November 10, 2026 · Opinion
Why I'd wait before hiring an AI SDR
Ten thousand emails, two percent reply, thirty meetings — the funnel maths works on a slide and fails on a new domain. What the pitch leaves out, and the version I would actually buy.
I know this one will annoy people, because a lot of money is being spent on the opposite argument.
The pitch: an AI SDR emails ten thousand prospects, personalises every message, handles the replies and books meetings while you sleep. The demo is genuinely good. The maths is where it falls apart.
Run the funnel with honest numbers. Ten thousand emails, a good list, a decent offer: two percent reply, maybe 0.3% book a meeting. That is thirty meetings. Sounds fine until you price the inputs.
Deliverability
A new domain sending ten thousand cold emails gets throttled, filtered or blacklisted inside a fortnight. The vendors solve this by rotating domains, which means your brand now lives on three lookalike addresses. Buyers notice. Some of them report it.
List quality
"We have 200 million contacts" means somebody scraped them. The overlap between that list and people who actually buy your product is the number that matters, and nobody sells you that number.
Brand risk
An AI that personalises at scale will eventually personalise wrongly. "I saw you recently expanded into Germany" sent to a company that just closed its German office. One of those a month is survivable. Three a week is a reputation.
The replies, which the demos skip
When someone answers "not now, but send me the spec", the AI has to know what to do. Most of them book a meeting anyway, or send a calendar link, and the human who could have converted the conversation never sees it. The reply is the asset. Handing it to a robot is how you lose the deal you already half-won.
Where it does work
You already have a narrow ideal-customer profile and can name it in one sentence. You have content — a case study, a calculator, a teardown — that a stranger would actually read. And a human replies within five minutes of any response, because that is when the reply is worth anything.
Notice that two of those three have nothing to do with the AI.
The version I would buy
An AI that researches fifty accounts properly, writes a first line that proves someone read their website, and hands the whole thing to a human who sends it from their own address. Fifty emails, not ten thousand. The maths is worse on paper and better in the bank.
I will be wrong about this eventually. The tools are getting better at reply handling, and the deliverability problem is a policy problem rather than a technical one. But right now, if you are a ten-person exporter, the money is better spent on one good list and a human who follows up.
Full disclosure. This take costs us nothing and might cost you a vendor relationship. If you want reviews that are not shaped by who advertises, this is the page that keeps them that way.
We publish the pricing model, minimum commitment and export-readiness notes for every tool in the library, and we flag it when a vendor changes one without changing the headline price.