Why "we already tried AI calling" isn't the same as "it doesn't work"

We talk to hundreds of recruitment agencies every month, and at least one out of ten tells us the same thing: they already tried AI calling, and it didn't work. That's a real, recurring number from our own calls, not a vague impression, and it's the tension this whole piece is built around.

It sits oddly next to how fast this market is growing. India's conversational AI market was worth an estimated $653 million in 2025 and is projected to grow at a compound annual rate of about 25.6% through 2034, according to IMARC Group. The AI recruitment segment specifically is on a similar path, from roughly $43 million in 2024 toward an estimated $90 million by 2035, per Market Research Future.

More agencies are trying AI calling every month, and a steady one in ten of the ones we talk to have already tried it and walked away.

Those two facts aren't actually in tension. A market growing this fast is full of new vendors, some built well and some built in a hurry, and agencies are the ones finding out the difference the hard way. What surprised us more than the objection itself was that when we asked what "didn't work" meant, we got a different answer almost every time.

What do agencies mean when they say AI calling didn't work for them?

Rarely the same thing twice. Some names have been left out or generalized below to keep this fair to the vendors involved, since we only heard one side of each story, but the reasons are real and came directly from agencies we spoke with this year.

One agency owner told us flatly: "It has wasted my money. I am not a big fan of AI. It's not that I'm against it, it's just that it has wasted my money." No elaboration on what specifically went wrong, just a clear conclusion that the money hadn't come back as results.

Another agency, when asked what he thought of AI calling, said the pricing never made sense against what his own recruiters already cost. He'd been quoted something in the range of ₹16 a minute by a previous vendor, and did the math himself: his recruiters earn a fixed monthly salary regardless of how many calls they make, so at that price, a machine that charges by the minute lost the comparison before it even started. That's a completely reasonable thing to conclude, and it says more about that vendor's pricing than about whether AI calling can ever pencil out.

At a materially lower per-minute rate, and once you count everything a human recruiter's time costs (redials, no-shows, WhatsApp back-and-forth, the hours spent on calls that never connect), the math looks different. See what your recruiters' screening calls are costing you for that fuller comparison. But he'd only ever seen the expensive version.

A third recurring complaint was about the product itself, not the price. More than one agency described a previous AI calling tool as feeling like a general-purpose voice assistant with a recruitment script taped on, not something built for screening candidates specifically. It could hold a conversation, but it didn't know what to do with a candidate who asked an unusual question, couldn't populate a shortlist the way a recruiter would, and treated every call like a generic support call rather than a job screening.

One agency owner running a large, established staffing operation put it simply when we finally got him on a call about our own product: "actually, I tried it last year, I won't say it was terrible, but it just made no sense for me to go ahead with it." He agreed to a two-week free trial precisely because he'd been burned by a vendor's demo before and didn't trust another pitch without seeing it run on real calls first.

And one more agency, cold-called about a completely unrelated staffing role, told us they'd already tried an AI calling tool: "they'd already tried it, but it didn't work well, and now they're not interested." No specifics volunteered, no follow-up wanted, no interest in hearing about a different vendor.

Do candidates hang up because the voice sounds like a machine?

Less than most agencies assume. A screening call isn't a sales call. Nobody is trying to extract money from the candidate or push them into a decision they weren't already leaning toward. It's a call about a job they applied for or already said they're interested in, so staying on the line and answering carefully is in the candidate's own interest, not something they need to be talked into.

A candidate who wants the outcome of the call, an interview, a shortlist spot, a next step, has a reason to engage even when they can tell they're talking to an AI. What breaks a call isn't the voice being artificial. It's the voice getting things wrong: mishearing a name, misunderstanding an answer, asking the same question twice because it didn't register the first response.

That's an accuracy problem, not a robotic-voice problem, and it's exactly the kind of thing a vendor with a scattered, unfocused build tends to get wrong more often than one that's tuned specifically for recruitment conversations, accents, and vocabulary. See what breaks in an AI screening call once you look past the pitch for the specific ways accuracy gaps show up.

Does a bad first experience mean an agency has written off AI calling for good?

Less often than you'd expect. Across almost every one of these conversations, the same agency owners who describe a bad experience will, in the next breath, say they think this is where the industry is going. The complaint is about a specific vendor, a specific price, or a specific product gap. It isn't usually "I don't believe machines can screen candidates." It's closer to "the one I tried wasn't good enough to keep using."

That's the difference between an agency that's closed to the category and one that's just waiting to see a better version of it. The agency owner above who agreed to a free trial is a good example: skeptical of another pitch, completely open to judging the product on its own results once he could hear it working.

Why do so many AI calling vendors get recruitment wrong in the first place?

Because for a lot of them, recruitment is one of ten or twenty use cases they sell voice AI into, not the thing they exist to solve. A vendor selling the same underlying voice bot into debt collection, appointment reminders, customer support, and recruitment is optimizing for a generic conversation engine, not for the specific judgment calls a screening call requires: knowing which answer disqualifies a candidate, which one needs a human's attention, and which follow-up question matters for the role being filled.

Spread that thin across a dozen unrelated use cases, and recruitment gets the same shallow treatment as everything else on the list.

A platform built only for recruitment voice screening doesn't have that problem, because it isn't solving anything else. Every decision in how the call is built, scripted, and routed is made with one job in mind.

Is the product being sold the actual point, or is it the outcome?

This is where a lot of vendors and buyers talk past each other. A vendor selling voice AI as a product hands over a tool and expects you to configure it into your process. That's selling process. What most agencies want isn't a tool, it's the outcome: more candidates screened, fewer stalled in the pipeline, a shortlist that shows up on time.

That's the difference in how we approach a new account. Before we propose anything, our team sits down on a discovery call to understand what the agency or business is trying to achieve, not to pitch a fixed feature set. We don't sell the same solution to everyone who calls.

We've pointed the same underlying platform at very different goals depending on the account:

• Reactivating a large pool of candidates who'd already said yes to a role but stalled before finishing their application (see how we approached that stalled-candidate problem for a high-volume hiring pipeline)

• Helping a small, growing academy with no dedicated hiring function screen candidates for occasional bursts of open roles, instead of the academy going without anyone recruiting at all (see that story here)

Neither one started from "here's our software, go use it." That's the difference between a software seller and a co-builder: a software seller hands over the tool and leaves you to make it work, a co-builder sits with you until the actual outcome shows up.

What should an agency check before trying AI calling again?

Four things came up repeatedly in what agencies told us about their last vendor. Check these before you sign up with anyone a second time.

Whether the pricing holds up against your actual recruiter cost, not just against a headline per-minute number. A rate that looks expensive in isolation often isn't, once you count the idle time, redials, and coordination overhead a human recruiter's salary already absorbs.

Whether recruitment is the vendor's whole business or one of many use cases. A team building a single, focused recruitment voice platform will have made a hundred small decisions specifically for screening calls that a general voice AI seller never had a reason to make.

Whether anyone sat down to understand your actual hiring problem before proposing a solution. If the pitch starts with a feature list instead of a question about what you're trying to achieve, that's a sign you're being sold a product, not a partner.

Whether you can start with a real pilot instead of a full commitment. A vendor confident in their own accuracy will let you test it on a real, limited batch of candidates first, so a bad fit costs you a small pilot instead of a signed annual contract you have to unwind later.

If your agency's last AI calling attempt ended in wasted spend, a confusing price, or a tool that never felt built for recruitment, send Recruit Bud one role and five candidate numbers. We'll start with a pilot, not a full rollout, so you can hear the actual call recordings and judge for yourself whether this one holds up where the last one didn't.

Frequently Asked Questions

Why do so many recruitment agencies say they already tried AI calling and it didn't work?

Because a lot of them genuinely did try it, with a vendor that didn't hold up. The reasons vary from agency to agency: money spent with nothing to show for it, per-minute pricing so high it never made sense against a recruiter's cost, a tool that mispronounced names and mangled answers because the speech recognition wasn't tuned for the accents it was calling, or a product that was really a generic voice assistant with a recruitment script bolted on. The common thread isn't that AI calling failed as an idea. It's that a specific vendor, usually one selling voice AI as one of many use cases rather than a recruitment platform built around this one job, failed a specific agency.

If an agency's last AI calling vendor didn't work, does that mean the category doesn't work?

Not on its own. Vendors in this space differ a lot in call quality, pricing model, and whether they were built for recruitment or just repurposed from general customer-support voice bots. A bad result from one vendor is a fact about that vendor. Most agencies who've had a bad first experience still say, in the same conversation, that they think AI calling is where the industry is headed. That's a very different statement from "it doesn't work."

Do candidates stay on the line for an AI screening call, or do they hang up because it sounds robotic?

Candidates generally stay engaged, more than most agencies expect, because a screening call works in the candidate's own interest. It's not a sales call trying to extract money or a decision from them. It's a call about a job they already applied for or expressed interest in, so they have a reason to answer questions carefully rather than brush the call off. Where calls do underperform, the cause is almost always accuracy, a voice that mishears an answer or mispronounces a name enough times to break the candidate's confidence in the call, not the fact that it's a machine talking.

What should an agency check before trying an AI calling vendor a second time?

Four things that came up repeatedly in agencies' own accounts of what went wrong before: whether the per-minute or per-call pricing beats what a recruiter costs once you count their idle time and redial attempts, whether the vendor was built specifically for recruitment screening or sells voice AI across a dozen unrelated use cases with recruitment as just one of them, whether the team sits down to understand the actual hiring outcome you're after before proposing a solution, and whether you can start with a real pilot instead of committing to a full deployment upfront.