This is where the hype train derails projects and burns budgets. The client had confused a complex retrieval system with a simple data collection script. Here's the difference, explained without the jargon:

The Wizard (What They Actually Needed)

Think TurboTax or a structured intake form. It's linear. It's predictable. The bot controls the conversation flow.

Example interaction:

• Bot: "What's your brand promise?"

• User: "We help small businesses automate without complexity."

• Bot: "Perfect. Now, what's your core offer?"

Tech requirements: Minimal. You could build this in Typeform, ManyChat, or a basic n8n workflow in an afternoon. No AI models required beyond basic natural language understanding.

Cost: Hundreds of dollars. Maybe a thousand if you want it pretty.

The RAG (What They Asked For)

Think reference librarian standing next to a file cabinet containing your company's entire knowledge base. The user drives the conversation with unpredictable queries.

Example interaction:

• User: "What's your refund policy for Tier 2 products purchased on Tuesdays during promotional periods?"

• Bot: (Searches vector database, finds specific paragraph in 50-page PDF, cross-references three policy documents, synthesizes answer) "According to section 4.B of our refund policy, purchases made during..."

Tech requirements: Vector databases (Pinecone, Supabase pgvector), embedding models, semantic search infrastructure, complex orchestration layers, ongoing maintenance.

Cost: Thousands to tens of thousands of dollars, plus monthly hosting and API costs.