Do AI SDRs Work? The Meeting Quality Problem

Every AI SDR vendor reports meetings booked. Almost none of them report what happened in those meetings — and that gap is where the money goes.

You bought the tool, the dashboard filled up, and three months later your pipeline looks roughly the same as it did before. The meetings were real. Somebody accepted the invite. But a chunk never happened, a chunk of the ones that did were the wrong company, and the rest went quiet after the call. So do AI SDRs work? The honest answer is that they work at the thing they are measured on, and the thing they are measured on is not revenue.

This is a criticism of the category BOSRAI sits in, so read it accordingly. We are not exempt from any of it.

Do AI SDRs work? It depends which number you are looking at

Every vendor in this space — 11x, Artisan, AiSDR, Amplemarket, us — reports some version of the same funnel: contacts sourced, messages sent, replies received, meetings booked. Those numbers are easy to instrument and they go up reliably. That is exactly why they are the numbers you get.

What nobody puts on the dashboard is the four-step decay between "meeting booked" and "revenue you can bank":

1. Did the meeting actually happen?
2. Was it the right company?
3. Did the person in the room have a problem worth solving and the authority to solve it?
4. Did it advance to a second conversation?

Each step removes a chunk. A tool that is excellent at step zero and untested on steps one through four can be simultaneously very good at its job and useless to your business.

The research points the same direction

Gartner surveyed 210 chief sales officers and senior sales executives in January and February 2026 and predicts that by 2028 AI agents will outnumber sellers ten to one — while fewer than 40% of sellers will say those agents improved their productivity. The analyst quote attached to that prediction is the whole argument in one line: "AI agents should not be viewed as a shortcut to sales productivity. They are only as effective as the systems they operate within."

The buyer side is stranger and more useful. In a separate Gartner survey of 646 B2B buyers run across August and September 2025, 67% said they prefer a rep-free experience and 45% had used AI during a recent purchase. Buyers want less friction. But Gartner also predicts that by 2030, 75% of B2B buyers will prefer sales experiences that prioritise human interaction over AI, and its analyst describes an "uncanny valley" effect where "the lack of genuine empathy, nuanced understanding, and the subtle cues that characterize human interaction can create a sense of discomfort or mistrust."

Read those together and you get the actual finding: buyers will happily let a machine handle scheduling and research, and they get suspicious when a machine tries to handle judgment. An AI SDR that books a meeting is doing the first thing. An AI SDR that decides the meeting is qualified is doing the second.

One more data point, because it reframes the whole conversation. The Bridge Group's 2025 research, across 351 B2B companies, found that 60% of SDRs hit quota — the lowest figure in the ten editions of that study — while raw pipeline generated per SDR rose to $3.78M from $2.83M in 2022. More pipeline, fewer reps making their number. The industry's own benchmark for human SDRs already measures raw volume and already has a quality problem. Automating the volume does not fix that. It scales it.

Where 100 booked meetings actually go

Only one number in the table below is sourced. The rest are the assumptions you have to fill in from your own CRM, and the point of laying it out this way is that most teams have never done it.

That held-meeting figure deserves its own sentence. RevenueHero analysed 6,428 B2B meetings across 15 industries and found a 6.5% outright no-show rate — but only 4,895 of those meetings, 76.1%, were completed. The rest were cancelled or rescheduled into the fog. So roughly a quarter of what a dashboard calls a "booked meeting" does not happen on the day.

Now attach money, with a made-up subscription price so the shape is visible. Say the tool costs $1,500 a month and produces those 100 booked meetings. That is $15 per booked meeting, which is a fantastic number and the one that ends up in the case study. Against the 12 that advance, it is $125. Both are true. Only one of them is a business input.

And the subscription is the smaller cost. Someone on your team sits through all 76 held meetings. At 40 minutes of prep, call and notes each, that is about 50 hours a month — a quarter of a full-time person — spent finding 12 real conversations. That labour never appears in the tool's ROI calculator. We walked through the equivalent arithmetic for human reps in the real cost of an SDR; the same discipline applies to software.

What vendors report vs. what predicts revenue

The right-hand column is harder to instrument and worse for marketing, which is why you have to ask.

Five questions to ask a vendor before you sign

1. Show me held-meeting rate, not booked. If the product cannot report it, it is not measuring the outcome you are buying.
2. What happens to a lead the model was wrong about? Ask who catches it and when. "The model learns" is not an answer for the prospect who already got the message.
3. Who signs off before a message goes out? Then ask how many customers actually leave that approval step switched on. Approval that everyone disables is a checkbox, not a control.
4. Can I see reply text, not just reply counts? Negative replies and unsubscribes get pooled into "engagement" more often than you would like.
5. Name three customers I can call. In March 2025, TechCrunch reported that 11x had displayed logos of companies that were not customers, one of which threatened legal action; sources described product issues and difficulty retaining customers, and the CEO stepped down that May. Logos are the cheapest thing on a website. References are not.

There is a longer version of this list in our autonomous sales agent checklist.

The uncomfortable part for us too

BOSRAI is an AI sales platform. We source leads, write outreach across email, WhatsApp and LinkedIn, follow up automatically, and keep it in a built-in CRM. Everything above applies to us.

So here is our position stated plainly, including the weak part.

We build human-in-the-loop by default: a person approves messages before they send, which is slower than full autonomy and deliberately so. We are WhatsApp-native rather than email-first, which matters mostly outside the US and UK, where a WhatsApp thread is a normal business channel and a cold email often is not. And we focus on founders and teams under about twenty people who are running outbound themselves. Plans run from a free tier through $79.99, $199, $499 and $999 a month, discounted annually — the full breakdown is on our pricing page.

The weak part: we have no published case studies, no named customers and no benchmark meeting-quality data. We are not going to invent any. Which means that when you apply question one from the list above to us, the honest answer today is that you should run the meetings through your own CRM for a quarter and judge us on your numbers, not ours. That is also the right answer for every vendor in this category, including the ones with a wall of logos.

If you want the fuller argument for why approval beats autonomy, it is in human-in-the-loop AI sales. If you are comparing what different vendors charge, we keep an AI SDR pricing index.

So do AI SDRs work

At sourcing, drafting, sequencing and following up: yes, measurably, and a solo founder doing this by hand is losing to one. Tomasz Tunguz relayed a chief revenue officer at a public company describing an AI SDR that generated substantial lead volume over nine months and produced no sales, and his read on it was not that the technology failed: "So it's not to say that AI won't work. It's to say many of us [still] do not know how to use AI."

At judging whether a conversation is worth your time: not yet, and the vendors reporting only booked meetings are the ones counting on you not to check. Buy the volume. Keep the judgment. Measure held meetings from the first week, because that is the number that tells you the truth about the other ones.

Sources

• Gartner, 28 July 2026 — AI agents to outnumber sellers 10:1 by 2028, fewer than 40% of sellers reporting productivity gains — the 10:1 and sub-40% figures, the 210-executive survey, and the Dan Gottlieb quote.
• Gartner, 9 March 2026 — 67% of B2B buyers prefer a rep-free experience — the 646-buyer survey run August–September 2025, the 67% and 45% figures.
• Gartner, 25 August 2025 — 75% of B2B buyers will prefer human-led sales experiences by 2030 — the 2030 prediction and the "uncanny valley" quote from Colleen Giblin.
• The Bridge Group — SDR Models, Motions & Metrics, 2025 research report — 351 participating B2B companies, 60% quota attainment as the lowest in the study's history, $3.78M raw pipeline per SDR against $2.83M in 2022.
• RevenueHero — B2B meeting no-show benchmark — 6,428 meetings across 15 industries, 419 no-shows (6.5%), 4,895 completed (76.1%).
• TechCrunch, 5 May 2025 — 11x CEO Hasan Sukkar steps down — the March 2025 reporting on non-customer logos, the threatened legal action, and the retention and product issues described by sources.
• TechCrunch, 26 December 2024 — AI sales rep startups are booming. So why are VCs wary? — the Tomasz Tunguz account of a public-company CRO whose AI SDR produced volume but no sales, and his quote.