An AI SDR is not an email bot. Here's what one actually did.

An AI SDR is a system that does the research and first-touch work of a sales development rep. It notices buying signals (a pricing-page visit, a new VP of Sales, a funding round, a job posting for the role your product replaces), enriches the account and the person, decides whether they fit, routes the ones that do to a human within minutes, and drafts the first message with the research already done.

The email-sending bot that vendors demo is the last ten percent of that. It's also the part that matters least, because a bot sending to a cold list converts like a human sending to a cold list. What changes the numbers is everything upstream: knowing who to talk to today, and getting them to a person while they're still looking.

A venture-backed software company selling to healthcare operators

A small SDR team, a pipeline that ran almost entirely on cold outbound from imported lists, and inbound that was a rounding error. In the quarter before the build, the team created about 60 deals, and the average won deal was in the mid $30Ks.

I didn't start with the bot. I started by wiring the company to notice who was already paying attention: a visitor identification tool pointed at the site and piped into Clay, every identified visitor enriched and scored against the ICP, the ones that passed dropped into a Slack channel the reps lived in. Conference registrant lists got the same treatment. A whitepaper campaign landed its downloads in the same pipe. Then the sequencer, which had sat idle for two months, got three lanes with three messages, each fed by Clay with the research already done. Within a quarter the team was enrolling about 600 contacts a month, and roughly half of all new deals that year touched one of those sequences.

The fully automated piece, cold contacts enriched by Clay and worked without a human picking them, pushed about 5,000 contacts into the CRM. Forty became opportunities, roughly $1.5M in pipeline, and a handful closed at about $80K each with most still open. That's 0.8% contact to opportunity. A human on the same list does about the same. The difference is the human cost a salary.

What moved the company was inbound. Form fills went from under twenty a month to over a hundred, and the quality held: 19% of form contacts became opportunities, which is a demo-request rate, not a content-download rate. Two quarters in, deal creation had tripled and the average won deal had nearly doubled, because the signals were surfacing bigger accounts. Same reps.

While the system above was running, the same team tried the other path: an off-the-shelf prospecting agent, the kind sold as an AI SDR in a box, switched on inside the CRM. It enrolled about 2,200 contacts. Three out of four errored out. Six finished the sequence. Some of that is configuration and some of it is the product, and from where the reps sat it didn't matter which; the bought version of the thing did a fraction of what the built version did, on the same contacts, in the same CRM.