MEDDPICC Qualification Research

Each MEDDPICC element maps to a specific Salesforce field on the Opportunity object:

During Setup, read existing values for all 8 fields. During Output, propose write-back for empty or improved fields. See references/queries.md for SOQL queries, DML patterns, and picklist values.

Workflow Overview

1. Setup: Resolve account in SF (including MEDDPICC fields), read org-context, create temp directory
2. Wave 1: Read existing artifacts (SF data, Obsidian account files, transcripts, Gmail)
3. Analysis: Synthesize findings per element, assess confidence, identify gaps
4. Wave 2: Dispatch gap-fill agents for elements at Low confidence
5. Output: Compile research document, save to Obsidian, propose SF write-back

Setup (Main Context)

1. Run the initial SF lookup. Read references/queries.md for SOQL queries. Extract: Account ID, primary Opportunity (highest Amount or most recent), stage, Amount, contacts with roles, and custom fields (Consequences_For_Not_Solving_Now__c, Timeline_for_Success__c, Purchasing_Process__c, Agent_Seats_input__c, AI_Agent_Use_Case__c, Phone_System__c, CRM__c, ICP_fit__c).

2. Prepare temp directory:

mkdir -p /tmp/meddpicc

3. Read org-context from Obsidian:

mcp__obsidian__obsidian_get_file_contents("Regal/org-context/org-context.md")

NEVER combine org-context reads with agent dispatches in the same message.

Named Sub-Agents

Use these specialized agents by name with the Task tool (subagent_type):

Each agent knows its tools. When prompting:

• Include the account name, Opportunity ID, and key contact names
• Specify which MEDDPICC elements the agent should focus on
• Tell agents to write findings to /tmp/meddpicc/<agent-name>.md
• Include the WRITE EARLY protocol: write partial findings in turns 1-2, update as discoveries are made

Sub-Agent Delegation Rules

Delegate to Sub-Agents When

• Reading multiple Obsidian account files
• Querying multiple SF objects
• Searching Slack or Gmail threads
• Extracting signals from transcripts

Acceptable in Main Context

• The initial SF lookup (Account + Opportunity + Contacts + custom fields)
• Reading org-context.md
• Quick obsidian_list_files_in_dir to check if account folder exists
• Final synthesis and document assembly

ALWAYS

• Spawn all independent sub-agents in a SINGLE message for parallel execution
• Use TodoWrite to track dispatch and completion
• Wait for ALL agents in a wave to return before proceeding

NEVER

• NEVER use run_in_background: true (MCP tools are unavailable in background subagents)
• NEVER synthesize until all agents return
• NEVER fabricate evidence or findings. Document only what is supported by source data.

Wave 1: Read Existing Artifacts

Dispatch in a single message (max 4 agents, then overflow):

Message 1 (4 agents):

• salesforce-researcher: Full deal data extraction. Prompt with Account name and Opportunity ID.
• obsidian-locator: Discover files in Regal/Accounts/<account>/
• transcript-locator: Find transcripts mentioning the account
• google-workspace-researcher: Search Gmail for threads with key contacts. Focus on: EB communication and budget language, champion email frequency and forwarding behavior, procurement/legal threads, approval chain discussions. Pass account name and contact names/emails from SF lookup.

Message 2 (gated on locator results, 2 agents):

• obsidian-analyzer: Read files discovered by obsidian-locator: Research docs, Context Brief, Contact Audit. Extract per-element MEDDPICC evidence. Pass the account name and instruct it to look for: pain statements, metrics discussed, evaluation process, decision makers, purchasing process, competitive mentions, champion indicators. Flag whether a Research doc exists (this gates company-researcher in Wave 2).

• transcript-analyzer: Read transcripts discovered by transcript-locator. Extract per-element MEDDPICC signals. Use three-state gating:

  • transcript-locator file exists with paths: dispatch with those paths
  • transcript-locator file exists but empty: skip, note [NO TRANSCRIPT DATA]
  • transcript-locator file missing: skip, note [TRANSCRIPT LOCATOR FAILED]

Three-Tier Fallback on Read

After Wave 1 completes, read each agent's output:

1. /tmp/meddpicc/<agent>.md exists: read it (primary)
2. File missing, Task result > 500 chars: use Task return text
3. File missing AND Task result < 500 chars: log data gap

Analysis: Synthesize Per Element

Read references/framework.md for the complete element definitions, evidence signals, and confidence criteria.

For each of the 8 MEDDPICC elements, synthesize Wave 1 findings into:

1. Table Answer: 20 words or less. The core answer for the Confidence Overview table. Format: H: <answer> or M: <answer> or L. Low confidence shows only the letter. For Economic Buyer and Champion, the answer is a single contact (Name, Title). See the table rules in template.md.
2. Current Answer: The best answer to the element's core question based on available evidence. Write in declarative prose, not bullet fragments. If unknown, state "Not yet determined" with what IS known.
3. SF Entry: 250 characters max. Mapped to the specific SF field for that element (see SF Field Mapping table above). For Contact lookups (EB, Champion): write the contact name. For picklist (Competition): must match an exact picklist value.
4. Confidence: High / Medium / Low based on evidence quality (see framework.md). Key rule: a direct, specific finding from ANY single source is High confidence. Direct = someone said it or a system recorded it. Inferred from behavior = Medium. No evidence = Low.
5. Rationale: Why this confidence level, citing specific evidence strength or weakness
6. Element-Specific Detail: Each element has unique sub-sections (e.g., Champion has "Champion Tests Passed", Pain has "Pain Cascade", Metrics has "Quantified Outcomes"). See the template.
7. Evidence: Every finding must cite its source type [SF/Obsidian/Transcript/Gmail/Slack/Web]
8. Gaps: What is missing or unverifiable
9. Discovery Questions: From framework.md, select the 2-3 most relevant questions to ask next based on what the gaps reveal

Table Answer Rules

• 20-word max per answer. No exceptions.
• No cross-section repetition. Each element's answer covers only its own domain. Metrics does not mention pricing, seat counts, or deal amounts. Decision Process does not restate who the EB is. Paper Process does not restate Decision Process steps.
• EB and Champion = one person. The table answer is a single contact's name and title. All rationale, evidence, and supporting context belongs in the detail sections.
• Metrics specifically: Answer what business outcome AI Agents will impact and how. Do NOT include POC pricing, seat counts, contract terms, or deal sizing.

Mark elements at Low confidence as gap-fill candidates.

Wave 2: Conditional Gap-Fill

Dispatch ONLY for elements at Low confidence. Match agents to gaps:

When multiple elements map to the same agent, consolidate into a single dispatch with all element prompts combined.

company-researcher Gate

ONLY dispatch company-researcher if obsidian-analyzer reported that no Research doc exists for the account. If a Research doc was found in Regal/Accounts/<account>/, the competitive landscape and company context are already captured there. Web search is expensive (~77K tokens, ~3 min); skip it when Obsidian already has the data.

Skip Wave 2 entirely if no elements are at Low confidence.

Dispatch gap-fill agents in a single message (max 4). Include in each prompt:

• The specific MEDDPICC element(s) to investigate
• What evidence is already known (avoid duplicate work)
• Write findings to /tmp/meddpicc/<agent>-gap-fill.md

Output

After Wave 2 (or after Wave 1 if no gap-fill needed), compile the research document.

Save to Obsidian using the filesystem path:

/Users/nick.yebra/Library/Mobile Documents/iCloud~md~obsidian/Documents/Core Vault/Regal/Accounts/<Account Name>/MEDDPICC - <YYYY-MM-DD>.md

Use the template in references/template.md.

Output Structure Summary

The document includes:

1. Header: Account, Opportunity, date, overall confidence summary
2. Confidence Overview Table: All 8 elements in a single | Element | Answer | table. Each answer is H: <20-word answer>, M: <20-word answer>, or L. EB and Champion answers are a single contact name and title. No cross-section repetition (Metrics excludes pricing/seats/deal size).
3. Per-Element Detail: For each element: current answer, SF Entry (250 chars max, mapped to specific SF field), confidence + rationale, element-specific sections, evidence citations, gaps, and discovery questions
4. Gap Analysis: Prioritized list of unknowns with suggested next actions
5. Recommended Discovery Questions: The highest-priority questions to ask in the next conversation, organized by element

SF Write-Back

After saving the document to Obsidian, propose SF field updates in the main context:

1. Compare each element's SF Entry against the value read during Setup
2. Skip fields where the existing SF value is equivalent or better
3. Present a summary table for user approval: Element, SF Field, Current Value, Proposed Value, Source
4. On approval, execute DML via mcp__Salesforce-Search__salesforce_dml_records. See references/queries.md for field-type-specific DML patterns (text, Contact lookup, picklist).
5. NEVER write to SF without explicit user approval

Critical Rules

1. Evidence-only findings. Every finding must cite specific evidence. No claims without a source.
2. Honest confidence. When evidence is thin or ambiguous, rate confidence Low. Inflating confidence creates false security.
3. No fabrication. If an element has no evidence, say so clearly. Never infer answers from adjacent elements.
4. Source attribution. Every evidence citation must include the source type (SF, Obsidian, Transcript, Gmail, Slack, Web).
5. Existing data first. ALWAYS read existing artifacts before dispatching gap-fill agents. The hybrid approach saves agent budget for real gaps.
6. Answers over scores. The goal is to document what is known and unknown, not to produce a number. A well-documented "we don't know yet" is more valuable than a guess.

Additional Resources

• references/framework.md - Complete MEDDPICC element definitions, discovery questions, evidence signals, and confidence criteria
• references/template.md - Output document template (single-account mode)
• references/batch-mode.md - Pipeline mode workflow and output template
• references/queries.md - SOQL queries for single-account and pipeline modes
• references/champion-tests.md - The 7 champion tests for Champion analysis