THE FORWARD TEST · SUBAGENT METHODOLOGY

The Forward Test.

Every message we send must be valuable enough that the recipient wants their boss, their peer, or their team to see it. If it fails the Forward Test, it doesn't send.

Why PVP wasn't enough.

Jordan Crawford's Permissionless Value Proposition framework at Cannonball GTM is one of the best pieces of outbound thinking in the last five years. Its core insight — that the best outbound messages deliver value a prospect would pay to receive — reshaped how serious operators think about cold email.

But PVP was designed for humans drafting messages. When you try to scale it through AI agents at volume, three things break:

  1. Quality drift. PVP's seven criteria are a great teaching tool. They're a worse quality control mechanism. Agents pattern-match on surface criteria and produce plausible-looking slop that technically passes but doesn't resonate.
  2. Sender bias. PVP asks "would the prospect pay for this?" That's still a sender-side question. The agent is guessing what the prospect values.
  3. No shareability test. A message can be valuable in isolation and still be something the recipient would never share. The shareability bar is the stronger signal of real value.

We needed a test that was simpler, harder, and agent-native.

The test.

Would the recipient forward this to their boss, their peer, or their team?

If yes, send it. If no, regenerate.

That's the entire methodology. One question. Two outcomes. Every subagent we build self-scores on this before any message leaves our infrastructure.

Why this is harder than PVP.

PVP asks one audience to like a message. The Forward Test asks two audiences to like it — the recipient AND whoever they might forward it to. A message that clears that bar has to be:

  • Valuable enough that sharing it makes the sharer look smart
  • Specific enough that it reads as research, not mass email
  • Timely enough that forwarding it is still relevant tomorrow
  • Actionable enough that the forwarded recipient has something to do

Most outbound fails at least three of those. Most AI SDR output fails all four.

The three criteria.

Every message our subagents draft is scored against three criteria before sending.

SPECIFIC Names. Numbers. Dates. A specific permit, a specific raise, a specific hire, a specific tech stack change. If the message could be sent to 50 other prospects with minor edits, it fails Specific.
TIMELY Last 30 days. Ideally 7. The signal must be from the last 30 days. Outbound based on a six-month-old data point reads as outdated research. The subagent checks signal dates before drafting and kills anything stale.
ACTIONABLE Something they can do today. The recipient must have something they can do today based on this message. Not "think about this next quarter." Something concrete, scoped, and doable this week.

How we implement it in Claude Code.

Every subagent we ship includes a forward-test.md file in the repo with the three criteria and a scoring rubric (0–3 per criterion, 7+ to send). Before any message reaches the send queue, the subagent self-scores the draft. Below threshold, it regenerates. After three failed regenerations on the same prospect, it skips and logs.

This means:

  • You can audit every message that was ever sent
  • You can adjust the threshold per campaign
  • Your team can modify the rubric as ICP understanding deepens
  • No human has to manually review each draft — the system self-regulates

The metric nobody else reports.

Every subagent/gtm engagement tracks forward rate — the percentage of sent messages that were forwarded by the recipient (measured by email thread replies, CC/BCC mentions, and post-meeting references).

Open rate is vanity. Reply rate is closer. Forward rate is the real signal. When a message gets forwarded, you've reached not just the recipient but everyone the recipient wanted to impress with their find.

We publish forward rate in every case study. No other AI GTM vendor reports it, because most can't.

Ready to ship?

Tell us what you want built. We'll reply within two business days with a fixed quote.

→ Book a build