As a replacement for human sellers, no: 77% of teams running AI SDRs cannot confirm a positive ROI and only 3% of SaaStr's survey respondents had closed revenue from one. AI earns its keep on the tasks around the conversation, qualifying records and predicting who will pick up, while a human does the talking. At Sentrama we run AI across more than five million dials, and it has never once spoken to a prospect.
Do AI SDRs actually work? As a replacement for human sellers, no. As of November 2026, 77% of teams running AI SDRs cannot confirm a positive return on the investment, and only 3% of respondents in SaaStr's community survey had closed revenue from one. At Sentrama we run AI across more than five million dials, and it has never once spoken to a prospect. It removes the wrong half of every list, finds the direct mobiles, scores who will pick up and dials them in the hour they do. AI does the tasks. Humans do the talking.
I made this argument in May, in a LinkedIn article called The AI SDR Experiment Has Failed the Maths. It was a contrarian position then. Six months later, it has become the verdict of the people who sell the category's data.
What did the AI SDR category just admit?
On 30 September 2026, Vladyslav Podoliako, CEO of Belkins, published a long teardown of AI SDRs. Belkins publishes some of the benchmark data this market runs on... their cold calling benchmarks get cited everywhere, including by me. In their 2026 study of 120 B2B sales professionals, the most common implementation challenge with AI SDRs was generic or robotic output, reported by 37.5% of teams. Poor data quality came second at 31.3%. The conclusion of the piece: AI earns a place where humans control the targeting and the judgement, and autonomous AI SDRs break when left to run the conversation alone.
Sit with who is saying that. A benchmark publisher has no commercial reason to talk down the category it measures. When the people selling the data start writing the case for human sellers, the debate is finished.
What do the 2026 AI SDR ROI numbers show?
The scoreboard, as it stands in November 2026:
- 77% of teams using AI SDRs cannot confirm a positive ROI. Within that, 45% have not measured it at all and 32% say it is too early to tell (Belkins, 2026).
- 3% of respondents in SaaStr's community survey had closed revenue from an AI SDR. 11% were building some pipeline. 83% had nothing to show (Jason Lemkin, SaaStr, November 2024).
- Gartner predicts more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating cost and unclear business value among the reasons, and estimates only around 130 of the thousands of vendors selling "agentic AI" are building the real thing (Gartner, June 2025).
Separate datasets, same direction. AI itself comes out of these numbers fine. The job it was given does not.
Why is a category built on automation rediscovering humans?
Look at what the failures have in common. The most common complaint in the Belkins study is generic, robotic output... a complaint about the conversation. When a senior prospect picks up the phone, the opening seconds decide what that lead becomes, and a script that cannot read the room loses them. People buy from people. That half of the job, the pitch half, belongs to humans, and it always did.
This piece is about the other half: the infrastructure problem, getting your ICP on the phone at all. The two are different problems with different fixes, and the same research explains why so many teams went shopping for a robot in the first place. The Bridge Group's 2025 study of 351 B2B companies puts median SDR tenure at 1.9 years and average ramp at three months. Only 60% of reps hit quota, the lowest in the study's history. The human SDR is expensive and does not stay long. So companies tried to remove the human. Wrong end of the problem. The waste sits in the dialling, and dialling is exactly the kind of task software is good at.
Where does AI actually belong in B2B outbound?
On the tasks, before a human ever speaks and after they hang up. We built the Sentrama Platform in-house on more than 5,000,000 dials, and before the call its AI does four jobs. It removes about half of every sourced list as wrong records before a single dial. It finds direct mobiles for about 70% of UK prospects, and nobody is dialled without one. It scores pickup propensity, which is how we find the prospects who actually pick up the phone. Then it dials each of them in the hour they answer. That last one changes the maths.
The result, from our platform data: of the numbers worth dialling, about 19% pick up in the UK. The best published benchmark I can find is Salesfinity's Call Report at 10.1% dial-to-connect, measured across 1.3 million dials and 13 teams. Roughly 1.9 times the best published figure. The gap is what AI produces when you point it at data and timing rather than at the conversation.
Salesfinity's own report carries a line worth sitting with. In their latest month, connects across the dataset rose 19.5% while meetings booked barely moved. The market is getting better at reaching people, and what happens after the pickup has become the constraint. That is the strongest argument I know for keeping the conversation human. It is the one part of the system where judgement wins.
Since February 2025 we have mapped more than 5,000,000 dials and delivered more than 600,000 pitches. Every one of those pitches was delivered by a human. The AI did its work before the call connected, and picked it back up when the call ended: every call recorded and transcribed, every lead tagged as an SQL or an MQL with the reason the prospect gave. AI before the conversation finds it. AI after it proves it. The human is in the middle.
What should you do before you buy an AI SDR?
Name your problem first. The diagnostic is simple: look at what happens after someone picks up.
If your ICP is rarely picking up at all... your SDRs live on voicemail and switchboards, and you cannot say how many real conversations next month will hold... that is an infrastructure problem. AI genuinely earns its keep there, and the return shows up in who answers.
If prospects pick up and nothing happens afterwards, that is a pitch problem. An AI rep will deliver a weak pitch faster and at greater volume, which is what the 2026 numbers above record at category scale. Fix the right half.
An AI SDR might be able to do volume, but it can't fix the infrastructure. And it sure as hell won't get a pitch through to the human on the other end once they realise it's an AI. Failure all round.
Sorry, guys. If you bought an AI SDR to do the talking, you were sold a dream that's never going to come true.
The Monday move: ask your team how many of last week's dials reached a human being. If nobody can answer, start at the data layer... with us or without us.
Frequently asked questions
Do AI SDRs actually work?
As a replacement for human sellers, the 2026 data says no: 77% of teams running them cannot confirm a positive ROI and only 3% of respondents in SaaStr's community survey had closed revenue from one. AI performs well on the tasks behind the call, qualifying data and predicting who will pick up, with a human doing the talking.
Why do most AI SDR programmes fail?
The most common implementation challenge is generic or robotic output, reported by 37.5% of teams in Belkins' 2026 study of 120 B2B sales professionals, with poor data quality second at 31.3%. Both findings point the same way: the conversation needs human judgement, and the data layer needs more attention than it gets.
What is the ROI of an AI SDR in 2026?
Mostly unproven. 77% of teams using AI SDRs cannot confirm a positive return: 45% have not measured it and 32% say it is too early to tell. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing escalating cost and unclear business value among the reasons.
Where does AI belong in B2B sales?
On the tasks around the conversation. The Sentrama Platform, built in-house on more than 5,000,000 dials, uses AI to remove wrong records before a dial, find direct mobiles, score when each prospect will answer and dial them in that hour, and about 19% of the UK numbers worth dialling pick up as a result. After the call, every call is recorded and transcribed, and every lead tagged with the reason the prospect gave. The AI never speaks to a prospect. AI does the tasks. Humans do the talking.
Will AI replace human SDRs?
The companies that publish the category's own benchmark data now say otherwise. Belkins' September 2026 analysis concludes AI SDRs need humans controlling targeting and quality, and that conversations with senior buyers stay human. The replacement experiment produced 3% closed revenue; hybrid is where the category has landed.
How we work out who picks up, and when, and what 100 of them become: see how Sentrama works for sales leaders.
Onwards and upwards.
Frequently asked questions.
Do AI SDRs actually work?
Why do most AI SDR programmes fail?
What is the ROI of an AI SDR in 2026?
Where does AI belong in B2B sales?
Will AI replace human SDRs?
Sources.
- AI SDR vs. Human SDR: What AI Can and Can't Do in Outbound, Belkins
- 83% of You Haven't Gotten AI SDRs to Work. Yet., SaaStr
- Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, Gartner
- Cold Call Benchmarks: The Call Report, Salesfinity
- SDR Models, Motions & Metrics: 2025 Research Report, The Bridge Group

18 years in B2B sales, seven building outbound SDR teams.
26 guides by RobbieLinkedIn