SMU ACADEMY — GENAI FOR BUSINESS GROWTH & INNOVATION · UNIT 2

The Executive's AI Primer

What the hierarchy actually means, how machines really learn, and where GenAI earns its budget — in three takeaways.

01

Cutting Through the AI Noise: The True Hierarchy

Artificial Intelligence

The broad umbrella: computational systems designed to simulate human intelligence, reasoning, and decision logic.

Machine Learning

Pattern recognition: statistical algorithms that learn patterns and rules directly from data without explicit procedural programming.

Deep Learning

Multi-layer neural nets: hierarchical representations that extract high-level features from complex, unstructured data.

Generative AI

Content synthesis: specialized models trained to generate net-new text, images, code, and synthetic media from input prompts.

Level 1: Basic Automation

Fixed rules & sorting

Deterministic logic, hardcoded formulas, and if-then sorting workflows. Essential foundational software, but not AI.

Level 2: Analytical ML

Price prediction & classification

Statistical pattern recognition, regression models, and classification. Learns from historical data to forecast known targets.

Level 3: Generative AI

Original text & media synthesis

Neural synthesis of unstructured text, code, audio, and imagery. Premium cognitive capability for creative and generative tasks.

Executive Decision Rule

If a tool merely calculates numbers or applies formulas, it is basic software; if it generates original content, it is GenAI. Don't pay premium AI SaaS prices for spreadsheet formulas.

02

Three Ways Machines Learn — And Where Business Fails

Supervised Learning

Historical car sales → Selling price prediction

Trained on mapped input-output pairs. Evaluates known ground-truth history to accurately predict defined numeric or categorical targets.

Unsupervised Learning

Raw buyer transactions → Behavioral personas

Discovers natural clusters, hidden groupings, and anomalies across unlabelled datasets without human guidance or pre-set targets.

Reinforcement Learning

Dynamic bidding & navigation

Direct policy optimization via continuous trial and error, guided by reward and penalty signals in dynamic, interactive environments.

The Bottleneck Barometer
20% Clean Structured DB

Standardized ERP records, normalized SQL tables, validated financial ledgers.

80% Messy Unstructured Operations

WhatsApp voice notes & customer chats, unformatted PDF log cards, handwritten dealer notes, vehicle inspection photos.

Operational reality: The 80/20 split is an illustrative industry rule of thumb, not a laboratory measurement. In production, the overwhelming bulk of enterprise friction lies in extracting, parsing, and cleaning unstructured field data—not in running the model.

Executive Decision Rule

AI models are only as good as the data feeding them. The vast majority of actual operational effort is data extraction and cleaning, not model tuning.

03

Strategic Value: Where AI Solves Real Bottlenecks

Technology: Computer Vision

Inspection & Dynamic Layout Adaptability

Automated vehicle damage appraisal Bounding-box detection frame Verified, structured tabular records Layout coordinate inspection overlay Translates unstructured physical surfaces Visual artifacts → tabular records Non-breaking scraping pipelines

Concrete Business Anchor: Automated vehicle damage appraisal and adapting to fluctuating website or document layouts without breaking scraping pipelines.

Value Delivered: Translates unstructured physical surfaces and visual artifacts into verified, structured tabular records.

Technology: Transformers & LLMs

Contextual Language & Non-Linear Attention

Sequential reading chain (word-by-word) dealer vehicle descriptions cryptic logs unstructured queries Simultaneous multi-directional self-attention web dealer descriptions cryptic logs unstructured queries Clean database fields in a single pass

Concrete Business Anchor: Parsing messy, informal dealer vehicle descriptions, cryptic mechanic logs, and unstructured buyer queries into clean database fields in a single pass.

Value Delivered: Resolves cross-sentence context, colloquial shorthand, and domain jargon where rigid keyword filters and brittle regex fail.

Executive Decision Matrix

Apply a strict ROI test. Deploy GenAI where human communication is messy and unstructured, but retain simple deterministic rules and standard software for straightforward bookkeeping and inventory.

SMU GenAI · Unit 2 Executive Primer