Run an E-commerce Win-Back Email Test
Lapsed shoppers are some of the cheapest revenue you can recover — but only if your win-back emails actually get opened. Here's how to design and run an e-commerce win-back email test with marginal, the hosted email MCP server, so your AI agent can find the subject line that brings customers back.
Why win-back tests matter for e-commerce
Win-back campaigns target customers who bought once or twice and then went quiet. For e-commerce, that segment is large, well-understood, and already trusts your brand — which makes the open rate the main bottleneck. A subject line that nudges curiosity or signals a real reason to return can lift recovery dramatically.
Because the audience is finite and decaying, you don't get many shots. An email A/B test removes the guesswork: instead of betting your whole lapsed list on one subject line, you split traffic, measure within-test lift, and roll the winner to the rest.
- High intent, low cost — these shoppers already converted before
- Small windows: stale lists shrink, so testing fast beats testing later
- Subject line is the dominant lever on open rate for win-back sends
- Clear success metric: re-engagement (opens, clicks) before discount spend
Designing the win-back subject line A/B test
Good win-back subject lines fall into a few testable angles. With marginal's generate_variants tool, your AI agent can spin up several distinct framings at once, then launch them as a single subject line A/B test against your lapsed segment.
Keep variants meaningfully different so the results teach you something. Comparing 'We miss you' to 'We've missed you' wastes a test; comparing an emotional hook to a concrete incentive does not.
- Incentive: 'Your 20% comeback offer is inside'
- Curiosity: 'Something changed since your last order'
- Restock/new-arrivals: 'The items you loved are back'
- Loss aversion: 'Your account points expire soon'
- Personal/honest: 'Did we do something wrong?'
Running it through marginal's email MCP server
marginal is a hosted email MCP server, so any compatible client — Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Codex, or Zed — can drive the whole test loop. You point your agent at https://marginal.sh/mcp with a Bearer API key, and the MCP tools handle generation, sending, and measurement.
The typical AI agent email marketing flow for a win-back test: generate_variants produces the subject lines, launch_test sends them to a split of your lapsed audience with open/click tracking, get_results reports within-test lift, and recommend_next suggests where to push traffic. The free tier covers 100 experiments per month, which is plenty to dial in a recurring win-back program.
- generate_variants — draft competing win-back subject lines
- launch_test — managed send with open/click tracking
- get_results — within-test lift between variants
- recommend_next — what to ship and what to test after
- Docs and setup: https://marginal.sh/docs/ · endpoint: https://marginal.sh/mcp
Reading results and scaling the winner
Don't call a winner on raw counts — let marginal's within-test lift metrics tell you which subject line genuinely outperformed on your holdout split. For win-back, watch click-through as well as opens: a clever subject can win opens but fail to actually drive a return visit.
Once you have a reliable winner, send it to the remaining lapsed segment and feed the result back into your next experiment. Over a few cycles, your e-commerce win-back email test becomes a repeatable engine rather than a one-off campaign.
- Promote the higher-lift variant to the rest of the lapsed list
- Track clicks, not just opens, to confirm real re-engagement
- Re-test seasonally — what wins in Q4 may not win in spring
- Use recommend_next to chain experiments without starting from scratch
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