Running a Number in Subject Line Experiment
Does adding a number to your subject line actually move open rates? Instead of guessing, run a controlled number in subject line experiment with marginal — the hosted email MCP server that lets an AI agent generate variants, launch the test, and report lift.
Why test numbers in subject lines
Numbers tend to draw the eye in a crowded inbox — "5 ways to cut churn" reads differently from "ways to cut churn." But the effect varies by audience, send context, and whether the digit feels specific or arbitrary. The only reliable way to know is a subject line A/B test against your own list.
A good number in subject line experiment isolates one change: the presence (or format) of a number. Keep the rest of the copy constant so the open-rate delta you measure is attributable to the digit and not a reworded promise.
- Digit vs. word: "3 tips" vs. "three tips"
- Number vs. no number: same copy, one with a count
- Specificity: "save 23%" vs. "save up to 25%"
- Position: leading number vs. number mid-sentence
How marginal runs the experiment
marginal exposes four MCP tools your AI agent can call directly: generate_variants, launch_test, get_results, and recommend_next. You describe the hypothesis in plain language, the agent drafts the numbered and non-numbered variants, launches a managed send, and tracks opens and clicks for you.
Because it's a hosted email MCP server, there's nothing to deploy. Connect from your editor or agent, authenticate with a Bearer API key, and the send + tracking pipeline is handled for you.
- generate_variants — draft numbered vs. word/no-number subject lines
- launch_test — split your audience and send both versions
- get_results — pull open/click rates and within-test lift
- recommend_next — suggest the next variant to try
Reading the results without fooling yourself
Within-test lift metrics from get_results tell you how the numbered variant performed relative to the control inside the same send — which controls for time-of-day and list-level swings. Look at the magnitude of the open-rate difference and whether your sample is large enough to trust it before declaring a winner.
Treat a single experiment as a signal, not a law. A number that wins on a product-launch email may lose on a re-engagement campaign. Re-run the test in different contexts and let recommend_next guide your follow-up subject lines.
Set it up for AI agent email marketing
marginal plugs into Cursor, Claude Desktop, Claude Code, Codex, Windsurf, Cline, Continue, and Zed, so your number in subject line experiment can run from wherever you already work. The free tier covers 100 experiments per month — plenty to A/B test numbers across several campaigns.
Point your client at https://marginal.sh/mcp, then ask your agent to test a numbered subject against its plain counterpart. Full tool reference lives 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