Testing Referral Program Emails for Online Courses
Referrals are one of the cheapest acquisition channels for online courses — your best students already know other learners. This guide shows how to run an online courses referral program email test with marginal, the hosted email MCP server, so your AI agent can generate variants, launch sends, and read lift without leaving your editor.
Why referral emails need their own test plan
A referral program email asks for a different action than a course promo: instead of "enroll now," you're asking a happy student to share a code or link. That changes how subject lines, incentives, and calls to action perform, so reusing your enrollment copy rarely lands.
Treat the referral flow as a distinct experiment surface. The questions worth answering early are whether students respond better to altruistic framing (help a friend learn) or reward framing (you both get a discount), and how prominent the incentive should be in the subject line.
- Test reward framing vs. social/altruistic framing
- Compare incentive-in-subject vs. benefit-in-subject lines
- Try personal sender names vs. course-brand sender names
- Vary the share mechanic: unique code vs. one-click invite link
Running a subject line A/B test with marginal
marginal exposes four MCP tools your AI agent can call: generate_variants, launch_test, get_results, and recommend_next. For a referral campaign, ask your agent to draft several subject line variants, launch the managed send to a segment of recent course completers, then check open and click metrics with within-test lift.
Because marginal is a hosted email MCP server, there's nothing to deploy — connect at https://marginal.sh/mcp with a Bearer API key and your agent handles the rest. The free tier covers 100 experiments per month, which is plenty for iterating on a single referral campaign.
- generate_variants — produce subject line A/B test options for the referral ask
- launch_test — send variants to your completer segment and track opens/clicks
- get_results — pull within-test lift between variants
- recommend_next — get a data-backed suggestion for the next iteration
An AI agent email marketing workflow for course teams
Most course teams are small and ship copy fast. With AI agent email marketing through marginal, a growth engineer can run the whole loop conversationally inside Cursor, Claude Code, Windsurf, or any supported client.
A typical referral test cycle looks like this: brief the agent on the program (reward, audience, share mechanic), let it generate variants, launch the email A/B test, and review lift after enough sends accumulate. Then have it call recommend_next to choose what to test in round two — perhaps preview text or CTA wording.
- Works in Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, and Zed
- Keep referral copy in your repo and let the agent draft test variants from it
- Use lift metrics to pick a winner before scaling the send
- Docs at https://marginal.sh/docs/ cover tool inputs and auth
Metrics that matter for course referrals
marginal reports open and click tracking plus within-test lift, which tells you how one variant performed against another in the same send. For a referral program, clicks on the share link are your leading indicator — they signal intent to invite, which is upstream of new enrollments.
Watch the gap between open rate and click rate. A strong subject line that wins opens but loses clicks usually means the body copy or incentive isn't compelling enough — a clear signal to test the offer next, not just the subject line.
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