What the approval loop actually costs

The loudest argument in AI sales is about autonomy: should an agent send outreach without a human looking first? The first wave of AI SDRs said yes, and 2025’s churn stories said otherwise. We run the other model, where agents draft and people approve, so our production data can put numbers on what that loop actually costs and returns.

What reps do with agent drafts

Across more than 20,000 agent-drafted sends reviewed on Stalar, 94 percent were approved exactly as written. About 6 percent were declined, and fewer than 2 percent were sent back for regeneration.

Read that from both directions. The approval loop is cheap: nineteen of twenty drafts need one glance and one click, seconds per send. And the loop earns its keep: the 6 percent that got declined are the messages that would have gone out under a rep’s name with something wrong in them. At machine volume, unreviewed, that 6 percent is how domains burn and brands sour.

What the outreach returns

The approved emails replied at just over 11 percent. Benchmarks for cold email reply rates commonly land between 1 and 5 percent. LinkedIn messages drafted and approved the same way replied at roughly 13 percent.

The gap is not magic prompting. It is the boring parts done properly: contacts found and verified at send time rather than pulled stale from a database, drafts written from what is actually known about the account, and a person who knows the buyer signing off. Quality in, replies out.

The trade the market is still arguing about

Full autonomy saves the approval click and risks the 6 percent. The approval model costs seconds and catches it. Measured against what a bad send costs, the argument resolves itself: the click is the best-priced insurance in outbound.

That is why Stalar keeps a person on every send and every CRM write: agents do the research, the drafting, the logging and the list building, and nothing leaves the building past the line you set. The rest of the field is compared honestly in the AI SDR guide.

Methodology

Figures are aggregated across outreach drafted by Stalar agents and reviewed by users at the B2B organizations running campaigns on the platform, measured in August 2026. Reply rates count replies to executed sends; approval figures count drafts approved, declined or regenerated by a human reviewer. No organization is identifiable. External benchmark ranges refer to commonly cited cold outreach reply rates; results vary by market and list quality.