Nobody running AI in outbound can prove it's working

The popular story is that AI already rewired outbound: autonomous SDRs booking meetings in their sleep, dialers that never get tired, sequences personalized at a scale no human team could match. Michael Maximoff, co-founder of Belkins and Folderly, put out the data that undercuts that story on July 7, 2026. AI lead generation ROI isn't a solved problem. For most teams, it isn't even a measured one.

The story everyone's telling

Walk through any GTM conference this year and you'll hear the same pitch, recycled across a dozen vendor booths: AI has already transformed outbound. The autonomous SDR is here. AI dialers close the gap human reps can't scale past. Every cold email platform now ships an "AI agent" tier, and the implicit claim is that the technology is mature enough to trust with revenue-critical work.

That story sells software. It doesn't match what teams running these tools report when someone actually asks them. Maximoff's team surveyed lead-gen practitioners for "Is AI in Lead Generation Actually Working? Belkins' 2026 Data Study," written by Iryna Yelisova and published on July 7, 2026. The headline number: 77% of teams using AI for lead generation cannot confirm it delivers a positive ROI. Not "it's mixed." Not "it's early but promising." They cannot confirm it, one way or the other.

This matters distinctly from the phone-versus-email debate that's dominated outbound budget conversations lately — see our breakdown of AI dialers against cold email spend. That piece is about which channel performs better. This one is about whether anyone can actually tell. Channel performance and ROI measurement are separate failures, and right now the industry has both, but the measurement failure is the one nobody's naming.

What the Belkins data shows about AI lead generation ROI

Three numbers from the study explain why the ROI question is so hard to answer honestly right now. First, adoption is new: 53% of surveyed teams adopted AI for lead generation only in the last six months. Most of what's running in outbound stacks today has been live for less than two quarters. That's not enough time to separate a real signal from a launch-quarter novelty effect, and it's certainly not enough time to have built a measurement process around it.

Second, AI hasn't actually touched the hardest parts of the job yet. Cold calling and objection handling remain 55-57% human-only in the same survey. The tasks vendors love to demo — a bot navigating a live objection, a bot booking a meeting off a cold dial — are still mostly done by people. AI is doing the easier surrounding work: list building, sequencing, drafting. That's useful, but it's a different and smaller claim than "AI runs outbound now."

Third, and this is the number that should stop the hype cycle cold: 45% of teams haven't measured ROI on their AI lead-gen tools at all. Not measured it and found it lacking — never measured it. Nearly half the market bought the tool and never built the loop to check whether it paid for itself. That's not a data problem. That's a decision not to look.

None of this is happening in a vacuum of easy wins elsewhere in outbound. Belkins' companion study, "B2B Cold Calling Benchmarks 2026," published June 26, 2026, put the per-dial connect rate at 9.9%, roughly one meeting booked per 370 dials, with Wednesday as the best day and the UK's 14.7% live-connect rate well ahead of the US's 9.0%. Outbound margins were already thin before anyone added an AI layer on top. Bolting an unaudited tool onto a channel that only converts once every few hundred touches doesn't make the arithmetic easier — it makes the case for measuring it more urgent, not less.

Why 43.8% feel misled

The study asked teams what unmet promise bothered them most about their AI lead-gen tools. The top answer, cited by 43.8% of respondents, was that the tool "requires no training or setup time." That's a specific and telling complaint. It's not "the AI writes bad copy" or "the leads are low quality." It's that the tool wasn't the plug-and-play product the vendor sold.

The satisfaction gap is the real headline

Here's the number that should reframe how you read all of the above: teams with confirmed positive ROI report 4.11 out of 5 satisfaction with their AI lead-gen tools. Teams without confirmed ROI report 3.0 out of 5. That's not a small gap, and it's not really about the tools.

Put those two numbers next to each other and the popular narrative collapses. It isn't that AI sales tools are good or bad. It's that teams who built a way to confirm ROI like their tools a lot more than teams who didn't — regardless of what the tool actually is. AI SDR ROI complaints are, disproportionately, a symptom of no one having built the scoreboard, not a verdict on the technology.

Crawford's fix: stop adding layers, start finding ground truth

Jordan Crawford, founder of Blueprint GTM, published "All AI in Go-To-Market Is Just This" on his Substack, On the Edge, on July 1, 2026. His argument is narrower and more useful than most AI-in-sales takes: AI in GTM should serve exactly one purpose — establishing ground truth about your customers. He ranks the sources of that truth in order of reliability: what your team believes internally is the weakest signal, what customers say is stronger, and what customers actually did is the strongest of all.

“Information is useless without action.”

This is the inverse of what most teams did with the 2026 AI outbound wave. They added an autonomous layer on top of a process they'd never fully audited, then hoped the new layer would somehow generate its own accountability. Crawford's framework says to do the opposite: mine the closed-won and closed-lost deals already sitting in your CRM before you add anything new. What actually separated the deals you won from the ones you lost? Most teams have never run that analysis, let alone let AI do the heavy statistical lifting on it.

That's also the more defensible use of AI SEO and AI outbound budgets broadly. Our reporting and analytics work exists for exactly this reason: pipeline that can't be traced back to a specific channel or campaign isn't pipeline you can optimize, it's pipeline you're guessing about. The same logic that applies to attributing pipeline to organic and AI applies to outbound tooling. If you can't trace the outcome to the input, you don't have a program, you have a subscription.

Fixing your AI lead generation ROI before you buy again

The contrarian position here isn't that AI in outbound is fake. Cold calling and objection handling being 55-57% human-only shows AI hasn't even reached the hardest parts of the job yet — there's real room to grow. The contrarian position is that most of the money going into AI outbound tools right now is an unaudited bet, and the fix being pitched — buy the next autonomous layer — makes the audit problem worse, not better. Every new tool you can't measure is another blind spot stacked on top of the last one.

None of this argues against AI touching outbound. It argues against buying the next tool before you've measured the last one. Something Inc.'s cold email work and our B2B engagements both start from the same place: trace the outcome to the input before you add anything else. Belkins' data says 77% of the market skipped that step. Crawford's framework is the fastest way to catch up without waiting for next year's version of this same study to say the same thing again.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.