OnlyLeads exists because of a 14-day trial that went very wrong.
One of us ran a competitor's "AI SDR" on a free trial: 40,000 credits to spend over two weeks. We used it for LinkedIn outreach only, the narrowest, lightest thing the product does. We burned through all 40,000 credits in three days.
At their published rates that's somewhere between six and ten thousand dollars of credits, vaporised on light usage. Not because LinkedIn outreach is expensive. Because the agent is bad at the job and burns credits flailing: re-researching, re-enriching, retrying, looping on its own confusion. The headline credit allowance isn't priced for the work. It's priced for the bug.
The agent is theatre
Here's the uncomfortable truth about most of the AI-SDR category: the "autonomous agent" on the marketing site is a front end. Behind it is a brittle, database-driven back end that a human support team quietly has to un-mess every day. You're paying agent-loop prices for deterministic database operations plus some LLM drafting plus human cleanup you can't see.
We looked at that and asked the obvious question: what if you just removed the agent layer? Same value. No theatre. No support team needed to rescue it. No credits set on fire while a model decides what to do next.
The agent isn't the product. It's the part of the product that makes the bill unpredictable.
What deterministic actually means
OnlyLeads is a pipeline you can draw on a napkin, and every stage does exactly one thing:
- Signal ingestion. Deterministic. We watch the signals you tell us to watch.
- ICP scoring. Deterministic rules, with one LLM judgement on fit and urgency where a rule can't see nuance.
- Draft generation. The LLM, used where it genuinely earns its keep: writing one message, in your voice, for one specific person.
- Human review. Non-negotiable. You read it. You approve it. Nothing leaves otherwise.
- Send. Deterministic. The approved text, sent.
There's no loop. There's nothing to debug at 2am. There's no "the AI is hallucinating, please open a ticket." The LLM writes; the human decides; the machine sends. That's the whole trick.
Why this is the pitch
We don't position OnlyLeads as an "AI SDR" or an "AI GTM engineer," because those frames invite you to compare us on agent capability, which is exactly the axis we deleted on purpose. Lead with what's absent: no agent loops, no credit burn on retries, no support team patching a database behind your back. Predictable cost. Predictable behaviour. Message-by-message human approval.
We've been live and running real outreach on this model since launch. The rest of this blog is the story of the specific problems we hit and fixed along the way. They're all downstream of this one decision.