DTC Brands: Referral Program Email A/B Tests with marginal
Referral programs only work if the invite email gets opened. This guide shows DTC brands how to run a referral program email test through marginal, the hosted email MCP server, so your AI agent can generate variants, launch managed sends, and report lift without leaving your editor.
Why referral emails are worth testing
For DTC brands, a referral program turns happy customers into a low-cost acquisition channel — but the share rate is highly sensitive to how the invite is framed. A small lift in opens on the 'invite a friend' email compounds across every customer who reaches the milestone.
Most referral emails fall into predictable patterns: reward-led, friendship-led, or scarcity-led. A subject line A/B test tells you which angle actually moves your audience instead of guessing from a swipe file.
- Reward-forward: lead with the discount or store credit both sides earn.
- Generosity-forward: frame it as gifting a friend a deal they'll love.
- Status-forward: position the customer as an insider with something to share.
- Urgency: tie the referral bonus to a limited window or restock.
Running the test with the email MCP server
marginal exposes four tools to your AI agent over the email MCP server at https://marginal.sh/mcp. You describe the referral campaign in plain language and the agent handles variant creation, launch, and measurement.
A typical referral program email test flows through these tools in order:
- generate_variants — produce several subject line angles for the referral invite.
- launch_test — start a managed send and split traffic across variants.
- get_results — pull open and click tracking with within-test lift metrics.
- recommend_next — let the agent suggest the next variation to try.
What to measure for DTC referral campaigns
Opens are the first signal, but referral campaigns live or die on downstream sharing. Use marginal's open and click tracking to find the subject line that gets the invite read, then watch which variant drives the most clicks through to your referral landing page.
Because lift is calculated within each test, you compare variants on equal footing rather than against historical averages that drift with seasonality and promo cadence — important for DTC brands sending around launches and sales.
- Compare open rate across variants to validate the subject line angle.
- Track clicks to the referral page to gauge intent to share.
- Use within-test lift to decide a winner with confidence.
- Feed results into recommend_next to iterate on the runner-up angle.
Fitting it into an AI agent workflow
marginal is a hosted service, so there's no server to run yourself — connect a Bearer API key and your agent in Cursor, Claude Code, Windsurf, Cline, Zed, or any supported client can drive the whole email A/B test. That makes AI agent email marketing practical for small DTC teams without a dedicated lifecycle marketer.
Start on the free tier with 100 experiments per month, run your first referral subject line A/B test, and review the docs at https://marginal.sh/docs/ for tool details.
- Hosted endpoint: https://marginal.sh/mcp — no self-hosting required.
- Works across Cursor, Claude Desktop, Codex, Continue, and more.
- Free tier covers early referral experiments before you scale.
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