AI in HR – Tracing the Incentives

THIS WEEK’S NUMBER: 77%

That’s how much of last quarter’s global AI funding went to just 63 companies.

If you’re buying HR AI outside that group, you’re betting on their survival, not your outcomes.

AI in HR is now an obligation rather than a strategy. Everyone feels pressure to say something. Far fewer can show anything. From Canada, the pattern looks even clearer. The winners are obvious. The value is not.

CB Insights’ Q3 2025 State of AI report found that 77% of private AI funding in one quarter – $37 billion of $47.8 billion – went to just 63 companies raising $100 million or more. Capital is piling into a few likely winners while everyone else fights for oxygen. If your vendor is outside that circle, you are betting on their survival, not your outcomes.

A Checkr survey of 3,000 HR managers found that most CHROs still can’t prove meaningful returns on their AI investment. The culprits: bad data and internal resistance. Neither of those is a technology problem.  (Checkr sells background screening and hiring software – keep that in mind when reading their findings on AI ROI.)

Greenhouse surveyed more than 4,100 job seekers, recruiters, and hiring managers. Hiring managers trust AI to make faster, better decisions. Only 8% of job seekers think it makes the process fair. Nearly half say their trust in hiring has dropped over the past year. Forty-two percent point directly at AI.  (Greenhouse sells hiring software. They have a stake in how this conversation goes.)

Connect those dots. Big money’s consolidating. Buyers can’t prove the tools work. The people subjected to them think the system’s rigged. That’s not a market that’s failing. That’s a market whose value proposition has already collapsed. The only thing left is for the invoices to catch up with the evidence.

MIT’s NANDA initiative released a report in July 2025 based on 150 interviews with leaders, a survey of 350 employees, and an analysis of 300 public AI deployments, finding that 95% of corporate generative AI pilots fail to deliver measurable business value. Ninety-five percent.

The RAND Corporation found that more than 80% of AI projects fail – twice the failure rate of non-AI technology projects. Your ERP implementation had better odds. We’re worse at this than almost every other technology deployment in the enterprise, and we’re doing it faster.

And the people who built the loudest case for urgency are now walking it back. Sam Altman told a Commonwealth Bank of Australia audience on May 26 that the ‘jobs apocalypse’ is ‘not coming’ – admitting OpenAI’s predictions were ‘pretty wrong.’ Dario Amodei shifted two weeks earlier, arguing onstage with Jamie Dimon that automating 90% of a job just expands the remaining 10%. These are the same people warning whoses sent HR budgets into panic mode. Both companies are heading toward IPOs. They need you confident, not scared. The fear was real. The evidence behind it apparently wasn’t.

What the Research Actually Says

A 2026 scoping review published in  Administrative Sciences  synthesized 33 studies and found that AI-based recruitment improves speed and consistency, especially in resume screening, candidate ranking, and automated communication. When the task is bounded and the criteria are clear, AI outperforms humans on speed and can match or exceed them on consistency.

A separate 2025 systematic review in  SAGE Open  reviewed 502 articles and confirmed that most AI applications in recruitment cluster around assessment, selection, and application processing. The efficiency gains in high-volume, structured hiring tasks are well-supported.

Notice the words that keep appearing. Structured. Bounded. Clear criteria. The research doesn’t say AI is good at judging character, potential, or culture fit. It says AI is good at fast, consistent, rules-based screening. That’s real value. It’s also a much smaller promise than what most vendors are selling.

A 2026 systematic review in  Sustainability  looked at 19 studies and found the same problems surfacing repeatedly: polarization, exclusion, and discriminatory outcomes when the data is biased, or the system isn’t transparent. A separate 2026 scoping review in  F1000Research  landed in the same place from a different angle. The issue isn’t the technology. It’s organizations deploying it without guardrails.

The same  Administrative Sciences  review that found efficiency gains also found this: candidates disengage when the process feels opaque or when the human disappears. They drop out, reject offers, and talk about it. The tool you bought to improve hiring can end up damaging the hiring process itself.

A 2025 systematic review in  Management Review Quarterly  examined 43 studies on AI and DEI. AI can make evaluation more consistent. It can also learn from old biases and scale them. Amazon is still the clearest example of that.

AI can help with narrow, structured parts of the hiring process. It can also create bias, erode trust, and collapse without strong human oversight. The research doesn’t resolve that tension. It confirms it.

The Canadian Picture

Eighty-one percent of Canadian HR professionals use AI tools at work – more than any other function tracked.  (Growclass/Angus Reid, September 2025 – Growclass sells AI training courses. Draw your own conclusions.)  Meanwhile, 63% of Canadian job seekers worry AI will significantly limit job opportunities. The people doing the hiring and the people being hired are anxious about the same technology. Nobody in this picture looks confident.

An Express Employment Professionals–Harris Poll from October 2025 found that 54% of Canadian hiring managers say their company uses AI, but 59% admit they don’t have the resources or training to help employees use it effectively. Tools are deployed. People aren’t ready. And a lot of them are worried the tools are coming for their jobs next.  (Express Employment Professionals is a staffing firm. Slow AI adoption is good for their business.)

A Capterra survey found that 67% of Canadian organizations have AI features in their HR software, but only 41% use them. The features are there. The adoption isn’t. Canadian companies are paying for AI they’re barely touching.  (Capterra sells software comparison and review services – their data reflects the market they profit from.)

Who Actually Gained

The platform incumbents.  Workday. Microsoft. LinkedIn. They didn’t have to invent the future. They just had to wait while you chased individual tools, got burned, and came back to the bundle. Workday’s $1 billion acquisition of Paradox, the leading conversational AI hiring platform – made the direction of travel obvious. The big players are getting bigger while everyone else is still comparing pricing tiers.

The VCs.  They made their bets when the story was cheap. Now every portfolio company press release helps reinforce the narrative. Even if most of those companies are gone in two years, the firms behind them may’ve already marked up the position, raised the next fund, and moved on.

And the consultants selling “AI readiness.”  Mercer’s 2025 report – and Mercer sells transformation services, so take that for what it’s worth – says 68% of AI investments fail because the basics aren’t in place. Who gets the call when the basics aren’t in place? Usually, the same people who sold the readiness assessment. Gartner puts a number on what happens next: through 2026, organizations will abandon 60% of AI projects that lack AI-ready data.

What to do with this

The HR AI market was valued somewhere between $6.99 billion and $8.5 billion in 2025, depending on whose estimate you use. Market size measures spend, not outcomes. A year later, the spending still looks a lot more like fear than evidence.

ISG’s 2025 State of HR Technology report found that AI budgets for HR will average $1.6 million in 2026 – a tenfold increase since 2023. Only 52% of organizations see quantifiable value from HR technology. The money’s committed. Whether it buys anything that lasts is a different question.  (ISG sells HR technology advisory services. More AI means more work for them.)

Here’s what the evidence tells you to do.

Kill any pilot that can’t name the workflow step it removes in one sentence. Not “enhances.” Not “streamlines.” Removes. If it adds AI to a process that was already broken, you bought a faster way to do the wrong thing.

Ask every vendor three questions and don’t let them leave until they answer with a number. What proprietary data are you training on that nobody else can access? If Workday or Microsoft builds this natively and gives it away, why do you still exist? Show me one client who removed a human step – not added your tool – and tell me the dollar amount they saved. If the answer starts with “great question” and ends with nothing, you’re looking at a wrapper with a marketing budget.

Let them beta-test the mess.

Sandra Stuart