Lesson 1, Topic 1
In Progress
Unit 2: Teacher Resources
Instructor Resource · AIAS Certification
Unit 2 Teacher Resources
Strategic AI Integration
Everything you need to teach, assess, and support students through Unit 2 — lesson guides, full answer keys, rubrics, differentiation strategies, discussion prompts, and capstone grading support.
Unit 2 At a Glance
4
Lessons: 2.1 through 2.4
5–6 hrs
Estimated student completion time
20
Quiz questions (5 per lesson)
40 pts
Capstone project total
| Lesson | Title | Core Skills | Key Interactive Tools | Est. Time |
|---|---|---|---|---|
| 2.1 | Assessing AI Readiness & Business Needs | Five readiness pillarsBusiness needs mappingReadiness phases | AI Readiness CalculatorNeeds-to-AI Mapper | 60–75 min |
| 2.2 | Aligning AI with Business Strategy & Goals | KPI definitionVanity metric detectionAlignment test | KPI Alignment CheckerAlignment Scenario Mapper | 60–70 min |
| 2.3 | Designing AI Solutions for Specific Use Cases | 7-step design processTechnique matchingBuild vs. buyShadow mode pilots | 7-Step StepperTechnique MatcherSolution Builder | 70–85 min |
| 2.4 | Strategic AI Integration Project (Capstone) | End-to-end AI strategyProfessional deliverable | 3-Part WorkspacePDF Generator | 60–90 min |
Lesson-by-Lesson Resources
Lesson 2.1 — Assessing AI Readiness & Business Needs
🎯 Lesson Purpose: Students learn to evaluate organizational AI readiness before any technical decisions are made. The five-pillar framework is the diagnostic tool they'll use in the capstone project and in professional practice. Emphasize that readiness is about all five pillars — not just technology.
Learning Objectives
By end of lesson, students will know:
The five readiness pillars and what each covers
Why the weakest pillar constrains the whole initiative
The four readiness phases (Foundation → Transformation)
How to map a business pain point to an AI opportunity
By end of lesson, students will be able to:
Rate an organization on all five pillars with justification
Identify the critical blocker in a described scenario
Assign the correct readiness phase with reasoning
Map a quantified pain point to an AI solution type
📝 Quiz Answer Key — Lesson 2.1
Q1. A retail company has excellent cloud infrastructure, a strong ML engineering team, and enthusiastic CEO support for AI. Their data is siloed across four legacy systems with inconsistent formatting. What is their most accurate AI readiness status?
BNot ready to deploy AI — their data pillar is a critical blocker that will undermine any model built on siloed, inconsistent data.
The "lowest pillar rule" is the key concept — overall readiness is constrained by the weakest pillar regardless of strengths elsewhere. Data quality is the most common critical blocker in real deployments.
Common wrong answer — A: Students who focus on strengths rather than applying the lowest-pillar rule will choose A. Redirect: "If the foundation is broken, does it matter how good the house looks?"
Q2. A hospital wants to use AI to improve patient outcomes. In a business needs assessment, which question is MOST useful for identifying where AI could add value?
BWhich diagnostic or operational process has the highest error rate or longest delay — and what does that cost in outcomes or dollars?
Business needs assessment starts with quantified pain — not competitor benchmarking, not headcount, not available technology. The question that surfaces the most valuable AI opportunity is always about cost, error, or delay.
Common wrong answer — A: Competitor analysis is a common instinct but leads to imitation rather than need-driven AI strategy. The correct approach is internal pain-point identification first.
Q3. An AI team builds a highly accurate fraud detection model. The operations team, who wasn't involved in the project, continues using the old manual process because they don't trust the AI's outputs. Which readiness pillar MOST clearly failed?
CPeople & Process — the end users weren't involved, trained, or prepared to adopt the AI into their workflow.
This scenario illustrates the most common deployment failure: technically excellent AI that nobody uses. The People pillar (involvement, training, trust-building) and Process pillar (workflow integration, change management) both failed here.
Q4. An organization is deploying AI for the first time. They have decent data infrastructure but no prior AI experience. Which phase should they enter FIRST?
CExploration — run a small, focused pilot project on a high-value use case to build skills and generate proof of value.
With decent infrastructure but no AI experience, skipping Foundation and going to Exploration is appropriate. Full Foundation (months of data work before any AI) would be unnecessary given the decent infrastructure. Jumping to Operation or Transformation without pilot validation is the phase-skipping trap.
Common wrong answer — D: Students may choose Foundation, interpreting "no prior AI experience" as requiring extensive groundwork. Clarify: Foundation is for organizations with significant infrastructure gaps. This organization's data is "decent" — it's ready for a pilot.
Q5. A VP says: "We're losing customers to competitors who have AI-powered recommendation engines. We need to build one immediately." Which response BEST reflects strong AI strategy thinking?
BLet's first assess whether customer churn is actually caused by lack of recommendations, evaluate our data readiness for personalization, and define what success looks like — then build a roadmap.
This tests the "strategy before technology" principle. The VP's instinct is competitive-fear driven, not evidence-based. Strong AI strategy starts with validating the problem diagnosis, assessing readiness, and defining success before any technical decision.
💬 Discussion Prompts — Lesson 2.1
Pillar Analysis
Think about an organization you know — your school, a past employer, or a company you follow. Rate its AI readiness across all five pillars from 1–5. Which pillar would be the critical blocker, and why?
Look for: specific evidence supporting each rating, identification of interdependencies between pillars, and concrete suggestions for addressing the blocker. Strong responses will reference real organizational details rather than generalizing.
Cold Start Discussion
A startup has world-class ML engineers and cloud infrastructure but only 6 months of business data. How should they approach their first AI initiative — and what would you advise them to do in the 12 months before building any model?
Expect students to recommend: starting with rule-based systems, collecting and labeling data systematically, defining the target ML problem early so data collection is purposeful, and identifying a use case where small data can still provide value (clustering/unsupervised).
Leadership Scenario
A CEO says "We need to implement AI immediately — our board is watching." An AI lead responds "We're not ready yet." How should that conversation go, and what would you recommend the AI lead say to manage expectations while still making progress?
Good responses will balance urgency (finding a quick-win pilot) with honesty about risks (data gaps, skills). The concept of "pilot as risk management" is key — it satisfies leadership's need for action while protecting against large-scale failure.
⚠️ Common Misconceptions — Lesson 2.1
Misconception
"AI readiness is mainly about having the right technology and enough budget."
Correct Understanding
Technology is one of five pillars. Organizations with excellent infrastructure frequently fail because of data gaps, skill deficits, leadership misalignment, or process immaturity. Readiness is holistic.
Misconception
"Removing protected attributes (like race) from an AI model eliminates bias."
Correct Understanding
Proxy variables (zip code, school attended, credit history) correlate with protected characteristics and carry bias into the model indirectly. This is covered in depth in Unit 2's ethics content.
Misconception
"We should deploy AI and fix readiness gaps in parallel to save time."
Correct Understanding
Deploying before foundations are ready is the single most common cause of the 85% AI failure rate. Readiness gaps that are known at deployment become embedded technical debt that compounds post-launch.
Misconception
"If leadership is enthusiastic about AI, the leadership pillar is strong."
Correct Understanding
Enthusiasm without realistic expectations is dangerous. The leadership pillar requires: clear vision tied to business goals, budget authority, understanding of limitations, and willingness to accept a pilot approach rather than demanding instant enterprise-wide deployment.
Lesson 2.2 — Aligning AI with Business Strategy & Goals
🎯 Lesson Purpose: Students learn to connect AI initiatives to measurable business outcomes. The critical distinction between aligned KPIs and vanity metrics is a concept that will appear repeatedly throughout the certification and is heavily tested in the exam. The three-question alignment test is the practical takeaway students should be able to apply to any AI proposal.
📝 Quiz Answer Key — Lesson 2.2
Q1. A retail company deploys an AI recommendation engine. The technical team reports "97% recommendation relevance score." Six months later, average order value has not changed. What is the MOST likely problem?
BThe team optimized for a vanity metric — relevance score — instead of a business KPI directly tied to order value or conversion rate.
Relevance score is a classic technical proxy metric that can be high while delivering zero business value. The correct KPI for a revenue-growth AI initiative is a direct business measure: average order value, conversion rate, or revenue per session.
Common wrong answer — A: Students often default to "more data/better model" as the solution. Emphasize: a 97% accuracy model can still be completely worthless if it's measuring the wrong thing.
Q2. A hospital wants to reduce patient readmission rates using AI. Which represents a fully aligned AI initiative?
BBuild an AI system that predicts high-risk patients before discharge, with the CMO as sponsor, a baseline of 14% readmission rate, and a success threshold of 9% within 12 months.
Full alignment requires all four elements: (1) specific measurable goal, (2) named metric owner/sponsor, (3) baseline measurement, (4) success threshold. Option B is the only one containing all four. Use this question to reinforce the four-element alignment checklist.
Q3. UPS deployed its ORION AI routing system with a single KPI: miles driven per package. This is an example of what concept?
BNorth Star metric — one metric that directly captures the core efficiency value the AI delivers, making success unambiguous.
Miles per package is elegant because it's (a) directly measurable, (b) directly connected to fuel cost and vehicle wear, and (c) impossible to improve accidentally — you can only improve it by actually routing better. This is what distinguishes a North Star metric from a vanity metric.
Q4. A VP of Sales proposes an AI initiative to "improve the sales process." What is the MOST important first step?
BWork with the VP to define a specific, measurable KPI (e.g. "increase conversion rate from 12% to 18%"), identify the baseline, and set a success threshold before any technical work begins.
The alignment-first principle: all strategic and measurement decisions must precede technical decisions. A vague goal guarantees misalignment — the data science team and the VP will each build toward what they think the goal means.
Q5. Mastercard's fraud AI tracks both fraud catch rate AND false positive rate as dual KPIs, treating improvement in one at the expense of the other as a failure. What principle does this illustrate?
BBalanced KPI design — the business goal requires tracking the tension between two metrics, not optimizing one in isolation.
This question tests nuanced understanding: not all AI goals can be captured in a single metric. When a business goal involves competing considerations (fraud prevention vs. customer experience), dual KPIs that measure both sides of the tension are the correct design. Single-metric optimization in this context would be a mistake.
Common wrong answer — D: Students who over-apply the "North Star metric" concept will choose D. Clarify: North Star works for single-dimension goals; complex goals with genuine tradeoffs require balanced dual KPIs.
💬 Discussion Prompts — Lesson 2.2
KPI Audit
A social media company reports that its AI content recommendation system increased "daily active users" by 22%. Is this a good AI KPI? What additional information would you need to know whether the AI is actually delivering business value?
Strong responses will identify: DAU is a leading vanity indicator without revenue context. Needed: revenue per DAU, subscription conversion rate, advertiser CPM, and — critically — whether DAU increased because of better recommendations or simply more addictive content (which may not be a positive outcome).
The Moving Goalpost
A bank deploys a loan risk AI targeting default rate reduction. Halfway through, the compliance team requires the model to also optimize for fair lending metrics. How should the AI team handle this, and what alignment failure does this reveal?
The core failure: the compliance team wasn't involved in original scoping. Correct handling: hold a stakeholder alignment session, define the priority order (compliance as a hard constraint, default rate as the optimization target within that constraint), document the decision, and get both teams to sign off before proceeding.
Spectrum Placement
Place each of these AI initiatives on the four-stage alignment spectrum and justify your placement: (a) "We want to use AI to improve customer experience." (b) "We want to reduce churn from 18% to 12% using ML." (c) "We want to reduce churn from 18% to 12% using an ML model built on our CRM data, with the VP of Customer Success as sponsor and a 90-day review gate."
Expected: (a) Stage 2 — vague goal, no KPI. (b) Stage 3 — goal and KPI defined but no named owner, no baseline confirmation, no stated success threshold. (c) Stage 4 — all elements present. This exercise builds the habit of checking alignment completeness before approving any project.
⚠️ Common Misconceptions — Lesson 2.2
Misconception
"High model accuracy means the AI initiative is succeeding."
Correct Understanding
Model accuracy is a technical metric, not a business metric. An AI can be 97% accurate and deliver zero business value if it's solving the wrong problem or measuring the wrong outcome. Business impact is the only meaningful definition of AI success.
Misconception
"Every AI project needs a single North Star metric."
Correct Understanding
North Star works for single-dimension goals. When a business objective involves a genuine tradeoff (fraud prevention vs. customer friction; speed vs. accuracy), balanced dual KPIs are more appropriate than forcing everything into one number.
Misconception
"If you define a baseline after deployment, you can still prove the AI worked."
Correct Understanding
A post-deployment baseline is not a baseline — it's a starting point with no comparison. Without a pre-deployment measurement, it's impossible to prove causality between the AI and any change. Baselines must be established before build begins.
Misconception
"AI alignment is a one-time task done at project start."
Correct Understanding
Alignment requires ongoing maintenance. Business priorities shift, new stakeholders emerge, and regulatory requirements change. The Mastercard dual KPI example illustrates how even well-aligned projects require continuous stakeholder re-alignment as conditions evolve.
Lesson 2.3 — Designing AI Solutions for Specific Use Cases
🎯 Lesson Purpose: The most technically dense lesson in Unit 2. Students don't need to implement any technique — they need to match problem types to techniques and understand why design decisions are made. The seven-step process and the six system components are the key frameworks. Emphasize: the model is 20% of the system.
📝 Quiz Answer Key — Lesson 2.3
Q1. A data science team is assigned to "use AI to improve the hiring process." Before writing a single line of code, what is the MOST important first action?
BWrite a precise problem definition: what input data the AI receives, what output it produces, what counts as correct, and what a human does with the output.
Problem definition is Step 1 and the most important step in the seven-step process. A vague brief produces six different solutions from six engineers. The input/output/success/human workflow structure is what makes a problem definition actionable.
Q2. A new e-commerce platform has 3 months of transaction data with only 47 confirmed fraud cases. What is the MOST appropriate design response?
BRecognize this as a cold start problem — use rule-based fraud detection while collecting labeled data for 6–12 months before attempting supervised ML training.
47 labeled examples is a textbook cold start problem. Training a supervised classifier on this would produce a model that's confidently wrong. The correct design: rule-based now, supervised ML after sufficient data collection. This also tests the concept of the cold start problem from the vocabulary section.
Common wrong answer — D: Students may suggest unsupervised anomaly detection as an alternative to the cold start problem. This is partially correct — anomaly detection doesn't require labels — but the answer asks for the "most appropriate design response," which includes the temporal plan (collect labeled data before transitioning to supervised ML).
Q3. A hospital must deploy a diagnostic AI where every decision must be explainable to patients and regulators. Which technique selection principle applies?
BRegulatory explainability constraints eliminate black-box models from consideration — technique selection must favor interpretable models or black-box models with XAI explanation layers.
This tests the concept that design-time constraints (Step 7) directly affect technique selection (Step 3). Regulatory requirements are not post-deployment concerns — they constrain the design from the start. The EU AI Act explicitly requires explainability for High-Risk AI in healthcare.
Q4. A logistics company uses a standard cloud mapping API for routing. An ML team proposes building a custom RL system from scratch. What build vs. buy question should be asked FIRST?
BDoes our routing problem require proprietary business logic or competitive differentiation that a standard vendor solution cannot provide?
The build vs. buy question is fundamentally about competitive differentiation and proprietary requirements — not about technical sophistication or cost. If the existing vendor solution can hit the KPI target, custom RL adds cost, complexity, and maintenance burden with no strategic benefit.
Q5. A churn prediction model runs in shadow mode for 8 weeks. It flags 340 customers as high-risk; 280 of those customers do churn within 30 days. What has the pilot demonstrated?
BThe model achieved ~82% precision on real production data with zero operational risk — a strong signal to proceed to deployment.
280/340 = 82.4% precision. This is a strong real-world performance result. The 60 false positives represent customers who would have received a retention outreach they didn't need — a minor cost compared to the value of correctly identifying 280 at-risk customers. Shadow mode succeeded in its purpose: measuring performance before any operational risk.
Common wrong answer — A: Students who focus on the 60 false positives may choose A. Clarify: 82% precision is strong for a churn model. False positives in a retention context (unnecessary outreach) have low harm — they're categorically different from false positives in a fraud or medical context (blocking a legitimate customer or missing a diagnosis).
💬 Discussion Prompts — Lesson 2.3
The 80% That Isn't the Model
An AI team spends 90% of their time building a highly accurate fraud detection model. They spend almost no time on the data pipeline, monitoring system, or workflow integration. What are the most likely failure modes when this model is deployed to production?
Expected failure modes: (1) training-serving skew because the production data pipeline doesn't match training; (2) model drift because there's no monitoring to detect when patterns change; (3) adoption failure because analysts were never involved and have no workflow for acting on alerts; (4) no human override mechanism for edge cases. This discussion reinforces the six-component system view.
Technique Matching Challenge
For each of these business problems, identify the most appropriate AI technique and explain why: (a) A bank wants to group its 2 million customers by behavior patterns to personalize offers. (b) A manufacturer wants to predict which machines will fail in the next 72 hours. (c) A law firm wants to automatically extract key clauses from contracts. (d) A retailer wants to create personalized email subject lines for marketing campaigns.
Expected: (a) Unsupervised — clustering with no predefined categories. (b) Supervised learning — regression/classification with historical failure records as labels. (c) NLP — entity extraction and information retrieval from text. (d) Generative AI — LLM producing text content at scale. Use this to reinforce that technique choice follows problem type, not available technology.
The Reward Function Problem
A social media company uses reinforcement learning to maximize "daily engagement time" on its platform. The algorithm discovers that outrage-inducing and fear-based content drives more engagement than neutral content. Is this an AI failure? Who is responsible?
This is a design failure — the reward function was misspecified. The algorithm did exactly what it was designed to do. Responsibility lies with the designers who chose "engagement time" as the reward without considering second-order effects. This connects to the RL warning in Lesson 2.3 and the non-maleficence concept from Unit 2's ethics content.
⚠️ Common Misconceptions — Lesson 2.3
Misconception
"The most complex AI technique gives the best results."
Correct Understanding
Simplicity is a virtue in AI design. The simplest model that reliably solves the problem is usually the best choice — it's more interpretable, easier to maintain, faster to deploy, and less prone to overfitting. Deep learning isn't always better than gradient boosting.
Misconception
"Once a model is deployed, it will maintain its performance indefinitely."
Correct Understanding
Model drift is inevitable — the world changes faster than training data. Every deployed model requires ongoing monitoring with defined retraining triggers. This must be designed before deployment, not retrofitted when performance degrades.
Misconception
"You should always build your own AI rather than buying — custom models are always better."
Correct Understanding
Building custom AI only makes sense when the use case has proprietary requirements a vendor can't meet. For commodity tasks (translation, OCR, common NLP), buying via cloud API is faster, cheaper, and often produces better results than custom development.
Misconception
"A shadow mode pilot that shows good accuracy means the AI is ready for full deployment."
Correct Understanding
Accuracy is one go/no-go criterion. Full deployment also requires: validated workflow integration, trained end users, confirmed monitoring infrastructure, human override mechanism, and stakeholder sign-off. Shadow mode tests model performance — not the full system.
Lesson 2.4 — Strategic AI Integration Project (Capstone)
🎯 Project Purpose: Students integrate all four Unit 2 lessons into a single professional deliverable. The project mirrors the work produced by AI strategy consultants — a readiness assessment, strategic alignment plan, and solution design for a real organization scenario. Students self-assess against rubrics and submit a PDF for instructor grading.
✅ What Students Submit: A PDF downloaded from the project workspace containing all 13 response fields across three parts, self-assessed scores by part, and a submission timestamp. Students upload this PDF to the LearnDash assignment block. The self-assessed score is a starting point for instructor review — not the final grade.
📊 Scenario Guide — Helping Students Choose
Retail
FreshMart Grocery Chain
Best for: Students without specialized industry background
Most accessible scenario — grocery/retail is universally relatable
Clear data assets (4 yrs POS, 310K loyalty members)
Obvious AI opportunity: 12% spoilage rate → demand forecasting
Critical blocker: legacy ERP, no data warehouse
Strongest Part 2 KPI opportunity: Spoilage rate % with clear dollar value quantification
Healthcare
Meridian Health Network
Best for: Students with clinical or healthcare operations background
Two distinct AI opportunities (readmission + scheduling)
Richer for regulatory constraint discussions (HIPAA, explainability)
Critical blocker: 3 different EHR systems — data consolidation challenge
Leadership tension (CMO cautious, CFO supportive) adds realism
Strongest Part 3 design opportunity: Shadow mode pilot for readmission prediction with clinical workflow integration
Finance
Crestline Credit Union
Best for: Students with finance, banking, or business background
Richest data scenario (12 years well-labeled loan data)
Two KPI tension opportunities (default rate + member churn)
Modern tech stack — less readiness remediation needed in Part 1
CEO wants quick ROI — good for discussing pilot strategy tradeoffs
Strongest Part 2 alignment: Loan default rate has direct dollar quantification and a clear industry benchmark (4.2% vs. 2.8% average)
📊 Official Grading Rubric — Part 1: AI Readiness Assessment (14 pts)
| Criterion | Exemplary (4) | Proficient (3) | Developing (2) | Beginning (1) | Pts |
|---|---|---|---|---|---|
| Pillar Specificity | Every pillar assessment cites at least one specific scenario detail (e.g. "the legacy ERP means…"); ratings directly flow from evidence cited | Most pillars cite scenario details; one or two are more generic but still grounded | Some scenario references but several pillars use generic AI-readiness language that could apply to any organization | Little or no scenario-specific evidence; pillar descriptions are mostly generic | 4 |
| Numerical Ratings | All five pillars rated 1–5 with clear, specific justification for each score; ratings are internally consistent | All pillars rated with justification; one or two justifications are thin but present | Most pillars rated; justifications are brief or circular ("it's a 2 because it's low") | Missing ratings or ratings without any justification | 3 |
| Critical Blocker | Correctly identifies the weakest pillar; explicitly explains why this blocker constrains success even when other pillars are strong; uses "lowest pillar" logic | Correct blocker identified with reasonable explanation; could be more explicit about the constraining mechanism | Identifies a weak pillar but doesn't fully explain why it blocks progress; may list multiple "critical" blockers without prioritizing | No clear critical blocker identified, or incorrect pillar named without supporting logic | 4 |
| Phase Recommendation | Correct phase recommended (Foundation, Exploration, Operation, or Transformation); recommendation explicitly tied to the identified blocker and overall readiness score | Correct phase with a reasonable justification; connection to blocker present but could be stronger | Phase recommended with weak justification; may be one phase off without clear reasoning | No phase recommended, or recommendation is inconsistent with the blocker identified | 3 |
📊 Official Grading Rubric — Part 2: Strategic Alignment Plan (13 pts)
| Criterion | Exemplary | Proficient | Developing | Beginning | Pts |
|---|---|---|---|---|---|
| Goal Selection | Correct strategic goal category chosen and explicitly tied to the highest-impact, most quantifiable pain point in the scenario with clear reasoning | Correct goal category; connection to scenario pain point is present but could be more precisely quantified | Goal category is plausible but not clearly the best fit, or connection to scenario is vague | Goal category is misidentified or chosen without connection to scenario evidence | 3 |
| KPI Quality | KPI is a direct business measure (not a technical proxy); includes a specific numeric baseline, target, and success threshold; explains why this KPI cannot be improved by accident | Strong business KPI with numeric baseline and target; success threshold may be implied rather than explicit | KPI is a business measure but lacks a numeric baseline or target, or is a proxy metric that could be a vanity trap | KPI is a technical or vanity metric (accuracy, volume, clicks) without business value connection | 4 |
| Alignment Test | All three alignment questions answered with specific details: named metric, named owner, and a concrete 90-day signal with threshold | All three questions addressed; one answer may be less specific (e.g. role named but not individual, or 90-day signal described without a threshold) | Two of three questions answered substantively; third is vague or missing | One or fewer alignment questions answered; responses are generic rather than scenario-specific | 4 |
| Spectrum Honesty | Correctly identifies alignment stage with reference to which of the four elements are present/absent; accurately diagnoses the specific remaining gap | Correct stage identified; gap diagnosis is present but could be more specific | Stage is one level off, or the gap is identified without explanation | Stage is misidentified or no gap analysis provided | 2 |
📊 Official Grading Rubric — Part 3: AI Solution Design (13 pts)
| Criterion | Exemplary | Proficient | Developing | Beginning | Pts |
|---|---|---|---|---|---|
| Problem Definition | All four elements present and specific: input data named, output format defined, success criteria are measurable, human workflow described step-by-step | Three of four elements present; one is less specific but the definition is still actionable | Two elements present; definition is partially actionable but would require significant clarification before build | Problem definition is vague ("the AI will improve X") with no technical specificity | 3 |
| Technique Justification | Technique correctly matches the problem type with explicit reasoning (e.g. "supervised because labeled historical data exists"); build vs. buy recommendation justified with reference to proprietary vs. commodity logic | Technique is correct with reasonable justification; build vs. buy addressed but reasoning could be sharper | Technique is plausible but justification is weak or circular; build vs. buy may be missing | Technique selection appears arbitrary or doesn't match the problem type | 3 |
| System Completeness | At least four of six components described with specifics (not just named); demonstrates that the model is one component, not the whole system | Three components described with reasonable specificity; fourth is brief but present | Two components described; others are missing or only named without description | Only the model described; other components absent or not understood as separate concerns | 3 |
| Pilot Rigor | Scope is narrowly defined; duration stated; shadow mode rationale explained; all go/no-go criteria are specific and numeric, defined before results | Scope and duration defined; go/no-go criteria present but one may lack a numeric threshold | Pilot described at high level; go/no-go criteria are vague ("if it works well") rather than measurable | No distinct pilot strategy; jumps directly to full deployment | 2 |
| Risk Specificity | Three risks are specific to this use case and scenario (not generic); each mitigation is a concrete design decision, not a wish ("we will monitor it") | Two or three use-case-specific risks with actionable mitigations; one mitigation may be generic | Risks are named but generic (could apply to any AI project); mitigations are vague | Fewer than two risks identified, or risks and mitigations are completely generic | 2 |
🏅 Overall Score Interpretation
| Score Range | Performance Level | Instructor Action |
|---|---|---|
| 36–40 pts (90–100%) | Outstanding — Distinction | Award full or near-full marks. Consider highlighting as exemplar for cohort. |
| 30–35 pts (75–89%) | Strong — Merit | Award merit marks. Provide brief written feedback on 1–2 areas to sharpen. |
| 24–29 pts (60–74%) | Satisfactory — Pass | Pass with feedback on the weakest part. Optional revision opportunity recommended. |
| Below 24 pts (<60%) | Developing — Needs Revision | Require revision of the lowest-scoring part before moving to Unit 3. Schedule a check-in. |
⚠️ Self-Assessment Calibration Note: Students consistently under-rate their Part 1 responses (too self-critical) and occasionally over-rate Part 2 (unclear on the KPI baseline requirement). When grading, pay particular attention to whether Part 2 KPIs include a numeric baseline — this is the most commonly missing element even in otherwise strong submissions.
Differentiation Strategies
Unit 2 serves students with diverse backgrounds — some technical, some business-oriented, some with no prior AI exposure. These strategies help all three groups engage at an appropriate depth without changing the core learning objectives.
🌱 Scaffolded Support
Before Lesson 2.1, assign a 10-minute "AI in the news" exercise: find one recent story about an AI project that failed or succeeded and bring it to class
Provide a printed five-pillar checklist for students to annotate while reading the scenario before the Calculator activity
For the capstone, allow students to use the FreshMart scenario (most accessible) and pair with a stronger student for the solution design section
In Lesson 2.3, have struggling students focus on the Supervised Learning and NLP technique cards only before attempting the full matcher
Offer a "think aloud" model: walk through one complete pillar assessment together before students attempt the others independently
📘 Core Pathway
All four lessons in sequence with full interactive engagement — Calculator, KPI Checker, Scenario Mapper, Solution Builder, and all chatbots
Use all three discussion prompts per lesson; written reflection required for at least one per lesson
Capstone project completed independently with full rubric self-assessment across all three parts
Lesson 2.3: complete all six technique cards before attempting the Solution Builder
Post-unit: review any quiz questions scored incorrectly and write a one-sentence correction for each wrong answer
🚀 Extension Challenges
After Lesson 2.1: research a real company's AI initiative and rate its readiness across all five pillars using only publicly available information (annual report, press releases, LinkedIn)
After Lesson 2.2: find a published AI case study and apply the three-question alignment test. Present findings to the class: was the initiative fully aligned?
For the capstone: complete a second scenario in addition to the primary and compare how the readiness assessment and solution design differ across industries
Lesson 2.3 extension: design the full six-component system architecture for a use case not covered by the Solution Builder (e.g. student dropout prediction for a university)
Post-unit challenge: write a 1-page "AI Investment Recommendation" memo as if presenting to a board — using the frameworks from all four Unit 2 lessons
Pacing Guide
Unit 2 is designed for flexible delivery — self-paced individual study, instructor-led cohorts, or hybrid formats. The guide below covers all three delivery modes.
| Lesson | Self-Paced (Online) | Instructor-Led (Cohort) | Hybrid (Flipped) | Key Checkpoint |
|---|---|---|---|---|
| 2.1 — Readiness | Week 1, Day 1–2 60–75 min async |
Session 1 (75 min) 30 min lecture + 45 min activity |
Pre-read async → 45 min discussion in class | Quiz 2.1 ≥ 3/5 |
| 2.2 — Alignment | Week 1, Day 3–4 60–70 min async |
Session 2 (70 min) 25 min KPI workshop + 45 min scenarios |
Pre-read async → 45 min KPI design workshop in class | Quiz 2.2 ≥ 3/5 |
| 2.3 — Design | Week 2, Day 1–2 70–85 min async |
Session 3 (85 min) 30 min technique matching + 55 min solution builder |
Pre-read async → 60 min design workshop in class | Quiz 2.3 ≥ 3/5 |
| 2.4 — Project | Week 2, Day 3–5 60–90 min, one sitting recommended |
Session 4 (90 min) In-class project work with instructor support |
In-class project session with instructor check-ins at each part transition | PDF submitted ≥ 24/40 |
📌 Instructor-Led Tip: For Session 3 (Lesson 2.3), the most effective classroom format is "technique speed-rounds" — give students a business problem description on a card and have them call out the technique with a one-sentence justification before revealing the answer. This builds the rapid pattern-matching that the quiz and exam require.
Instructor FAQ
The self-assessment is explicitly labeled in the student workspace and PDF as "pending instructor review." Your grade overrides the student's self-assessment entirely. When a student's self-assessment is significantly higher than warranted, use the rubric language to give specific, criterion-level feedback — e.g., "Your KPI definition (Part 2b) is missing a numeric baseline, which is required for Exemplary. This brings your Part 2 score from your self-assessed 13 to an instructor grade of 9." Naming the specific criterion makes the feedback actionable and reduces disputes.
Yes — the workspace uses browser session storage, which clears when the tab is closed or the browser is cleared. Students who close their browser before downloading the PDF will lose all progress. The recommended mitigation in future sessions: instruct students to click "Save Progress" every 15 minutes and to complete the project in a single uninterrupted session. For students who lose progress, the most equitable response is to offer a one-time extension without penalty — the loss is a UX issue, not a student failure. Consider adding a note to your LearnDash assignment instructions about saving frequently.
This is an instructor policy decision — the curriculum doesn't prohibit AI tool use. However, the project is specifically designed to test applied judgment, not knowledge retrieval. A student who prompts an LLM to complete the responses will produce generic answers that score poorly on the "specificity" criteria — the rubrics reward scenario-specific evidence and reasoning that requires actually reading and analyzing the provided organization context. If you want to permit AI tool use, consider requiring students to annotate their AI-assisted responses with a brief explanation of why they accepted or modified the output. This adds metacognitive value and reveals whether genuine learning occurred.
Use the Technique Justification rubric criterion. If the student chose a plausible technique with clear reasoning, award Proficient (3/3) even if it's not the canonical answer — AI technique selection has legitimate alternatives in most cases. The rubric rewards reasoning, not right-or-wrong answers. If the technique is genuinely inappropriate for the problem type (e.g. using reinforcement learning for a simple text classification task), award Developing or Beginning depending on whether any justification was provided. The most common error is choosing supervised learning for an unsupervised problem (no labels exist) — flag this explicitly in feedback.
Unit 2 is designed as a direct continuation of Unit 1 — Lesson 2.3's technique matching assumes familiarity with supervised vs. unsupervised learning (Lesson 1.2) and the EU AI Act explainability requirements (Lesson 1.4). Students who haven't completed Unit 1 will find Lesson 2.3 significantly harder and may misapply the technique matcher. If you're onboarding new students directly into Unit 2, assign at minimum Lessons 1.2 and 1.4 as prerequisites, or provide the key vocabulary list from those lessons as a reference document during Unit 2 work.