What does Inside Dr. Jonah Tebaa's Case Against the Default Review Gate mean in practice?
Dr. Jonah Tebaa argues against default pre-send review gates because response latency drastically lowers B2B conversion rates. While fast, unsupervised replies yielded a 31 percent meeting-booked rate, fourteen-minute human review delays dropped bookings to 12 percent. Tebaa replaces default blocking with a consequence-based framework that evaluates whether an imprecise reply costs more than a delayed one, reserving human checks for sensitive exceptions while allowing routine qualification and scheduling to send instantly.
Dr. Jonah Tebaa has spent much of the past year studying how B2B service companies deploy AI in customer-facing chat, and one finding keeps repeating across the deployments he has reviewed: the biggest driver of conversion is rarely the quality of the AI's writing. It is where a human review step sits in the pipeline — before a message goes out, or after. That deployment pattern is not a niche one: Eurostat's enterprise survey records that 34.70% of EU enterprises using AI technologies applied them to marketing or sales in 2025, the most common purpose it measured.
In a recent piece of work, he compared two configurations of the same AI sales-chat system running on identical lead volume, roughly 100 inbound leads a week over four-week windows. The setups are illustrative of a broader pattern he has observed across service-industry deployments, rather than a single case study.
Two Ways to Place the Same Checkpoint
In the first configuration, the AI replied to an inbound message on its own — typically within about eleven seconds — asked a short set of qualifying questions, and offered a calendar slot without waiting for approval. A human reviewed a sample of the resulting transcripts afterward, correcting course on later messages when something needed adjusting.
In the second configuration, every AI-drafted reply waited for a human to read and approve it before it reached the lead. That review added an average delay of just over fourteen minutes per message.
Tebaa's argument is that these two setups look, from the outside, like a simple tradeoff between speed and caution. His data suggests the tradeoff is far more lopsided than most teams assume.
The Cost of Waiting, in Numbers
Leads who received a reply within a minute converted to a booked meeting at roughly 34 percent. Leads who waited more than five minutes converted at about 9 percent. At the configuration level, the fast-reply setup produced a 74 percent contact rate against 39 percent for the review-first setup, and a 31 percent meeting-booked rate against 12 percent.
What stands out in Tebaa's analysis is what did not change: the AI's underlying language, tone, and qualifying logic were identical in both configurations. The only structural variable was the position of the review checkpoint relative to send.
His conclusion is that most companies treat response latency as an incidental byproduct of process design rather than a decision in its own right. Review gates get placed at the start of a workflow by default and applied uniformly to every message, regardless of what is actually at stake if that message is slightly imperfect versus slightly late. Outside auditors have reached the same verdict about response speed: Harvard Business Review's study of how firms handle online sales leads concluded that most companies are not responding nearly fast enough.
A Framework, Not a Formula
Rather than asking whether an AI's replies are good enough to send unsupervised, Tebaa reframes the question around consequence: for a given category of message, does an imprecise answer cost more than a delayed one, or less? Pricing exceptions and sensitive complaints, in his framing, still warrant a pre-send human check. Routine qualifying questions and scheduling offers usually do not — the cost of a minor error is low and correctable on the next exchange, while the cost of delay compounds immediately.
He has distilled the finding into a short set of questions he now recommends before any team touches an AI-assisted messaging workflow:
- How much delay will this specific lead tolerate on this specific channel?
- Is the review step positioned to catch errors after the fact, or to block the message before it reaches anyone?
- Which failure is more expensive — a slightly wrong reply sent now, or a correct one sent late?
- Should review apply to every message, or only to the categories genuinely worth slowing down for?
- Is response-time distribution actually being measured, or assumed?
That distinction, Tebaa suggests, is the real lever available to teams running AI in customer-facing roles — not a faster model or sharper copywriting, but a more deliberate placement of the single checkpoint that determines how quickly a lead hears back.
