Matrix Discovery: Extracting the AI Guardrails

In an agent-first world, SMEs don't separate their current pain from their future desires.

When you ask business leaders how they handle a midterm amendment today (the As-Is), they will immediately complain about manual credit memos and jump straight into how they want the AI to automate it tomorrow (the To-Be). The idea that Discovery is a neat, chronological march from "As-Is" to "To-Be" is a theoretical fallacy.

You are no longer observing a "day in the life" to map a human's screen clicks. You are observing the SMEs to extract the invisible, deterministic business rules they use to make decisions—so you can hardcode those rules into the Commercial Matrix. Here is how to embrace the chaos of a workshop and organize Discovery so it actually builds your agentic guardrails.

1. Organize by "Intent Cluster", Not by Phase

Stop scheduling generic "As-Is Workshops." Instead, organize your Discovery sessions by the Intent Cluster or Revenue Thread. If the workshop is focused on the "Renewal Lifecycle," you let the conversation flow naturally through the past, present, and future of just that specific conversational boundary.

2. The In-the-Room "Micro-Cycle"

When you are facilitating, guide the SMEs through a continuous loop for every specific capability they bring up. You are translating chaos into architecture in a 10-minute window:

The Human Friction (As-Is): "Reps keep discounting below our margin floor, so I have to review every deal manually."
The AI Intent (To-Be): "The agent should automatically stop bad discounts before they are ever proposed to the customer."
The Matrix Constraint (Pre-Gap Steering): "We will build a deterministic rule in the Commercial Matrix: If Margin < 20%, the Agent is blocked from generating the Quote payload. Are you open to standardizing this floor across all product lines?"

3. Matrix-Guided Design

When business leaders ask, "What can the AI do?", they are inviting you to shape their To-Be state to fit standard Salesforce Agentforce and RLM architecture. You aren't just taking orders; you are advising them on best practices for headless execution.

If their To-Be aligns with standard AI guardrails: You confirm it immediately, mapping the rule directly to Data Cloud or a Revenue Cloud Pricing Procedure.
If their To-Be requires dangerous LLM reasoning: You challenge them. "We can let the AI try to guess the proration, but it's an extreme hallucination risk. If we standardize your service periods into a hard matrix rule, we guarantee 100% backend predictability when the payload hits the financial ledger."

4. The Unified Matrix Discovery Document

Because the conversation happens all at once, your documentation must capture it all at once. For every row, you track the friction, the intent, and the exact mathematical guardrail the AI needs to function.

The Agent-First Matrix Discovery Template

Intent Cluster	Business Intent (The Goal)	Required Matrix Guardrail (Agent Constraint)	Architectural Gap / Data Risk
Mid-Term Add-On (Amendments)	System should automatically calculate the exact prorated cost for new seats based on the days remaining.	Grounding Rule: The Agent must query the existing Contract End Date and apply the 'Standard Proration Rule' via Data Cloud before finalizing the JSON payload.	High Data Risk: If legacy contract data lacks accurate end dates, the Agent will hallucinate the proration. Data cleanup is a prerequisite.
Margin Floor Enforcement	Automate a hard stop on any hardware discounts exceeding 15%.	Execution Rule: The Agent's Prompt Template restricts negotiation. It must route through the RLM Pricing Procedure to validate the Net Price.	Process Gap: Business must agree to standardize the 15% rule across the entire hardware catalog without manual overrides.
Hardware + SaaS Bundling	Reps need to sell the physical sensor and a 12-month software subscription in the same chat prompt.	Orchestration Rule: The Agent must select the 'Hybrid Quote Archetype' to ensure the OmniStudio IP constructs the payload perfectly for downstream SAP FICO split revenue recognition.	Medium Gap: The Agent can capture the intent easily, but the downstream billing schedules must be pre-mapped to accept mixed Product Selling Models (One-Time vs Termed).

Extend L1-L3: Process-to-Intent Mapping Hierarchy

Compiling the Intent Hierarchy before Discovery is exactly how you lock down the project scope. If a specific intent is on the L3 list, the agent will be trained on it; if it isn't, it is out of bounds.

Hierarchy Level	Agile Issue Type	Agent-First Definition & Example
L1: Intent Domain	Initiative	The comprehensive conversational boundary. Example: Revenue Lifecycle (Quote-to-Cash).
L2: Intent Clusters	Component/Theme	Grouped user objectives mapped to Commercial Archetypes. Example: Subscription Modifications.
L3: Discrete Intents (Agent Actions)	Epic	The precise objective extracted from language to execute an API. Example: Co-term a mid-year SaaS add-on.
L4: Payload Blueprint	User Story	The narrative of data execution. Example: "As the AI Agent, I need to extract the target quantity from the user and format the Amendment Archetype payload so the RLM engine accepts the order."
L5: Guardrail Verification	Acceptance Criteria	The rules that prove the Agent did not hallucinate. Example: "Agent successfully retrieves Regional Discount Limit from the Matrix before proposing a price."

The Architect's "Cheat Sheet": Writing Agent-First Acceptance Criteria

Agile boards are just a mechanism for tracking work. The magic happens when you shape those tickets to force the development team to build headless architecture correctly. Business Analysts (BAs) must write Acceptance Criteria (L5) that actively prevent developers from slipping back into UI-heavy, rigid screen builds. Enforce these three rules:

1. The "No Hallucination" Rule (Ground the AI)
The AC must mandate that the Agent queries the Commercial Matrix before taking action.
- Bad AC: "The system applies a 10% discount for volume."
- Agent-Ready AC: "The Agent successfully retrieves the Volume Discount Tier matrix from Data Cloud and applies the exact percentage before generating a response."

2. The "Headless Payload" Rule (Remove the UI)
The AC must prove the Agent can construct the data via intent, without a user clicking a screen.
- Bad AC: "The 'Hardware Type' dropdown is visible on the quote page."
- Agent-Ready AC: "The LLM successfully extracts the hardware type from natural language and maps it correctly into the 'Hardware Archetype' JSON payload."

3. The "Downstream Proof" Rule (The Ledger Check)
Every payload story must prove it survives the backend integration without manual intervention.
- Bad AC: "The user can generate an upgrade quote."
- Agent-Ready AC: "The Agent's completed payload successfully traverses the OmniStudio integration, updates the Asset, and registers the deferred revenue schedule properly in the downstream financial ledger."