I've sat on both sides of this argument in the same quarter. In April I told a peer group AI SDRs were a solved problem. In June I watched a team I respect rip one out after three weeks because it kept confidently misquoting their own pricing page. We were both right, and we were both looking at the same category.
That's the confusing part about where GTM is right now. Nobody's lying to you on purpose. The category got oversold by vendors promising autonomous revenue with a five-minute setup, and it got undersold by teams who bolted one on, watched it produce fifty forgettable drafts, and decided the whole idea was smoke. Even the loudest voice in the category spent two years telling companies to stop hiring humans, then quietly retired the slogan and hired a human BDR. That's a real data point. It's just not the one most people took from it.
It's honest to say the first wave underdelivered. It's not honest to conclude the category doesn't work, I've now sat close enough to a few good deployments to know the difference isn't the model. It's what the model was handed.
Here are the five beliefs I hear most often at dinners with other revenue leaders, and what actually happens to each one once you run it against a real pipeline instead of a demo environment.
The short version
- The failures were briefing failures, not model failures.
- An AI SDR behaves like a new hire, not a feature you flip on.
- It earns its keep on judgment applied at volume, not on “simple” tasks.
- Buyers object to bad automation. Not to automation.
- What it actually removes from a rep's week is the unpaid integration work nobody put on a job description.
1. “We tried one and the output was bad, so they don't work”
This is the most common objection I hear, and it's also the most reasonable one, because the experience behind it is real. I've watched smart RevOps people run an AI SDR, read the first fifty drafts out loud in a Slack huddle, wince, and quietly switch it off by Friday.
But go look at what those deployments were actually given to work with. A company name. A one-line description field. A “tone of voice” box that fits about a tweet. Sometimes a list of contacts with nothing attached but an email address.
Here's the thing nobody wants to hear in week one: what separates a good deployment from a bad one is almost never the model underneath it. It's what the thing was handed to be right about.
What's actually true: An AI SDR should be doing the volume work: research, drafting, first replies, chasing a reschedule, inside rules you set. It's not there to replace the judgment in a discovery call, and any vendor pitching it that way is quietly setting you up to be disappointed by week three, once the novelty wears off and the drafts start repeating themselves.
Give it nothing to be right about, and a model does exactly what any of us would do when asked a question we can't answer honestly. It produces something plausible-sounding instead.
2. “It's software, so it should work on day one”
Software gets configured. An agent gets briefed. That distinction sounds like a marketing nuance right up until your first week, when it shows up in the phrasing of every single draft it writes for you.
Think about what you'd actually hand a new AE before letting them anywhere near a live prospect: the product walkthrough, the pricing sheet with the actual discount floor written on it, the battle cards for the three competitors that come up in nearly every call, the objections your team hears every week and what's genuinely worked against them, and the part everyone forgets to write down, a clear definition of who you sell to, and more usefully, who you don't.
What's actually true: Treat it like a hire, not a feature. A new rep gets the deck, the pricing sheet, the battle cards, the objection handlers, and your ICP. Hand an AI SDR a 300-character tone-of-voice box instead, and it'll produce something fluent, confident, and wrong, because that's the entire universe of things you gave it permission to be right about.
Where it still needs you: Keep sends gated for human approval until you've earned the right to loosen that. Not because the drafts are bad, but because every wrong one is pointing directly at a gap in the knowledge base you didn't know you had. Read them. They're doing your onboarding audit for you.
3. “AI only automates the simple stuff”
Half right. The half that's wrong is the part worth sitting with, because it changed how I think about where the ROI actually lives.
Yes, it automates the obvious layer: drafting, chasing, logging, rescheduling. But the argument that this is all it's good for misses the actual point. Most qualification judgments aren't hard. They're uneconomic. Any decent SDR can read a careers page and tell you whether a company is hiring for the role that signals they need what you sell. None of them can do that for six thousand companies before lunch.
What's actually true: The case for running AI over every single lead isn't that the work is simple, it's that applying real judgment to 6,000 leads a year was never affordable with humans alone. Reading a careers page, weighing a title against company size, deciding whether a paragraph signals urgency, a person does every one of those better than a model, for maybe forty leads. Not three thousand.
Where it still needs you: Someone still has to define what a good lead actually looks like. An AI SDR will apply your definition at scale and say it with a completely straight face. Give it a vague definition and you'll get vague qualification, just faster, and at more volume than you had before, which is arguably worse.
4. “Buyers can tell, and they hate it”
Buyers can usually tell. Whether they mind depends almost entirely on whether the message was actually good.
I've never once seen a prospect complain about getting a fast, accurate, relevant answer to a question they asked at 9pm on a Tuesday. What generates the complaints is the other thing: a message that's obviously merged a company name into a template and stopped there, or a bot that answers a straight pricing question by asking to book a discovery call, because that's the only move it knows.
What's actually true: Nobody minds a fast, accurate, relevant reply. What people mind is feeling obviously processed, a message that name-drops their company and nothing else, or an assistant that can't answer the one question they actually asked. The objection has always been to bad automation. It was equally true of mail merge twenty years ago, and it'll be equally true of whatever comes after this.
Worth separating two situations, because they behave very differently in practice. Early-stage engagement (answering a question, confirming a fit, finding a time) is exactly where an immediate response beats a delayed human one, every time. A complex negotiation is where it doesn't, and no amount of model quality changes that math. Know which one you're in before you decide where to point this.
5. “This is a headcount reduction plan”
Sometimes it does get sold that way, and it's a mistake, honestly, it's the reason this category has a credibility problem it never needed to have.
Watch what actually changes in a rep's week instead of what a slide deck promises. They stop opening five browser tabs before a call, because the brief was already assembled. They stop manually re-routing a lead that landed on the wrong desk. They stop quietly losing the meeting that was cancelled on a Friday afternoon and never got rebooked.
What's actually true: The measurable effect on most teams I've talked to is that reps stop doing unpaid integration work. The tab-switching before a call. The manual re-routing. The chasing of a meeting that died somewhere between two tools that don't talk to each other. That was never really a job. It was the part of the job everyone already wanted off their plate.
The honest version of this pitch is genuinely unglamorous: your reps were never the bottleneck. The tools sitting between the lead and the rep were. That's a smaller, less headline-grabbing claim than replacing people. It's also one you can actually go check for yourself.
How to tell which one you're dealing with
If you're evaluating an AI SDR, there's one test that separates the serious tools from the demo-day ones. Take the hardest question a prospect asked your team this month, the pricing edge case, the competitor comparison, the integration that doesn't technically exist yet, and ask the tool the exact same thing.
If the answer comes back fluent and confidently wrong, you don't have a model problem. You have a tool that was never given anything to be right about in the first place. Ask what you can actually upload, how the knowledge gets stored, and what happens when the documents are silent on a question, the answers to those three questions will tell you more than any curated output sample a vendor puts in a deck.
And ask one more thing: what does it do with a reply it genuinely can't handle? The serious tools hand it over with the full context attached. The rest just keep talking.
How Lazure approaches this
The pattern underneath all five myths is the same one: the failures came from what the tool was given, not from what the tool fundamentally is.
Lazure's AI SDR gets briefed from an actual document library, not a text field. It ingests real product specs, pricing with the real discount floor, battle cards, objection handlers, your ICP definition, even your email swipe files, so it has the whole picture. It reads the entire lead record before it drafts a single line, including everything already said across both inbound and outbound, so it isn't guessing at context a human already has.
Sends stay gated for human approval until you decide to loosen that, and the autonomy level is something you set explicitly, not something buried three menus deep in a settings page. And when it hits a reply it genuinely can't handle, it hands it over to a person with the full context attached. It doesn't bluff its way through the next message just to keep the conversation moving.
The category isn't broken. Most of the deployments people point to when they say it is were just never given anything to be right about.
