How to Run a Promotional Email Split Test
A promotional email split test compares two or more versions of a campaign — usually the subject line — to see which one earns more opens and clicks. marginal turns that workflow into a few MCP tool calls your AI agent can run directly from your editor.
What a promotional email split test actually measures
When you split test a promotional email, you send competing variants to slices of your list and measure which performs best. The most common version is a subject line A/B test, since the subject is the single biggest lever on open rate before anyone reads your offer.
marginal handles the mechanics for you: it splits the audience, sends both versions, and tracks opens and clicks so you can read the result instead of building the plumbing.
- Open rate — did the subject line earn the click into the inbox?
- Click rate — did the content and CTA drive action?
- Within-test lift — how much better the winning variant performed
- Sample slices so each variant gets a fair share of the send
Running the test through the email MCP server
marginal is a hosted email MCP server, so there's nothing to deploy. Your AI agent connects to the endpoint at https://marginal.sh/mcp with a Bearer API key and calls four tools to run an entire split test end to end.
A typical promotional email split test flows like this: generate candidate subject lines, launch the test, wait for engagement data, then ask for a recommendation on what to send next.
- generate_variants — draft subject line and copy options for the promo
- launch_test — start the managed send and split the audience
- get_results — pull open/click tracking and within-test lift
- recommend_next — get a data-backed call on the winning variant
Why AI agent email marketing fits split testing
AI agent email marketing shines on repetitive, measurable tasks, and subject line A/B tests are exactly that. Instead of context-switching to a separate dashboard, you describe the promotion in Cursor, Claude Code, or another MCP client and let the agent generate variants and launch the test.
Because marginal exposes results back through MCP, the agent can close the loop — reading lift metrics and proposing the next experiment without you copy-pasting numbers around.
- Works with Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Codex, and Zed
- Keep your testing workflow inside the tools you already use
- Free tier includes 100 experiments per month to get started
Tips for a cleaner split test result
A promotional email split test is only useful if the result is trustworthy. Change one thing at a time and give the test enough volume before declaring a winner.
- Test a single variable — usually the subject line — per experiment
- Use a meaningful audience size so lift isn't just noise
- Let opens and clicks accumulate before reading get_results
- See the marginal docs at https://marginal.sh/docs/ for tool details
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