Champion/Challenger Email Subject Testing with marginal
A champion challenger email subject test keeps your best-performing subject line in play while continuously pitting it against new challengers. With marginal, an AI agent can run that loop end to end — drafting challengers, launching the send, and reading lift before recommending the next round.
What a champion/challenger test actually does
In a champion/challenger framework, the champion is your current winning subject line and the challenger is a new contender you want to beat. Unlike a one-off subject line A/B test, the pattern is meant to repeat: each round either confirms the champion or crowns a new one, so your baseline improves over time.
The discipline matters because subject performance drifts. Audiences fatigue, seasons change, and a line that won last quarter may lag today. Treating it as an ongoing contest — rather than a single experiment — protects open rates against that decay.
- Champion: the incumbent subject line you currently send
- Challenger: a fresh variant designed to outperform it
- Decision: keep the champion or promote the challenger based on measured lift
- Repeat: feed the new champion into the next round
How marginal runs the loop
marginal is a hosted email MCP server for AI agents. Connect it to your editor or agent runtime and the model gets four tools that map directly onto the champion/challenger cycle: generate_variants for challengers, launch_test for the send, get_results for open and click data with within-test lift, and recommend_next for the promote-or-keep decision.
Because the server is hosted at https://marginal.sh/mcp, there's nothing to deploy — you authenticate with a Bearer API key and the agent handles sends and tracking for you.
- generate_variants — draft challenger subject lines against your champion
- launch_test — start a managed send and split the audience
- get_results — pull opens, clicks, and within-test lift metrics
- recommend_next — decide whether to promote the challenger
Reading lift and calling a winner
The point of any subject line A/B test is a clean comparison, and lift is the number that tells the story. marginal reports within-test lift so you can see how much the challenger gained or lost against the champion on the same audience and send window — not against a stale historical average.
For an AI agent email marketing workflow, this is where recommend_next earns its place: it weighs the result and suggests the next move so the loop keeps running without you babysitting every round.
- Compare challenger vs. champion within the same send
- Use open and click tracking to separate curiosity from value
- Promote a challenger only when lift is meaningful, not noise
- Let the agent queue the next champion challenger email subject test automatically
Getting started with your agent
marginal works with Cursor, Claude Desktop, Claude Code, OpenAI Codex, Windsurf, Cline, Continue, and Zed, so you can drive tests from whatever environment you already use. The free tier covers 100 experiments per month — enough to run several champion/challenger cycles before committing.
Point your client at the endpoint, add your API key, and ask the agent to draft challengers for your current champion. Full setup details live in the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key (marg_live_...)
- Free tier: 100 experiments/month
- Docs: https://marginal.sh/docs/
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