Who runs the best outbound?

We made four AI models run our real outbound pipeline, same brief, 51 times, and graded every email. One of them isn't close.

What we did

We built our outbound pipeline as five specialist AI agents.

Then we asked the obvious question: which AI model should run them? So we made four fight it out. The exact same pipeline, on identical briefs, 51 times, in a sandbox that never touches a real send queue. Each model had to find the trigger and verify the facts itself. Meet the five agents every model had to operate:

RidgeDiscovery

Finds a real company with a real, timely trigger — from public signals only.

CairnResearch

Verifies the buyers and the facts. Fabricates nothing.

SetterPlay design

Designs the specific multi-step sequence and channel fit.

ScribeCopy

Writes the offer-led emails. No AI tells, no banned phrases.

AnchorAdversarial QA

Re-checks every claim against sources and blocks anything wrong.

A taste of what we found

One model caught a backwards fact before it hit a prospect's inbox.

During the runs, a draft email to a Berlin-based climate-tech marketplace got a key fact about the prospect's own partnership exactly backwards:

What the draft claimed: The draft said Lufthansa Group was “one of six providers in the prospect’s 14-project climate portfolio.”

What was actually true: The May 2026 press coverage says the inverse: the prospect is one of six providers in Lufthansa Group’s portfolio.

Anchor, the adversarial QA agent at the end of the pipeline (running Claude Opus), re-fetched the primary sources, caught the reversal, and blocked the send pending a rewrite.

Imagine that email landing with their commercial lead, wrong about their own deal. That's the send the QA layer exists to stop. It got stopped.

The full results

So — who won?

Every model scored, the real costs, the failure modes, and the one we trust with a client's name. Drop your email — it all unlocks instantly.

No spam. One email, from Amit.

We run this rigor before a single email goes out.

Every client system gets stress-tested like this. Want it pointed at your outbound?

Book a call