Phase 1: KNOW

Contextual Intelligence

Stop looking at tools and start looking at the problem. Understand AI capabilities, limitations, and security requirements for your specific workflow context.

The Goal

Before mentioning "Generative AI," map the specific workflow step-by-step. Where is the bottleneck? Is it ideation, summarization, or coding?

Workflow Contextualization

Map the Workflow

Break down your workflow into discrete steps. Identify where friction occurs, where time is wasted, and where quality suffers.

  • Document each step in the current process
  • Identify bottlenecks and pain points
  • Measure time spent on each step

Identify the Bottleneck

Determine the specific type of AI assistance needed. Different bottlenecks require different AI capabilities.

Ideation
Creative generation, brainstorming
Summarization
Condensing information, extracting insights
Coding
Code generation, debugging, refactoring

Data Classification: The Critical Gate

This is the most important decision point. The sensitivity of your data determines which AI tools are viable options. Get this wrong, and you risk security breaches or compliance violations.

Public Data

Safe for open models. Information that is already publicly available or can be shared without security concerns.

Marketing copy
Public research
General knowledge
Tool Options:
Open models (ChatGPT, Claude, etc.)

Internal/Confidential

Requires enterprise instances with "zero-retention" policies. Data that must remain within your organization's control.

Internal documents
Proprietary processes
Customer data (non-PII)
Tool Options:
Enterprise SaaS with zero-retention

Strictly Private/IP

May require on-premise or private cloud small language models (SLMs). Highest security requirements.

PII (Personally Identifiable Information)
Trade secrets
Regulated data (HIPAA, GDPR)
Tool Options:
On-premise or private cloud SLMs

Constraint Mapping

Latency Requirements

How quickly does the user need a response? This determines whether you need real-time processing or can use batch processing.

Instant Response
Chatbots, real-time assistance
Requires: Low-latency models, edge deployment
Overnight Processing
Batch analysis, reports
Allows: Higher-latency, more powerful models

Accuracy Requirements

Can you tolerate hallucinations, or do you need factual grounding? This determines whether you need RAG (Retrieval-Augmented Generation) or can use pure generative models.

Creative Tolerance
Creative writing, brainstorming
Allows: Pure generative models
Factual Grounding Required
Financial reporting, legal documents
Requires: RAG, grounding capabilities

Key Questions to Answer

What data sensitivity level?
Public, Internal, or Private/IP?
What output quality needed?
Creative or factual accuracy?
What time constraints exist?
Real-time or batch processing?