Running a B2B Software Upsell Email Test
Upsell emails to existing accounts are some of the highest-leverage sends in B2B software — but the difference between a 12% and a 24% open rate is usually just the subject line. This guide shows how to design and run a B2B software upsell email test using marginal, the hosted email MCP server your AI agent already talks to.
Why upsell emails deserve their own A/B test
In B2B software, upsell campaigns target a known, qualified audience: customers already paying for a lower tier, seat count, or usage band. That makes them ideal candidates for an email A/B test because the segment is consistent and the conversion event — an upgrade — is clearly measurable.
Small lifts compound here. A subject line that pulls more opens on an upgrade announcement feeds directly into more trial-of-feature clicks and more expansion revenue, without acquiring a single new account.
- Test feature-led vs. value-led framing ("Now with SSO" vs. "Stop sharing logins")
- Compare urgency angles like usage limits approaching vs. seasonal pricing
- Isolate one variable at a time so wins are attributable
Designing the test with marginal
marginal exposes four MCP tools your agent can call directly: generate_variants drafts subject line candidates, launch_test ships them to your audience as a managed send, get_results returns open and click data with within-test lift metrics, and recommend_next suggests the follow-up experiment.
For an upsell, start with a subject line A/B test on the announcement email. Let your AI agent draft three to five variants that each lean into a different upgrade motivation — capacity, collaboration, security, or savings — then launch and read the lift before committing your full list.
- generate_variants — produce upsell subject line options at scale
- launch_test — managed send with open/click tracking built in
- get_results — within-test lift so you know which framing won
- recommend_next — chains your next email A/B test automatically
Driving it from an AI agent
Because marginal is an email MCP server, AI agent email marketing workflows fit naturally into the tools you already use. Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, and OpenAI Codex can all reach the endpoint at https://marginal.sh/mcp once configured with a Bearer API key.
A practical loop: ask your agent to propose upsell variants, launch the test to a holdout slice of your B2B software customer base, wait for results, then roll the winner to the remainder. The agent reads metrics and decides — you review.
- Endpoint: https://marginal.sh/mcp · Auth: Bearer key (marg_live_...)
- Free tier covers 100 experiments/month — plenty for iterating on upsell copy
- No self-hosting required; marginal is fully hosted
Reading results and iterating
After launch_test runs, pull get_results to compare variants on opens and clicks with within-test lift. For upsell campaigns, treat the click-through to your pricing or upgrade page as the signal that matters most — opens tell you the subject worked, clicks tell you the offer landed.
Use recommend_next to keep momentum: once you've found a winning subject line angle, your next B2B software upsell email test can move to body copy, CTA wording, or send timing. See the docs at https://marginal.sh/docs/ for tool parameters and examples.
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