# The Three-Pillar Framework

Paul Cheek's framework for developing an AI-Driven Enterprise in any sector. Every strategy in the playbook is expressed as a combination across the three pillars: WHERE the AI lives, WHAT it mechanically does, and HOW it gets implemented. Teach each pillar before questioning within it; the executive should leave able to draw all three from memory.

## Pillar 1: Application Domains (The "Where")

Where does the AI live within the organizational chart? It usually falls into one of four buckets.

### 1. Back-Office / Administrative
Streamlining internal operations that don't directly touch the product or the customer.
Use cases: resume screening in HR, anomaly detection in Finance (fraud prevention), contract analysis in Legal.

### 2. Front-Office / Customer Experience
Enhancing how the organization interacts with the market.
Use cases: hyper-personalized email marketing, 24/7 customer service chatbots, dynamic pricing models in retail.

### 3. Core Operations & Supply Chain
Optimizing the actual production or movement of goods and services.
Use cases: predictive maintenance on factory floors (predicting when a machine will break before it does), inventory forecasting, route optimization for logistics fleets.

### 4. Product / Service Enhancement
Embedding AI into what the company actually sells.
Use cases: a tractor manufacturer adding computer vision to its machinery to spray only weeds and not crops, or a bank offering an AI-driven personal finance assistant in its app.

Teaching note: most organizations start in back-office because it is safe, but the durable competitive advantage almost always lives in core operations and product enhancement. Surface that tension explicitly when the executive ranks the domains.

## Pillar 2: AI Capabilities (The "What")

This pillar defines the actual mechanical role the AI is playing. What is the tool actually doing?

### 1. Automation (Doing)
Handling repetitive, rules-based tasks without human intervention (often combining AI with Robotic Process Automation, but not in cases where RPA alone is sufficient).
Example: automatically extracting invoice data from PDFs and inputting it into the accounting system.

### 2. Augmentation (Assisting)
Acting as a co-pilot to make a human worker faster or better.
Example: summarizing a 50-page regulatory document into a one-page brief for a compliance officer.

### 3. Analytics & Predictive (Foreseeing)
Analyzing massive datasets to find hidden patterns and predict future outcomes.
Example: predicting next quarter's demand for a specific SKU based on historical sales, weather patterns, and economic indicators.

### 4. Generative (Creating)
Producing net-new text, code, audio, or imagery.
Example: generating initial drafts of marketing copy or creating variations of product packaging designs.

Teaching note: the RPA boundary from Pillar 2's Automation bucket matters. If a task takes identical steps every run with zero judgment, plain RPA or deterministic code is the cheaper answer and it does not belong in the AI strategy. This is the same discipline the AIDE Opportunities Audit enforces.

## Pillar 3: Implementation Strategy (The "How")

Organizations do not need to build AI from scratch to implement it. This spectrum ranges from low-barrier/low-cost to high-barrier/high-customization.

### 1. Ready-to-Use SaaS (Off-the-Shelf)
Adopting standalone AI tools that require little to no technical integration.
Examples: ChatGPT Enterprise for drafting communications, Midjourney for marketing asset ideation, Otter.ai for meeting transcriptions.

### 2. Embedded AI (Vendor-Supplied)
Utilizing the AI features newly built into the software the organization already uses.
Examples: Microsoft 365 Copilot, Salesforce Einstein, SAP's embedded AI for ERP. The barrier to entry is low, but the organization is reliant on the vendor's roadmap.

### 3. Configured / Fine-Tuned AI
Taking existing foundational models and training or pointing them at the organization's proprietary data (often using Retrieval-Augmented Generation, or RAG).
Examples: an internal chatbot trained strictly on a company's HR policy documents, or an IT helpdesk bot trained on past resolution tickets.

### 4. Custom / Ground-Up AI
Building proprietary machine learning models for a highly specialized, core business function.
Examples: a mining company building a unique computer vision model to detect specific ore variations on a conveyor belt, or a pharmaceutical company building a model to predict molecular binding.

Teaching note: the right posture is a portfolio, not a single point on the spectrum. Commodity tasks belong at the low-barrier end; anything that touches the company's proprietary advantage deserves consideration further right. Vendor dependence is a strategic variable, not a footnote.

## Combining the pillars

A strategy line item is a triple: Domain x Capability x Implementation. Example: "Core Operations (Where) + Predictive (What) + Configured/Fine-Tuned (How) = demand forecasting tuned on our own sales history." The Strategic Outlook document renders the executive's choices as this matrix so the board can see coverage and gaps at a glance: domains with no capability assigned, capabilities concentrated in one domain, or everything piled into off-the-shelf SaaS.