AI in Recruiting: The 73% That Doesn’t Exist Yet
If you’ve spent any time in the recruitment world lately, you know AI is the most overused but least understood term in the business. Every vendor has suddenly become an AI company. Every demo promises transformation. Every pitch deck shows a future that feels perpetually six months away.
But here’s the problem: most of it doesn’t actually exist.
According to Fosway Group’s latest research, only 27% of AI features vendors claim are “live and usable” actually work in production. That means nearly three-quarters of what you’re being sold is roadmap, aspiration, or outright fiction.
Demos Are Easy, Integration Is the Beast
The chasm between what vendors promise and what they deliver comes down to two things: underestimation and noise.
Vendors underestimate the hard work required – data integration, governance frameworks, organizational change management. Building a demo is easy. Making it work across 16 different job boards, multiple ATSs, and disparate CRMs with real client data? That’s the challenge most can’t meet.
On the buyer side, there’s massive uncertainty about what actually creates value and where AI is still smoke and mirrors. When every vendor suddenly slaps “.ai” on their domain and calls their legacy chatbot “generative,” separating signal from slideware becomes nearly impossible.
“LLMs turned every vendor into an AI provider overnight,” says Sven Elbert, a senior analyst at Fosway Group who leads independent research in the talent acquisition space. “Our research wanted to find out who’s walking the talk. What we found is a few with very high ambitions who talk more than they walk, and others who’ve actually embraced it and are putting things out that work.”
Programmatic Ads and Recommendations: The Short List of What’s Live
Not everything is vaporware. Some AI applications in talent acquisition are delivering measurable impact today.
Programmatic job advertising is seeing real traction. Campaign optimization, A/B testing, automated media planning – AI accelerates and automates work that used to require constant manual intervention.
Recommendations at scale are also live and functional. Job suggestions, audience targeting, content personalization – these are happening in production environments with enterprise clients.
Basic automation around FAQs, interview scheduling, and candidate communications is working, though this often gets conflated with AI when it’s really just workflow automation. The distinction matters.
Where it’s still struggling: explainability, audit trails, bias controls. Only 5% of live AI features are designed primarily for hiring managers, which means vendors are still building mostly for recruiters and missing a massive opportunity to streamline intake and manager workflows.
How to Vet Vendors Before the Sales Demo
David Weinstock, Vice President of Talent Acquisition at New Story, doesn’t wait for vendors to tell him what works. He does the research first.
Before running a formal demo, Weinstock finds practitioners who’ve already bought the tool – through communities like RL 100 or the recruiting subreddit – and asks to see the product in real time, in real use, with real data. Then he runs the vendor’s scripted demo and compares what he’s being pitched against what actual users showed him. The gap between those two experiences tells him everything.
His evaluation criteria are straightforward:
- What’s actually live, not on a roadmap?
- How transparent is the vendor about what’s under the hood?
- What does the integration actually look like?
- Is it usable without heroics?
“My tech stack has to be fluid,” Weinstock says. “Everything I use – ChatGPT, Joveo, BrightHire, Ashby, Juicebox – works together seamlessly. No heroics required.”
When he implemented Joveo for programmatic advertising, cost per application dropped over 30%. That became the bar. Now he’s pushing for phase two: decreasing cost per hire. Then phase three: improving quality of hire. Each technology in the stack has different ROI metrics, and he holds vendors accountable for delivering on all of them over time.
Elbert offers two questions every buyer should ask:
“What happens when the AI is wrong? How can I see it and correct it?” This is especially critical in Europe with the EU AI Act, but transparency matters everywhere.
“Show me three live customers where AI features changed behavior and produced measurable outcomes.” If a vendor can walk you through the baseline, what changed, and what the measured result was, it’s worth continuing the conversation. If they can’t, it’s not.
Backend Complexity, Integration Hell, and Six Months of Compliance
KJ, founder and CEO of Joveo, is blunt about why most AI implementations fail: “Anybody can spin up a demo in a day. Seeing is no longer believing. Where the rubber meets the road is delivering – and delivering happens on two fronts: outcomes and deep integration.”
The backend work is brutal. You need multiple data sources. You need to marry those with LLMs. Different models work better for different use cases, so you’re constantly testing and optimizing which services to use where. Joveo uses Snowflake as its data infrastructure and has become a case study for how startups should build their entire tech stack around AI.
But data is only half the problem. Integration is the harder challenge.
“Every client is unique,” KJ explains. “Some have multiple ATSs, some have multiple CRMs, some have different brands across geographies. Larger staffing companies operate in different languages. You start thinking about multivariate systems and disparate technologies, and you realize you can’t build point-to-point integrations. It doesn’t scale.”
Joveo built its own intelligence layer that sits between the front end and back end, anticipating how systems interact and making the integration repeatable. “If it’s not repeatable, you’re not building best-in-class,” KJ says.
The company also invested six months in earning ISO 42001 certification for AI guardrails and compliance – possibly the only US recruiting tech company to have it. “The liability of one interaction gone wrong could be a PR disaster,” KJ warns. “There will be lawsuits. Smart people learn from other people’s mistakes.”
His benchmark for whether a company is truly an AI-first organization: if you bring them a new problem, they should be able to solve it in under three months. “If somebody says they’ll have it in three quarters, they’re not an AI company,” he says.
Intake Automation and AI Agents: What’s on the Horizon
AI agents are the next wave. Currently, only 10% of organizations in Europe are using AI agents in talent acquisition, but adoption is accelerating.
The more interesting shift is happening earlier in the funnel: intake automation. Some platforms are already using AI to run intake meetings with hiring managers, prepare interview questions, and shorten time-to-hire by seven to eight days – all before a job even gets posted.
“This is a persona that’s been ignored,” Elbert notes. “But there’s fundamentally cool stuff happening here that companies should be looking at.”
KJ’s roadmap goes deeper into what Joveo already does well – job advertising with automated geofencing, location expansion, and labor market dynamics baked in – while expanding into areas like AI-powered CMS (build an entire career site in under 30 minutes by showing a design and prompting the system), omnichannel campaign creation, and AI-driven CRM.
“CRM NPS scores have been negative for 20 years,” KJ observes. “The problem is recruiter adoption. People have to click buttons, change filters, do everything manually. Can AI handle the tasks? Campaign creation, timing, tone of voice, messaging – all done automatically based on best practices? That’s where intelligence comes in.”
AI in recruiting isn’t magic. It’s math. But when applied thoughtfully, it’s already transforming how companies find and hire talent.
The vendors who are winning are the ones solving real integration problems, delivering measurable outcomes, and building guardrails so their systems don’t hallucinate or create liability.
And if a vendor can’t show you three live customers with measurable results? Move on.
