Win-back subject lines in Claude Desktop
Lapsed subscribers are the hardest audience to write for — they already ignored you once. With marginal connected to Claude Desktop, you can draft, test, and ship win-back subject lines without leaving the chat window, then let the open and click data tell you which line actually pulls people back.
Why win-back campaigns need a real subject line test
Win-back emails fight inbox fatigue. A subscriber who hasn't opened in 90 days won't click unless the subject earns the tap. Tone matters enormously here — curiosity, a discount tease, and a plain 'we miss you' can produce wildly different open rates on the same list.
Guessing wastes your re-engagement window. A structured subject line A/B test removes the argument: you launch competing lines, track opens and clicks, and read the within-test lift instead of relying on intuition.
- Win-back segments are small, so every percentage point of open rate counts
- Subject tone (apologetic, incentive-led, curiosity) drives most of the variance
- Tracking opens and clicks together reveals which line drives action, not just curiosity
Running the test from Claude Desktop
Once the marginal email MCP server is connected, Claude Desktop can call the tools directly in conversation. You describe the campaign in plain language and the agent handles the experiment lifecycle.
A typical win-back flow uses four tools. Ask Claude to generate candidate lines, launch the split, pull results once data lands, and recommend what to send next.
- generate_variants — produce several win-back subject lines for your lapsed segment
- launch_test — start the subject line A/B test as a managed send
- get_results — read open/click tracking and within-test lift metrics
- recommend_next — get a data-backed suggestion for the winning line or a follow-up
Prompting Claude for win-back variants that convert
The quality of generated subject lines depends on the context you give the agent. Tell Claude who the segment is, how long they've been inactive, and whether an incentive is on the table. Specific prompts produce sharper variants than 'write some subject lines.'
Treat AI agent email marketing as a loop, not a one-shot. Run a test, let marginal report the lift, then ask Claude to riff on the winning angle for your next batch.
- Specify the inactivity window: 'subscribers with no opens in 120 days'
- Name the offer if any: '15% off, no minimum'
- Ask for a tone spread so the test compares genuinely different approaches
- Feed get_results output back into generate_variants for the next round
What you get out of marginal
marginal is a hosted email MCP server — there's nothing to deploy or maintain. You authenticate with a Bearer API key (marg_live_...) and point Claude Desktop at https://marginal.sh/mcp. The free tier covers 100 experiments per month, which is plenty for iterating on a win-back sequence.
Because the server handles managed sends and open/click tracking, your win-back subject line test stays end-to-end inside the agent workflow. Full setup details live at https://marginal.sh/docs/.
- Hosted endpoint at https://marginal.sh/mcp — no self-hosting
- Managed sends with open and click tracking built in
- Within-test lift metrics so winners are obvious
- 100 free experiments/month to test win-back angles
Get started with marginal
marginal is a hosted email marketing MCP server at marginal.sh. Sign up free, create an API key, and connect https://marginal.sh/mcp from Cursor, Codex, or Claude Desktop.
- 100 experiments/month on the free tier
- Four MCP tools: generate_variants, launch_test, get_results, recommend_next
- Listed in the MCP Registry as sh.marginal/mcp