Analyse 10,000 emails with AI using the MAP framework

Executive overview

Most business decisions rely on memory, which is inherently subjective and incomplete, even though the raw data needed for objective analysis already exists in inboxes, calendars, and messaging tools. The obstacle is that large datasets exceed the context window of any AI chat tool, causing it to compact earlier work and lose progress. The MAP framework — three files: Mission, Activity log, and Payoff — solves this by giving the AI a persistent roadmap that survives memory resets. A practical example shows how to surface "open loops": commitments made or received that were never followed up. Documented data is the dividing line between businesses that will extract compounding value from AI and those that will not.

Exporting your raw data

• Gmail users export email via takeout.google.com; deselect all items except Mail, then download.
• The same export approach works for LinkedIn DMs, Slack messages, Microsoft calendar events, and social-platform message archives.
• After download, create a single local folder and place the exported data folder inside it.
• The folder becomes the AI agent's workspace; everything it reads and writes stays within it.

How the MAP framework works

• M — Mission file: the only file you create manually; tells the AI what to do and how to recover after a memory reset (named agents.md for ChatGPT/Codex, CLAUDE.md for Claude).
• A — Activity log: a to-do checklist the AI creates and updates; prevents duplicate work across memory cycles.
• P — Payoff file: an insights markdown file where the AI writes every finding as it processes each batch.
• The AI runs a tight loop — wake up → read mission → check to-dos → process one batch → write insights → update to-dos — and restarts this loop each time its memory is compacted.
• Batching emails in groups of 20 keeps each processing cycle manageable and auditable.

What goes inside the mission file

• Primary task section: states the objective in plain language (e.g., "find open loops and broken commitments across all emails").
• Session start section: instructs the AI to re-read the mission file first after every reset, then check or create the activity log and insights file.
• Step zero section: defines what the AI should create if no to-do or insights file exists yet, including required sections and an example structure.
• What to find section: specifies the exact signals to look for — for open loops, that means outbound promises with no reply and inbound commitments never fulfilled.
• Rules section: enforces one batch per cycle, mandatory updates to both files after each batch, and full-access permissions so the AI can run without constant human approval.

Choosing and setting up the right tool

• Claude Desktop (Mac and Windows) supports this workflow via the Cowork tab with Opus 4.6 and extended reasoning enabled.
• OpenAI Codex desktop app (Mac only at time of recording) provides a similar local-agent capability through a ChatGPT-style interface.
• Both tools require opening the project folder directly inside the app and setting permissions to full access.
• Full-access mode removes per-action approval prompts so the AI can run autonomously for 30–40 minutes unattended.

Open loops as a practical use case

• An open loop is any email thread where a commitment was made — by you or the other party — and never resolved.
• Running this analysis monthly can surface 10–50 unchased opportunities per inbox: leads, partnerships, or client follow-ups that quietly died.
• The insights file separates findings into outbound open loops (you dropped the ball) and inbound open loops (they dropped the ball), plus a ranked top-10 recovery list.
• The same mission-file structure can be repurposed for any analytical goal: time allocation, relationship health, recurring complaints, or sales patterns.

The documentation advantage

• AI can only leverage data that has been captured; undocumented activity produces no fuel for analysis.
• Recording every meeting to generate transcripts, time-blocking calendars, and keeping communications in writing are the three highest-leverage documentation habits to start immediately.
• Businesses that document consistently and feed that data into AI will compound their advantage over those that rely on memory and intuition.
• The gap between documented and undocumented organisations will widen every year as AI tooling matures.