ilya-a11y Reply Analysis
Read the inbox via MCP, categorize every reply, surface patterns, and return one actionable recommendation.
What you need
- Inbox access via LinkedIn MCP and/or email MCP
- Timeframe (default: last 7 days)
- Campaign name or hypothesis to filter by (optional — if blank, covers all)
Categories
- Interested: any positive signal (question, meeting request, "tell me more", "send it over")
- Not interested: explicit rejection or unsubscribe
- Out of office: auto-reply, vacation message, or equivalent
- No reply needed: spam, wrong person, bounce
- Needs follow-up: ambiguous reply that requires a human response
Process
- Pull replies from inbox via MCP for the specified timeframe
- Read each reply and assign one category
- For "Interested" replies: flag the company, note what they responded to, and suggest a next step
- Scan for patterns across all replies: what signal/angle/step got the most engagement?
- Identify one specific copy or targeting change the data supports
Output format
Reply analysis: [date range]
Total replies: [n]
INTERESTED ([n])
- [Name] @ [Company] | replied to: [step + subject] | signal: [what they responded to]
→ Next step: [suggested action]
NOT INTERESTED ([n])
- [summary, no need to list individually unless pattern is notable]
OUT OF OFFICE ([n])
- [list names for follow-up queue]
NEEDS FOLLOW-UP ([n])
- [Name] @ [Company] | [what they said] | suggested reply: [one option]
PATTERNS
[2-3 observations: which step got the most replies, which angle resonated, which segment responded]
ONE CHANGE TO MAKE
[Specific recommendation — a subject line, an opening line, a targeting filter — based on the data above]
Notes
- "Interested" always gets a named entry with a suggested next step — never leave it as a count
- If inbox access fails via MCP, report the error and list what was accessible
- Out-of-office replies should be queued for re-contact, not marked as closed