Subscription Boxes: Referral Program Email Test Playbook
Referrals are the cheapest acquisition channel a subscription box has — your happiest subscribers already love telling friends. This guide shows how to run a focused subscription boxes referral program email test using marginal, so your AI agent can write variants, launch managed sends, and report lift without you leaving the editor.
Why referral emails matter for subscription boxes
Subscription boxes thrive on recurring revenue and word-of-mouth. A referral program email is the moment you ask a loyal subscriber to share their box — and small wording changes can swing how many people forward the offer. Testing those changes beats guessing.
The variables worth isolating are usually narrow: how you frame the reward (free box vs. account credit), the urgency, and the subject line that gets the email opened in the first place.
- Highlight the mutual reward — both referrer and friend get something
- Test whether 'free box' or 'account credit' framing drives more shares
- Trigger the email after a positive moment, like a renewal or a 5-star rating
- Keep the share link prominent and the ask single-purpose
Setting up the email A/B test with marginal
marginal is a hosted email MCP server, so your AI agent connects to https://marginal.sh/mcp with a Bearer API key and drives the whole workflow through four tools. For a referral program email, the loop is generate, launch, measure, repeat.
Start by asking your agent to call generate_variants for the referral offer, then launch_test to send them to a sample of your subscriber list. Open and click tracking captures real engagement, and get_results returns within-test lift so you can see which variant actually pulled more referral clicks.
- generate_variants — draft multiple referral subject lines and copy angles
- launch_test — run the email A/B test as a managed send
- get_results — open/click metrics with within-test lift
- recommend_next — let the agent suggest the follow-up variant to try
Subject line A/B tests that lift referral opens
Most referral emails fail at the inbox, not the offer. A subject line A/B test is the highest-leverage experiment you can run: if no one opens, no one shares. For subscription boxes, contrast a reward-led subject against a community-led one and let the data decide.
Run two or three subject variants per send, keep the body identical, and isolate the open-rate effect before you optimize the referral copy itself.
- Reward-led: 'Give a free box, get $15 credit'
- Curiosity-led: 'Your friends keep asking about this'
- Identity-led: 'Help us build the box club'
- Always send identical bodies so you measure subject impact cleanly
Running it from your AI coding agent
Because marginal speaks MCP, AI agent email marketing fits inside the tools you already use — Cursor, Claude Code, Claude Desktop, Windsurf, Cline, Continue, Codex, and Zed. You describe the referral campaign in natural language and the agent calls marginal's tools to run it.
The free tier covers 100 experiments per month, which is plenty to validate a referral program email test across a few subject lines and copy angles. See the docs at https://marginal.sh/docs/ to wire up your client and start sending.
- No self-hosting — marginal is a hosted MCP endpoint
- Auth with a Bearer key (marg_live_...) once, then drive tests by prompt
- Iterate fast: test, read lift, and let recommend_next propose the next round
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