contributions without becoming a substitute for the accountable specialists.
For each duty in the table, write a name, not a department. Where you cannot, you have found the gap.
6:00 – 12:00
Design the work with the people doing it
- A recommendation needs a place in the working day: who receives it, how it is prioritised and what evidence the person sees. A separate application that makes users re-enter records into maintenance or billing systems can add enough burden to undermine adoption.
- Observe the task with its users. Ask what happens during a communications failure, a disputed record or an urgent competing incident. These discussions identify requirements; they should not legitimise unsafe shortcuts.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. A fictional generation company's maintenance planners receive equipment alerts. The interface shows the basis of each alert, the age of the evidence and a route to request review, so a planner with better field evidence can challenge it. Overrides are recorded with reasons and reviewed.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. In a fictional distribution company's billing-dispute queue, an officer who finds a wrong customer match must be able to challenge it, not accept it because senior management introduced the tool. Same loop, seen twice: challenge can reveal a model limitation, a record defect or a misunderstood process.
- THE DOCTRINE
Investigate the pattern of overrides without assuming every override is either resistance or wisdom. The outcome of that review belongs in improvement of the service and workflow.
If nobody ever overrides the tool, ask whether they can.
12:00 – 17:00
Train people for their actual duties
- Each group needs something different. Board members need investment evidence, limitations, accountability and reasons to stop. Managers need to know how priorities and measures change. Frontline users need practice with ordinary cases, incorrect outputs and escalation. Data owners need the consequence of late records and ambiguous identities; model owners need enough operating knowledge to see when a technical change affects the decision.
- Training should use the approved workflow and actual permissions. Attendance at a generic AI seminar does not establish that someone can use a recommendation safely. Verify that staff know when to act, when to withhold and what record to keep.
- Plan continuity: name deputies and preserve fallback instructions. Dependence on one enthusiastic employee is an operational weakness even when that individual is highly capable.
- Training and support belong in the cost. A demonstration sustained by a project team differs from a service the receiving department can run with normal resources.
A certificate of attendance tells you who was in the room, not who can act safely on the next shift.
17:00 – 21:00
Align measures with the intended result
- Performance measures influence behaviour. Rewarding closure alone can encourage staff to close alerts without establishing findings. Short-term collection targets can encourage inappropriate treatment of disputed bills. Punishing early escalation can conceal the signals a project-risk model is supposed to reveal.
- A verified correction, a safely completed intervention or an accepted deliverable means more than an output viewed or a task marked complete. This is the Operations-Excellence Lens applied to people: activity is not outcome.
- Review protective measures alongside the primary outcome, as you set them in Session 2, so an apparent gain cannot conceal unfair treatment or an unacceptable operating risk.
- Explain what staff information will be collected and why. A decision-support tool should not quietly become undisclosed monitoring.
Read your current scorecard as a frontline user would, and ask what it rewards them for doing.
21:00 – 25:00
The inheriting manager, and your wrap
- Different teams read the same result differently. A data scientist sees predictive improvement; a planner sees alerts arriving too late; finance sees benefit that has not reached cash. A useful intervention must satisfy each test.
- THE DOCTRINE
Before adoption, the receiving manager reviews the complete service: workload, staff cover, budget, escalation and known limitations. A success sustained by unusually intensive attention may establish feasibility without establishing a sustainable operating model.
- Protect time for monitoring and review, let staff raise concerns without being treated as disloyal to the programme, and make retirement possible. Organisational readiness includes the ability to stop a service with the same accountability used to approve it.
- Before you close: write a name beside each duty for your Session 1 decision. That is Section 7 of your 90-Day AI Readiness Plan. Write your assignment now.
Ask which temporary project supports disappear on the day the project team leaves.
Key Terms
Process owner
The manager with authority over the work, accountable for action policy, capacity and operating result.
Model owner
The person responsible for the technical service: evaluation, deployment and monitoring.
Override
A user's documented decision not to follow a recommendation, recorded with a reason and reviewed.
Receiving manager
The manager who inherits the service after the project team leaves and confirms workload, cover, budget and limitations first.
Self-Check — Before You Submit
1. A fictional distribution company rewards its collections team only for the number of disputed-bill cases closed each month. A new recommendation queue goes live. What is the main risk, and the right response?
- A. Remove all measures from the team
- B. Retrain the model every month
- C. Cases closed without findings; add verified-correction and protective measures
- D. None, because closures show adoption
C. Cases closed without findings; add verified-correction and protective measures — Rewarding closure alone can encourage staff to close cases without establishing findings, and short-term collection targets can encourage inappropriate treatment of disputed bills. Reading closures as adoption is the tempting error: it counts activity, not a verified outcome.
2. How does the book treat the six duties around an AI-supported decision (sponsor, process, data, model, control and finance owners)?
- A. As duties that each need a named person with time and authority, not as compulsory new positions
- B. As six new job titles to create before a pilot
- C. As roles that a delivery manager can perform on everyone's behalf
- D. As a committee whose size shows readiness
A. As duties that each need a named person with time and authority, not as compulsory new positions — The book says these are duties, not compulsory new positions, and a small company may combine roles. A delivery manager can coordinate but is not a substitute for the accountable specialists, and an impressive organisation chart is not evidence of operating capacity.
3. A fictional generation company's pilot succeeded while a project team gave it intensive daily attention. The sponsor wants to adopt it next month. What must happen first?
- A. Adoption can proceed, because the pilot proved the service works
- B. The project team should stay indefinitely to keep results stable
- C. Users should attend a general AI seminar before go-live
- D. The receiving manager reviews the complete service and which temporary supports will disappear
D. The receiving manager reviews the complete service and which temporary supports will disappear — A success sustained by unusually intensive attention may establish feasibility without establishing a sustainable operating model. Treating the pilot result as proof of an operable service is the trap, and a generic seminar does not show anyone can use the recommendation safely.
Today’s Assignment
Write Section 7 of your plan: authority beside responsibility for your Session 1 decision, and the manager who inherits the service.
- Complete the six duties for your decision in your company, with a named role holder, contribution and evidence for each. Mark any duty where nobody has the time and authority.
- Name who must help design the work, where the output sits in their day, and how challenge and overrides work.
- State the training each group needs for its actual duties, with deputies named.
- Name one current measure that rewards the wrong behaviour, and its replacement.
- Describe what the manager inheriting the service must confirm before adoption, and how the service would be retired.
- The honest part: where your company currently assigns responsibility without the authority to act. If no AI is in use yet, that is normal; work from the decision or proposal in front of you. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 7 — Authority beside responsibility, the work designed with its people and the inheriting manager.
How this is marked — the rubric
| Names a role holder for each duty and identifies where authority is missing | 30 |
| Describes the work designed with users, including challenge and overrides | 20 |
| Specifies duty-based training and continuity cover | 15 |
| Replaces a measure that rewards the wrong behaviour with one of verified result | 15 |
| States the inheriting manager's checks and retirement route, with an honest example | 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
Accountability mappingWork design with usersDuty-based trainingMeasure designService handover
Chapters 5 to 7 · ProcurementSESSION 8
Avoiding the AI Money Pits
Why This Matters
Sessions 6 and 7 tested whether your evidence and your organisation are ready. Today you test the purchase. A vendor presentation can combine a software licence, a model service, an integration project, training and a promise of business improvement in one attractive proposal. A company can receive licensed software while still lacking the records and authority needed to obtain value from it. This session teaches you to separate what is being bought, to tell a demonstration from an evaluation, to cost the complete service including currency exposure, to negotiate evidence and exit, to link payment to delivery you can inspect, and to close a pilot deliberately. Each is a place where money can disappear quietly.
25-Minute Module
0:00 – 5:00
Be precise about what is being bought
- Begin with an operating specification: the task, the available evidence, the required output, the review process and the acceptance conditions. Suppliers then offer approaches against the same basis, rather than each demonstrating a favourite capability.
- THE DOCTRINE
Include the simpler alternative. If the AI proposal cannot establish an advantage sufficient to justify its additional cost and exposure, management should choose that alternative. The label attached to the technology should not decide the funding outcome.
- Figure 7.1 is the author's proposed procurement framework: each stage requires evidence before the next commitment. It is the Sequencing Discipline applied to money.
If you cannot write the task on one page without the vendor's slides, you are not ready to buy.
5:00 – 10:00
Separate demonstration from evaluation
- A demonstration shows what a service can do under selected conditions. An evaluation examines it on evidence representative of intended use, including the inconvenient cases. A demonstration dataset should not become the basis for an investment decision unless it is also appropriate evaluation material.
- Control the test records. Ask which the supplier used for development and which remain unseen for the agreed test. For forecasts, preserve information available at the decision time; for document services, include superseded versions, conflicting sources and questions with no supported answer.
- Agree measures before the supplier sees the result, or a favourable metric can be chosen after observation. Acceptance covers useful performance, workflow, integration and controls; a strong technical score cannot compensate for unauthorised access or an inability to export the records the business process requires.
- A bounded investigation can proceed without a production-platform commitment. State its question, cost limit, duration, outputs and stopping route, so it does not quietly become the first instalment of a much larger obligation.
Ask the supplier which of your records they have never seen. Those are the ones to test on.
10:00 – 15:00
Cost the complete service
- The licence is only part of cost. Data preparation, integration, hosting, access, engineering review, training, security, monitoring and support may all consume resources, including internal capacity. Distinguish one-time work from costs that recur with users, data volume, requests or locations.
- State currency exposure. A foreign-currency service can have a different naira cost over the investment period. Finance should provide exchange-rate assumptions, payment dates, applicable taxes and indexation terms, and sensitivities should vary material inputs rather than present one attractive estimate as a commitment.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. A fictional distribution company tests a network analytics service on a single feeder, and the quotation looks modest. Under the pricing schedule, a new feeder, a new document source, a new integration, a model refresh or a migration can each create additional fees. The operating owner learns these triggers before approval, not at expansion.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. Finance converts the foreign-currency fee into naira over the investment period on its own assumptions and shows the sensitivity. A service that fits the first intervention but charges heavily for ordinary expansion may become difficult to justify later.
- For generated language, usage can depend on document length and request frequency; a cheap test on a few short documents may not represent operation across a large collection.
Ask for the price of the second feeder before you sign for the first.
15:00 – 21:00
Negotiate evidence and exit, and pay for what you can inspect
Contract issue Question for procurement Evidence to retain Performance What is tested, on which records and against what alternative? Agreed evaluation and acceptance record Data rights Who may use, retain or train on company information? Purpose, access and retention terms Integration Which systems and operating conditions are supported? Interface specification and failure behaviour Support Who responds when the service fails or changes? Responsibilities and escalation route Change cost Which events create additional charges? Pricing schedule and assumptions Exit Can the company retrieve records and continue the process? Export format, migration test and termination terms - THE DOCTRINE
Demonstrate portability. The company should retrieve inputs, findings, decisions and outputs in a form another service or process can use. Exporting unreadable proprietary files is not a practical exit.
- Set confidentiality and processing conditions: approved locations, relevant subcontractors, and no use of company information in a supplier's wider service or training process without an authorised agreement. Require version information, notice of material changes and regression testing, and keep a route to suspend a changed service that behaves unacceptably.
- Payment milestones should describe deliverables the buyer can inspect: an interface working, records transferred correctly, operating acceptance achieved, each with a named acceptor and evidence. A broad assertion that the AI platform has been installed leaves too much unresolved, and a promised percentage saving without a baseline or comparison can become a dispute rather than protection.
- Test difficult conditions before acceptance: a failed source connection, stale information, changed permissions and an export request.
The time to test the exit is while the relationship is working.
21:00 – 25:00
Close deliberately, and your wrap
- A pilot needs a decision when funding ends: retire it, extend a defined investigation, or adopt it with a funded owner and support. Preserve the evidence, manage dependencies and plan the transition.
- THE DOCTRINE
An extension must name the unresolved question, the evidence required and the additional cost limit. Repeating an experiment without resolving uncertainty postpones the decision.
- Before you close: for your Session 1 decision, write what your company would actually be buying, element by element, and the one cost you have not yet seen on any quotation. That is Section 8 of your 90-Day AI Readiness Plan.
- Pop Quiz 2 opens now: five questions, closed book, on Sessions 6 to 8. Sit it before Session 9, then write your assignment.
Challenge an extension as carefully as you challenged the first commitment.
Key Terms
Operating specification
The task, evidence, required output, review process and acceptance conditions every supplier is compared against.
Evaluation
Testing on evidence representative of intended use, including unseen records and inconvenient cases, against measures agreed in advance.
Complete service cost
All one-time and recurring costs, including internal capacity, currency exposure and fee triggers.
Portability
The demonstrated ability to retrieve inputs, findings, decisions and outputs in a usable form.
Self-Check — Before You Submit
1. A fictional mini-grid developer's supplier offers to link part of its fee to a promised percentage saving. The developer has no agreed baseline for the decision. What should the commercial director do?
- A. Accept, because it protects the developer
- B. Reject all performance terms
- C. Pay now and measure later
- D. Agree the baseline, comparison, dependencies and attribution evidence first
D. Agree the baseline, comparison, dependencies and attribution evidence first — Results can depend on customer policy, available crews, engineering approvals and conditions outside the supplier's control. A promised percentage saving without a baseline or comparison can become a dispute rather than protection, which is why accepting it as protection is the tempting error.
2. Which is the strongest procurement milestone for releasing a payment?
- A. Records transferred correctly, accepted by a named owner on stated evidence
- B. The supplier confirms the AI platform has been installed
- C. A successful demonstration to the executive committee
- D. The project schedule date for go-live has arrived
A. Records transferred correctly, accepted by a named owner on stated evidence — A milestone should describe a deliverable the buyer can inspect, with a named acceptor and supporting evidence. A broad assertion that the platform is installed leaves too much unresolved, and a demonstration or a date arriving does not establish delivery.
3. When should the measures used to accept a supplier's service be agreed?
- A. After the test, once the most informative metric is known
- B. Before the supplier sees the result
- C. When the supplier proposes its preferred metrics
- D. Only at contract renewal
B. Before the supplier sees the result — If measures are agreed after observation, a favourable metric can be chosen to fit the result. Choosing the most informative metric after the test sounds reasonable, but it is how a weak service can pass acceptance.
Today’s Assignment
Write Section 8 of your plan: what your company would buy for your Session 1 decision, its complete cost and the evidence, exit and payment terms you would require.
- Write the operating specification for the purchase your company would consider, separate what is being bought, and name the simpler alternative.
- Separate demonstration from evaluation: unseen records, inconvenient cases and measures agreed in advance.
- Cost the complete service: one-time and recurring items, internal capacity, currency exposure with the assumptions finance must supply, and the events that trigger additional fees. Do not invent figures; where you do not yet have a number, name who will supply it.
- Set your terms on performance, data rights, integration, support, change cost and exit, inspectable payment milestones and closure.
- The honest part: where your company has previously paid for something it could not inspect, could not exit or could not operate after the supplier left. If no AI is in use yet, that is normal; work from the proposal in front of you. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 8 — What we would buy, the complete cost and the evidence, exit and payment terms.
How this is marked — the rubric
| Writes an operating specification that separates the purchased elements and names the simpler alternative | 25 |
| Separates demonstration from evaluation with unseen records, inconvenient cases and measures agreed in advance | 20 |
| Costs the complete service, including internal capacity, currency exposure and fee triggers, without invented figures | 25 |
| Sets evidence, exit, inspectable payment and closure terms, and gives an honest example from past purchasing | 30 |
| 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, People and Money Pits 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
Operating specificationEvaluation designComplete service costingExit and portabilityInspectable payment milestonesDeliberate pilot closure
Chapters 8 to 13 · The First InterventionSESSION 9
The Right First Step
Why This Matters
You now have the evidence, organisation and procurement tests. Today you write the charter for a first intervention on your decision, before anyone arranges a demonstration. The book's first step is deliberately modest: one useful decision the company can investigate at a controlled cost. A narrow question is easier to assess than an ambition to transform the whole organisation. The charter states the question, the baseline and comparison, the design around the capacity your teams actually have, and an observation period long enough to reveal the result. The aim is not a successful pilot. The aim is an intervention that lets management choose, on evidence, whether to scale, investigate further or stop.
25-Minute Module
0:00 – 5:00
A question the intervention can answer
- The book frames the first question in one sentence: whether improved evidence can change this decision, for this population, through this workflow, compared with the method used today. Every part of that sentence is a boundary.
- Bound the intervention by process, population and authority: one equipment class, account-investigation process, document category or project package. A tiny demonstration can miss relevant variation; a broad rollout can create exposure before the team understands the service.
- THE DOCTRINE
The operating owner should nominate the question and explain why it deserves attention. Technology assesses feasibility, finance examines the value path and control owners determine whether the use is acceptable.
- The book's Figure 8.1 is the author's pilot framework, from a bounded question to a decision on further investment. It is explicit that ninety days is a planning horizon, not a guarantee of sufficient evidence.
If your question needs the word transform, it is not yet a question.
5:00 – 10:00
Write the charter before the demonstration
- The charter states the current process and proposed change, the decisions the service may support and the actions it may not take, the baseline, evaluation method, cost ceiling, responsibilities and the evidence required at the final review. It should be understandable without a vendor presentation.
- Success requires both a useful operating result and appropriate controls. For inspections, useful findings matter alongside false accusations, complaint handling and safe execution; for maintenance, treatment must remain within approved engineering practice.
- State stopping conditions in advance. A material privacy breach, unsafe behaviour or failure of an essential control can justify suspension regardless of the reported benefit; other weaknesses may call for repair and retest. Name who can suspend the intervention and who can authorise its restart.
- The cost ceiling includes internal capacity: record what is deferred or redirected so the committee sees the opportunity cost.
Write the stopping conditions while everyone is still calm.
10:00 – 16:00
Baseline, comparison and the review stages
- A baseline is the agreed starting measure; a comparison is what you set the result against, so the intervention's effect can be told apart from everything else that changed. Before deployment, agree the outcome, population, measurement period, exclusions and source records.
- THE DOCTRINE
Where a credible baseline cannot be established, the first task is measurement or reconciliation. Do not create a starting value simply to keep a project schedule intact.
- Changes in supply, weather, tariffs, staffing or collection policy can affect results. Identify plausible alternative explanations and, where feasible and appropriate, compare similar groups or introduce the change in stages, respecting customer duties, safety and practical operation.
- A shadow stage shows what the service would have recommended without changing the decision: whether records arrive in time and alerts are manageable. It cannot prove the consequence of actions never taken.
Review stage Required evidence Available decision Scope approval Decision, owner, baseline plan and permitted action Fund preparation or decline Data review Required records, access and critical defects assessed Proceed, repair or revise scope Shadow review Timely outputs, manageable workload and controls tested Permit bounded live use or retest Live review Verified actions, results, costs and incidents Stop, investigate further or scale with conditions - These are proposed management arrangements: a low-risk administrative task and a safety-related operating service should not automatically share the same review requirements.
A starting value invented to protect a schedule will be defended long after everyone has forgotten it was invented.
16:00 – 21:00
Design around capacity, and observe long enough
- Determine how many outputs staff can review and how many actions the company can complete. If the queue runs far beyond that capacity, users may select cases arbitrarily or ignore it, and the operating method then differs from the evaluated design.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. A fictional transmission asset owner's crews can complete only so many inspections in a period. The useful test is whether the cases the service selects within that capacity lead to better verified findings than the cases selected by the current method, not its classification score across every asset.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. A fictional distribution company's account-investigation team faces the same arithmetic. Recording why recommendations did not lead to action, such as no access, further evidence needed or urgent competing work, can show that the next investment belongs in operating capacity rather than another model.
- Choose an observation period that can reveal the result. A frequent document task generates evidence quickly; a rare equipment failure may not occur during the intervention. Distinguish evidence sufficient to assess the workflow from evidence sufficient to establish a reduction in failure exposure.
- An absence of observed harm does not, by itself, establish safety or reliability. Report evidence limits rather than present a clean incident log as a universal outcome. Monitor whether comparison groups remain comparable.
A ranking is judged at the capacity you have, not the capacity you wish you had.
21:00 – 25:00
Learn without rewriting, and your wrap
- Keep a dated record of changes. If the target, population or action policy changes, distinguish the original test from the revised one. Presenting several different tests as one uninterrupted success obscures what the evidence supports.
- THE DOCTRINE
Report failed cases as well as successful ones. An intervention showing that a simpler method is adequate, or that available information cannot support the proposed use, can prevent a much larger wasted commitment.
- The final recommendation is specific. Scaling requires ownership, funded support, monitoring and approvals; further investigation requires an unresolved question and a limit; stopping requires an orderly return to the existing process and appropriate retention of records.
- Before you close: write the one-sentence question for your Session 1 decision, and the baseline you would agree before anything is switched on. That is Section 9 of your 90-Day AI Readiness Plan. Write your assignment now.
The intervention has done its job when management can choose on evidence, whatever the choice.
Key Terms
Intervention charter
The written basis for a first intervention: scope, permitted actions, baseline, evaluation, cost ceiling, responsibilities and stopping conditions.
Baseline
The agreed starting measure for a defined population, period and source records, fixed before deployment.
Shadow stage
Running a service alongside the current process to see what it would have recommended, without changing any operating decision.
Stopping condition
A condition stated in advance, such as unsafe behaviour, that justifies suspension regardless of reported benefit.
Self-Check — Before You Submit
1. A fictional distribution company cannot yet reconcile last year's collection records for the population in scope. The project team proposes an estimated starting value so the pilot can start on schedule. What should the sponsor decide?
- A. Approve the estimate, since it can be corrected later
- B. Start the pilot without any baseline and compare informally
- C. Make measurement or reconciliation the first task before deployment
- D. Use the sector aggregate as the baseline instead
C. Make measurement or reconciliation the first task before deployment — Where a credible baseline cannot be established, the first task is measurement or reconciliation, and a starting value should not be created to keep a schedule intact. A sector aggregate is no substitute for the company's own reconciled baseline, as Session 2 showed.
2. What can a shadow stage establish?
- A. Whether records arrive in time, alerts are manageable and explanations are useful
- B. The financial benefit of actions the service recommended
- C. That the service is safe in live operation
- D. The consequence of actions that were never taken
A. Whether records arrive in time, alerts are manageable and explanations are useful — A shadow stage shows what the service would have recommended without changing the decision. It cannot prove the consequence of actions never taken, which is why a bounded live stage is needed before any benefit is claimed.
3. A fictional generation company's pilot on a rare equipment failure mode records no incidents during its observation period. What does that establish?
- A. That the service has reduced failure exposure
- B. That the service is reliable and can be scaled
- C. That the observation period was long enough
- D. Little by itself: the absence of observed harm may reflect limited exposure or a short period
D. Little by itself: the absence of observed harm may reflect limited exposure or a short period — A rare failure may not occur during the intervention, and an absence of observed harm does not, by itself, establish safety or reliability. Reading a clean incident log as reduced exposure is the tempting error; examine control design and failure routes too.
Today’s Assignment
Write Section 9 of your plan: the charter for a first intervention on your Session 1 decision.
- Write the one-sentence question for your company's first intervention, bounded by process, population and authority, and name the operating owner who nominates it.
- Write the charter: proposed change, permitted and prohibited actions, cost ceiling including internal capacity, and stopping conditions with who may suspend and restart.
- Set the baseline and comparison, with plausible alternative explanations.
- Design around capacity: how many outputs your teams can review and act on, and how the reasons for no action will be recorded.
- Choose the observation period and state what it can and cannot reveal.
- The honest part: where your company has previously judged a project against a baseline created to keep a schedule. If no AI is in use yet, that is normal; the charter may conclude that a simpler method should be tested first. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 9 — The intervention charter: question, baseline, comparison, capacity and observation period.
How this is marked — the rubric
| States a bounded question by process, population and authority with a nominating operating owner | 20 |
| Writes a charter with permitted and prohibited actions, a cost ceiling including internal capacity and stopping conditions with named authority | 25 |
| Sets a baseline and comparison with alternative explanations identified | 25 |
| Designs the intervention around actual review and action capacity | 15 |
| Chooses an observation period with its evidence limits stated, and gives an honest past example | 15 |
| 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
Bounded question designIntervention charterBaseline and comparisonCapacity-based evaluationEvidence-limited reporting
Chapters 8 to 13 · The First InterventionSESSION 10
Buy, Build, or Wait
Why This Matters
In Session 9 you chartered a first intervention. Today you decide who will carry it. Buying and building are not always opposite choices: a company can buy a model service, configure its own workflow and retain its data controls, or decide that the immediate investment should repair records or processes rather than introduce a model at all. Compare complete operating arrangements, not technology labels. A demonstration can be assembled quickly by a small team; a supported service needs security, continuity, monitoring, documentation and incident response for its whole operating life. This session closes Week 2. You will choose to buy, build, combine or wait, on the responsibility your company can actually sustain.
25-Minute Module
0:00 – 4:00
Responsibility the company can sustain
- Start with strategic relevance. A well-supported product can be suitable for a common task. A decision depending on distinctive operating evidence, proprietary processes or specialist integration may justify greater internal control, without meaning that every technical component should be built internally.
- Commissioning a plant is a different commitment from running it, and the same is true of an analytical service. Assess the cost of maintaining security, continuity, monitoring, documentation and incident response before choosing the sourcing route.
- THE DOCTRINE
Compare complete operating arrangements rather than technology labels. The outcome sought is a capability the company can maintain at a defensible cost, with responsibilities it can meet.
Ask who will answer for this service long after the launch, and whether they know it yet.
4:00 – 9:00
What a purchased service must demonstrate
- Require evidence of relevant interfaces, access controls, export capability and failure behaviour. Reference customers and marketing material can inform diligence; they cannot replace evaluation on your intended use.
- Establish who manages configuration changes, such as account categories or asset mappings, how they are tested and whether they create fees.
- Treat continuity as an operating question: who responds if a source system fails or the supplier changes a material component? A general availability promise may not describe whether useful evidence reaches staff.
- Configuration should preserve the narrower approved use. A product might technically support automatic action your company has not authorised. Do not expose every feature and rely on staff to remember the charter; keep the service on the rung of the Autonomy Ladder that was approved.
- Demonstrate export and fallback while the relationship is working, as Session 8 required.
A reference customer tells you the product worked somewhere. Your evaluation tells you whether it works here.
9:00 – 14:00
What an internal build must carry
- An internal team needs a defined service boundary, a product owner, a release process and access to the people who understand the decision. Skilled developers are necessary; technical skill alone does not establish operating fitness.
- Maintenance involves more than retraining. Source schemas can change, identifiers can be revised, and a new operating policy can change the meaning of an output. A model may remain statistically stable while becoming unsuitable because its action policy has changed.
- ON THE PLANT AND NETWORK · THE WORKED CASE
- Plant and network. A fictional generation company's engineers build an equipment-condition model on their own records. The build carries its dependencies: revised asset identifiers, changed source schemas and a maintenance policy that changes what an alert means. Each needs an owner.
IN THE COMMERCIAL CHAIN · THE PARALLEL- Commercial chain. Finance treats the knowledge the build creates about company records as a learning objective with its own cost ceiling, described separately from the expected operating return. Learning should not become an after-the-event explanation for a financial claim the project failed to establish.
- THE DOCTRINE
A critical capability dependent on an undocumented personal laptop is not a sustainable operating arrangement. Documentation and access should allow another qualified person to diagnose a problem.
- Record how the service was built and evaluated, keep access and secrets under company control, and budget for support and skill renewal.
Ask who could fix it if the person who built it moved on.
14:00 – 19:00
When waiting is the better investment
- Waiting can be rational where required evidence is unavailable, action cannot be executed or an essential control remains unresolved. It should be an active decision with a repair plan: what must change, who owns it and when the candidate will be reconsidered.
- Record repair may have value independently of AI: correct asset identities, verified work findings and reconciled accounts improve ordinary operations. If that work resolves much of the original problem, the eventual model investment may become smaller or unnecessary; that is a useful result of the Sequencing Discipline.
- Waiting is less useful when it means refusing any bounded use until the whole organisation reaches an undefined level of digital maturity. Assess readiness at the decision level: a low-consequence document task may be feasible while a maintenance prediction is not.
Choice Appropriate condition Responsibility retained Buy A product fits the task and can pass local acceptance Lawful records use, operating policy and supplier oversight Build Distinctive needs justify control and support capacity exists Full service ownership, evaluation and maintenance Combine Commercial components fit a controlled company workflow Integration, boundaries and complete accountability Wait and repair A critical evidence, control or execution gap remains A funded repair plan and dated reconsideration - These choices remain revisable, but significant changes belong in a fresh investment review, not informal extensions of an earlier approval.
Waiting is a decision with an owner and a date. Without them it is drift.
19:00 – 25:00
Compare consistently, and close Week 2
- Use a common scope and period: implementation, recurring service, internal support, controls and exit. Comparing a full-service supplier quotation with an internal estimate covering only developer time would mislead the committee.
- THE DOCTRINE
Risk belongs in the comparison. A cheaper option may require access or dependence the company cannot accept. A single weighted score cannot settle a safety or legal condition that requires separate approval.
- Deployment location changes responsibility. A non-urgent analytical task may wait for connectivity; a working queue may need locally available approved records; a safety-related process needs its existing authorised control arrangements. Hosting choice does not establish that access, patching and monitoring controls are in place.
- Before you close: write buy, build, combine or wait for your Session 1 decision, with the responsibility your company would retain. That is Section 10 of your 90-Day AI Readiness 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.
If the route changes during delivery, revisit the case. The approval was for an arrangement, not a label.
Key Terms
Combine
Commercial components inside a controlled company workflow; the company keeps integration, boundaries and accountability.
Wait and repair
An active decision to defer a model while a critical gap is repaired, with a funded plan and a dated reconsideration.
Action policy
The rule for what people do with an output. If it changes, a statistically stable model can become unsuitable for the decision.
Rollback
A tested route back to the previous working configuration after a failed deployment, supported by the release record.
Self-Check — Before You Submit
1. A fictional mini-grid operator cannot reliably link meters to customer accounts, and its collections staff cannot act on more cases than they already handle. A supplier offers a ready product used by other operators. What is the most defensible choice?
- A. Buy now, because the product is proven elsewhere
- B. Wait and repair, with an owner and a dated reconsideration
- C. Build internally, because control removes the record problem
- D. Combine the product with an internal workflow now
B. Wait and repair, with an owner and a dated reconsideration — Waiting is rational where required evidence is unavailable or action cannot be executed, provided it is an active decision with a repair plan. Buying because it is proven elsewhere is the tempting error: no sourcing route supplies missing records or capacity.
2. A committee compares a supplier's full-service quotation with an internal estimate covering only developer time, and favours the build as cheaper. What is wrong?
- A. Nothing, provided both figures are in naira
- B. Internal builds always cost more
- C. The supplier quotation should be discounted for risk
- D. The scopes differ: the estimate omits support, controls and exit
D. The scopes differ: the estimate omits support, controls and exit — Costs must be compared on a common scope and period, including implementation, recurring service, internal support, controls and exit. Assuming either route is always cheaper is also a trap.
3. A purchased product can technically take automatic action on customer accounts, but your company has approved it only to recommend cases for review. How should it be configured?
- A. To preserve the narrower approved use
- B. With every feature enabled, relying on staff to follow the charter
- C. With automatic action on for a trial period
- D. As the supplier's standard operating method recommends
A. To preserve the narrower approved use — Configuration should preserve the narrower approved use rather than expose every feature and rely on staff to remember the charter. Enabling automatic action for a trial grants authority the company has not approved, which is a different decision on the Autonomy Ladder.
Today’s Assignment
Write Section 10 of your plan: buy, build, combine or wait for your Session 1 decision, decided on the responsibility your company can sustain.
- Choose buy, build, combine or wait and repair, justified on strategic relevance and the responsibility your company can sustain.
- If buying or combining: what the purchased service must demonstrate on your intended use, and how configuration keeps the approved use narrow.
- If building or combining: what the internal build must carry, with any learning objective kept separate from the operating return.
- If waiting: write the repair plan, naming what must change, who owns it and the date of reconsideration.
- Compare the options on a common scope and period, and name any safety or legal condition that needs separate approval.
- The honest part: where your company has previously compared a full quotation with a partial internal estimate, or called drift a decision to wait. If no AI is in use yet, that is normal; the right answer may be to wait and repair. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 10 — Buy, build or wait, decided on the responsibility we can sustain.
How this is marked — the rubric
| Justifies the route on strategic relevance and sustainable responsibility | 25 |
| States what the service must demonstrate or the build must carry, or a dated repair plan | 25 |
| Compares options on a common scope and period, with separate approval conditions named | 30 |
| Keeps the choice revisable, with an honest example | 20 |
| 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, Procurement and the First Intervention 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
Sourcing decisionsSupplier diligenceInternal build ownershipActive waitingConsistent cost comparison