What is an AI SDR? The definition, and the two rows where the label oversells
An AI SDR automates the reading and drafting parts of outbound sales development. What the term means, how it differs from an SDR and a BDR, and where it stops.
An AI SDR is software that does the preparation a sales development rep would do: finding accounts that fit, researching them, and drafting the first outreach. The name suggests it replaces the role. What it really replaces is the reading, which is most of the hours, and it stops short of the two things the role is actually judged on.
An AI SDR automates account research and first-draft outreach. SDR stands for sales development representative, the role that qualifies interest and books meetings rather than closing deals. The software handles the preparation well. It doesn't reliably decide who isn't worth contacting, and it can't be trusted to hold a conversation once someone replies.
SDR vs BDR vs AI SDR: what's the difference?
The two human job titles differ by who started the conversation. An SDR works with interest that already exists: a form fill, a trial signup, a webinar attendee. A BDR (business development rep) reaches out to accounts that haven't shown any interest yet. Plenty of companies swap the definitions, and some use one title for both.
The bottom two rows of that table are the important part. Very few vendors sell an "AI account executive" with a straight face, because closing means answering questions nobody anticipated. Those same two rows are where an AI SDR stops too, which is easier to see when the roles are lined up side by side than when a product is being demoed.
Where the term came from
"AI SDR" became a common label around 2023, when language models got good enough at reading messy web pages that the research step could be automated without producing nonsense. Before that, outbound tools automated sending and data enrichment, and the research stayed manual.
That's why the category name is a little misleading. What got automated wasn't the rep. It was the reading the rep did at 9am before writing anything.
Outbound is really five jobs, and vendors tend to price all five as one. Finding and drafting are close to solved, qualifying depends entirely on the evidence available, and the last two are where an unsupervised system loses you deals rather than saving you time.
What an AI SDR does well
The reading. An account worth contacting needs someone to open the careers page, the changelog, a couple of reviews and a LinkedIn post, and pull out the two facts that matter. That's thirty-odd pages to find one sentence, and it's exactly the kind of work models are good at. It's also where most of the hour goes in prospecting, so speeding it up isn't a small win.
The drafting. You get fluent, well-structured first drafts instantly. Fluency was never what held cold email back, but not starting from a blank page is still worth something.
The consistency. It applies the same standard to account four hundred as to account one, which no tired human manages at four o'clock on a Thursday.
Where it stops
Deciding not to send is the first gap. A tool that's judged on output produces output: ask it for fifty accounts and you'll get fifty, including the eleven where the honest answer was "nothing here this quarter". Restraint is the hardest behaviour to build and the most valuable, because sending to accounts with no real reason attached is what damages your domain.
We see how often "nothing" is the right answer in our own product. Across 792 opportunity-finding runs Revtive's agents made between July and September 2026, 52% came back empty. That's partly our own filters rejecting weak matches and partly that most markets don't produce something worth acting on every few hours. A system that never returns an empty result isn't more capable, it's just less willing to say so.
The reply is the second gap. Demos end when the email goes out, but real outbound starts at the reply, and replies are rarely a clean yes or no. "We looked at this in March, what's changed?" is a question about your roadmap, your pricing, and a conversation you weren't part of. An agent that answers it confidently and wrongly has cost you the account, and it sounds just as confident when it's wrong as when it's right.
What to ask in a demo
Four questions, none of them about features.
"Show me an account where it found nothing." If the tool has never come back empty, it doesn't have the concept, and every account you feed it will produce a message.
"Open the source link for that claim." Not a citation label, but the actual page, clicked, in front of you. Check the date on the page it lands on.
"What happens if I reply with a question?" Ask them to show it on a reply that isn't yes or no.
"Whose domain does this send from?" And then: what happens to that domain if the error rate turns out higher than expected?
The first question is the one vendors least expect, and the answer usually tells you more than the rest of the call.
How much does an AI SDR really cost?
More than the seat licence on the quote.
| Cost | Billed by | On the quote? |
|---|---|---|
| Seat licence | User, per month | Yes |
| Enrichment and data | Records looked up, including failed attempts | Usually not |
| Inboxes, domains, warm-up | Mailbox, plus the domain you're putting at risk | Usually not |
| Human review | The hours of someone already busy | Never |
Here are the five questions worth asking before you buy one, with a two-week test that answers them better than four vendor calls.
A definition that predicts how it behaves
If you want one sentence for a slide: an AI SDR is a research and drafting system with a sales job title. That framing predicts its behaviour correctly. The other popular framing, "a rep that never sleeps", predicts it'll handle a reply, and it won't.
Who shouldn't buy one
If your outbound doesn't work at ten hand-written messages a week, automating it won't help. These tools multiply a process, and nothing multiplied is still nothing, now spread across several domains.
If you have fewer than about fifty target accounts, the research is a morning a week and the tool is extra overhead you now have to review.
And if nobody can spare time to review drafts, think twice. Review isn't optional, and a tool bought to save time that creates a daily reading task tends to get cancelled, often after the domain has already taken some damage.
Revtive sits deliberately in the first half of that pipeline: it researches and drafts, cites a source for every signal, and never sends anything by itself. That split follows the table above rather than the market's habit of selling the whole pipeline as one thing.