Do AI SDRs actually work?
Partly, and it depends on the job you give them. No neutral study of meetings held per dollar across AI SDR products is published, so claims of big wins come from vendors and claims of failure often come from competitors. This page separates what the tools plainly do from what has to be tested on your own list.
What the data shows
Each figure is labelled with who produced it and what kind of source it is. None is a neutral measure of meetings held per dollar.
| Figure | Source | Type |
|---|---|---|
| 77% of teams using AI for lead generation cannot say whether it has delivered a positive return. 45% have not measured it and 32% say it is too early to tell. | Belkins, 120 B2B sales professionals, May to June 2026 | Survey run by a vendor (Belkins sells outbound services) |
| "AI-assisted with human review" was the most common answer for ten of thirteen tasks. Fully autonomous use never exceeded 20% on any single task. | Belkins, same survey | Survey run by a vendor |
| Of an audience polled by SaaStr: 3% had generated real revenue from AI SDRs, 11% were building some pipeline, 83% had got nothing. | SaaStr, November 2024; number of respondents not stated | Informal poll |
| "AI SDRs" appeared as a distinct category for the first time, at 1% of respondents. | Bridge Group, 2025 SDR report, 351 B2B companies | Survey |
| 15 to 20 hours a week actively managing five AI SDRs, and 60 to 90 days for each to reach peak performance. | SaaStr, on its own deployment | Self-reported |
Read together, they say the tools are in use, most teams cannot yet show a return, and the human time is real. They do not say your list will or will not work.
What they do work for
An AI SDR reliably produces output: a researched list, an enriched contact record, a personalised first email for every row. Producing the work is not the same as producing meetings. Output is the part you can check in an afternoon.
Where they fail, and the fix
| Failure | What it looks like | Fix |
|---|---|---|
| Email that lands in spam | Volume climbs, replies fall, complaints rise. Google's sender guidelines treat senders of 5,000 or more messages a day to personal Gmail accounts as bulk senders and ask them to keep spam complaints low. | Send from a separate domain, start small and watch complaints. |
| Wrong facts in a personalised line | A message praises a product the prospect does not sell. | Read every message at first. SaaStr advises reviewing every conversation in the first months. |
| Replies nobody handles | A prospect asks a real question and gets a canned answer or silence. | Name one person who owns replies and checks them every day. |
| Meetings that are booked but not held, or not wanted | The calendar fills and the pipeline does not. | Write down what counts as a meeting, and count meetings held. |
| Rules for your target countries | Email and calling rules differ by country. | Have your own counsel check them before you send at volume. |
| No phone | The tools reviewed centre on email and LinkedIn. | If your buyers answer calls, plan on a person. |
Ready to start? Check four things
- A named owner for replies. One person, with time set aside each day.
- A written playbook. The segment, the offer and what a good reply looks like. If a person cannot yet book meetings from it, software will not.
- A separate sending domain, so a mistake does not damage your main one.
- A meeting definition. Who attends, with what title, and what has to be agreed, so held meetings can be counted the same way every time.
If one is missing, fix it first. The worth-it page says who should buy now and what it costs per meeting held.
Why results vary so much
Three things differ between a good result and a bad one: the list (a narrow, well-defined one works better than a broad one), the offer (one a short email can carry), and the person watching the tool. A study of one product on one list tells you little about yours.
A four-week test
- Pick one segment you already know converts, and a list size you can review by hand.
- Run the tool on half the list and keep the other half for a person to work, so the comparison uses the same accounts.
- Count meetings held, not emails sent or replies. Note who handled each reply.
- Add up the human time the tool needed: checking copy, answering replies, fixing the domain.
- Compare cost per meeting held with the human half.
If the tool wins on cost and the meetings are as good, expand it. If it books meetings nobody wants to take, you have learned that cheaply. For the wider question of headcount, see will AI replace SDRs; the rest of the topic is on the AI SDR hub. To work with a person, compare fractional SDR providers.
FAQ
Do AI SDRs book meetings?
Some do. Whether those meetings are held, and turn into pipeline, depends on your list, your offer and who handles replies. Test on a small list before you commit to a plan.
Why do AI SDR results differ so much between companies?
List quality, offer, and how closely a person watches the tool and the sending domain all change the result, so one company's numbers rarely carry over.
Can an AI SDR make phone calls?
The tools reviewed in the sources below centre on email and LinkedIn. If calls matter for your buyers, plan on a person for them.
Can I trust vendor case studies?
Treat them as claims. They come from the seller and rarely show the human time the tool needed. Ask for meetings held and cost per meeting, then run your own test.
Sources
- Artisan, "AI SDR vs human SDR: which should you hire?": artisan.co
- Bain Capital Ventures, "The Case for Humans in Sales: Why BDRs Still Win in the Age of AI": baincapitalventures.com
- Instantly, "AI BDR vs human SDR": instantly.ai
- Google, "Email sender guidelines" FAQ: support.google.com
- Belkins, "AI for lead generation study" (vendor-run survey): belkins.io
- SaaStr, "83% of you haven't gotten AI SDRs to work yet": saastr.com
- SaaStr, "6 months of AI SDRs: what's worked": saastr.com
- The Bridge Group, "SDR Models, Motions & Metrics: 2025 Research Report": bridgegroupinc.com