Applied Artificial Intelligence in the MBA.
Annex A specifies how Applied AI is to be taught within any program aligned to the MBA Council Framework. It governs orientation, module composition, assessment, and the required governance overlay. It does not prescribe vendors, models, or specific tools.
Six principles governing every unit.
AI as an Operator's Tool, Not a Discipline
MBA programs must treat Artificial Intelligence as an instrument the business operator wields — comparable to spreadsheets in 1990 or the web in 2005. The framework rejects the treatment of AI as a computer-science subject inside a management syllabus. The learner is not being trained to build a model; the learner is being trained to deploy, govern, and monetize one.
Prompt Architecture Before Prompt Engineering
Prompt Engineering — the writing of clever prompts — is a low ceiling skill. Prompt Architecture — the structural design of multi-step, tool-augmented, verifiable LLM workflows — is the mandated competency. Coursework must move learners from single-turn prompting to agentic workflow design within the first module.
Decision Augmentation, Not Automation
The pedagogical goal is decision quality, not job elimination. Every applied unit must produce a decision — a hiring call, a pricing move, a capital allocation — where the AI's recommendation is explicitly reconciled with the human operator's judgment. Cases must terminate in a defended decision, not a produced artifact.
Data Literacy as Prerequisite
Applied AI presumes data literacy. Adopting institutes must place a bounded quantitative reasoning unit — descriptive statistics, distributions, base rates, and Bayesian intuition — before any generative or predictive AI unit. Learners who cannot read a distribution cannot audit an LLM's answer.
Governance, Bias & Provenance
Every Applied AI unit must carry an integrated governance thread: model provenance, training-data disclosure, bias audit, and human-in-the-loop escalation. Governance is not an ethics elective at the end of the program; it is embedded inside every applied deliverable.
Vendor-Neutral Instruction
Coursework must be portable across at least two independent model providers and at least one open-weights model. A syllabus that trains a learner to operate a single vendor's product is not a management education — it is a certification course, and it is out of scope of the framework.
Five modules, 102 contact-equivalent hours.
- Tokens, context windows, and cost economics
- Prompting patterns: role, format, retrieval, reflection
- Structured output contracts (JSON, tool schemas)
- Evaluation harnesses: correctness, groundedness, latency, cost
- Descriptive vs. predictive vs. prescriptive framing
- Feature intuition without derivation of algorithms
- Reading and challenging a data scientist's model card
- Cost-of-error framing: false positives vs. false negatives at the P&L
- Deterministic pipelines vs. agentic loops — when to pick which
- Tool-use, memory, and retrieval-augmented generation
- Human-in-the-loop checkpoints and rollback strategy
- Deployment envelope: sandbox, staging, production, audit
- Data provenance, IP exposure, and confidentiality boundaries
- Bias audit: measurement, mitigation, and disclosure
- Jurisdictional regulation: EU AI Act, sectoral frameworks
- Incident response and post-mortem discipline
- Buy-vs-build-vs-fine-tune under uncertainty
- Unit economics of inference: gross margin, latency, retention
- Organizational adoption: change management, upskilling, ROI
- Terminal case: an AI-native P&L defended in viva
Defended workflows, not multiple-choice.
The Applied AI annex may not be assessed through multiple-choice examinations. Learners must submit a functioning workflow, a governance memo, and a defended decision case at the viva. Terminal examinations for this annex take the form of a live design problem in which the learner sketches, defends, and revises an AI-augmented decision system in front of an examiner.