Everybody Bought the Future. Who Made It Work?
Nobody has ever watched an HR tech demo go badly, which is sort of the point of a demo. The data is clean, the workflow is tidy and the hiring manager in the sample scenario responds to feedback requests within the hour (a detail that, on its own, should get the whole thing reclassified as speculative fiction).
Then the contract gets signed, and the product goes from the People function to the people who actually need the product to function. And that’s where the real mess usually begins.
That gap, between what looked enticing before the contract was actually signed and which feature actually exists when that contract is up for renewal, is the subject of my session at UNLEASH World in Paris this week, “Beyond the Demo: Spotting the HR Innovation Worth Betting On.”
Thousands of HR leaders will spend the next couple of days walking the expo floor at Porte de Versailles, and just about every one of them will be told, repeatedly and with tremendous conviction, that they’re looking at the future of work (which, let’s face it, probably involves a combination of prompt engineering and skilled trades).
A few of them actually will be something innovative (at least by the modest standards of our industry); the hard part is figuring out which ones are telling the truth, and which ones are selling a roadmap to nowhere.
None of this is an argument against buying technology. The opportunity is real (which is exactly why hype sells so damn well), and the session is about telling the two apart: how to recognize innovation that will change somebody’s working day, and what it takes to make that change stick after the vendor’s solutions engineer has flown home.
Ball Don’t Lie, but Averages Fib a Little
The foundational question every one of those attendees need answered is a simple one, but one that’s rarely asked: what did your last HR technology purchase actually improve?
I don’t mean how many seats got provisioned or what the vendor’s quarterly business review deck said. I mean what changed, for whom, and how anyone knows. If you need to pull up the vendor’s case study to answer, you have your answer. The business has never once asked HR for a more innovative procurement department, and a longer technology inventory makes a lousy business outcome.
Part of the problem is as simple as defining the problem. The OECD’s Oslo Manual, which is about as unglamorous as a source gets (and is a pretty reliable one for exactly that reason), defines innovation as a “new or improved product or process that differs significantly from what came before and has actually been made available to users or put into use.”
I want to underscore the last part of that clause, because it means innovation has to survive contact with work. For anyone writing the checks, I’d add one more test, which is, “demonstrates value.” By that standard, rebrands and acquisitions are press releases, and an “AI-powered” badge slapped on a legacy codebase is product marketing.
To be clear, the opportunity is real, and the best evidence for it is a lot sturdier than anything you’ll find on a booth banner this week. Economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied the staggered rollout of a generative AI assistant across 5,172 customer-support agents and found it raised issues resolved per hour by 15% on average, in a paper published in the Quarterly Journal of Economics last year.
The average hides the interesting part, though. Less experienced and lower-skilled agents improved the most, in both speed and quality, while the most experienced agents picked up small gains in speed and small declines in quality. Put plainly, the tool did its best work spreading what top performers already knew to the people who hadn’t learned it yet. It’s also one company, one well-defined task and a workflow with plenty of historical conversations to learn from.
HR work comes with policy exceptions, legal exposure and employees who are, understandably, having a bad day, so the study is solid proof that well-scoped AI can move a real number and a much shakier forecast for your HR service center.
Then there’s the number every global HR leader should remember when they’re walking the expo floor. The International Labour Organization estimates that one in four workers worldwide holds a job with some exposure to generative AI, a share that rises to 34% in high-income countries and falls to 11% in low-income ones.
Exposure measures what the technology could touch, which says nothing yet about what organizations will adopt and even less about jobs lost; the ILO itself concludes that because most occupations still require human input, transformation is the likeliest outcome. Global averages have never met your workforce, either.
An employer whose headcount sits mostly in frontline roles across lower-income markets faces a very different set of priorities than one with a headquarters full of knowledge workers in Paris and London, and a strategy built on the global headline describes nobody’s org chart in particular.
Don’t Believe the Hype
None of this requires a PhD or a procurement framework with a trademarked acronym. Something that’s truly “innovative,” at least in the context of HR Tech, should be able to answer five relatively straightforward and simple questions, which form the backbone of today’s presentation:
- What problem changes?
- What can people do differently as a result?
- What evidence supports the claim?
- What has to change locally to make it work?
- Who owns the result?
That last one is the hardest to answer, and if nobody in the room can, the accountability gap will outlast every feature on the roadmap.
Run those questions honestly and the most useful innovation often turns out to be decidedly unsexy: predictable scheduling for hourly workers, internal openings that employees can actually find and apply for, and employee service that resolves a case without three handoffs and a ticket number.
Interview technology belongs on that list too, when it captures what was actually said and ties it to the role’s explicit requirements, so hiring decisions rest on evidence instead of whoever had the most memorable handshake.
None of these will headline a keynote, but all of them show up in somebody’s working day.
The demo deserves the same sort of skepticism. Vendors write demo scripts specifically engineered to succeed, so bring a few difficult employees of your own: conflicting policies, missing records, a request so ambiguous that two people on your own HR team would answer it differently.
And “global” needs to mean more than a language menu. Local policy fit, accessibility, device access for frontline workers on shared or personal phones and where employee data actually lives all decide whether a tool works in Marseille as well as it did in the sandbox environment back at headquarters.
In several European countries, works councils and employee representatives get consulted before certain workplace technologies go live, which you can treat as a free stress test of whether your rollout plan holds up to scrutiny. The model that works is shared outcome standards across the business, with local differences allowed whenever there’s a genuine reason for them.
The Room Where It Happens
Adoption doesn’t get decided by vendors; it gets decided by end users; somewhere between a manager’s third back-to-back meeting and an employee’s lunch break, and no steering committee has ever actually been in that room. That’s why incentives and training drive usage far more than mandates and monitoring.
A talent marketplace looks fantastic in a board presentation, but mostly underdelivers when managers block internal transfers to protect their own headcount, which, given how most of them are measured and rewarded, is a perfectly rational thing for them to do. Fix the incentive or make peace with the outcome. Either way, measure successful task completion and sustained use, since logins only tell you who showed up, not who actually leveraged the technology to drive incremental improvements or tangibl
The nausea inducing, but ubiquitous, “human in the loop” deserves similar scrutiny, because it has become the phrase vendors reach for whenever someone asks what happens when the system gets it wrong.
A human in the loop needs a job description: who reviews which decisions, what evidence they see when they do, when they can intervene, who has the authority to pause the system, and how an employee can challenge an outcome and get an actual answer; European lawmakers have landed in roughly the same place.
The EU AI Act classifies AI used in recruitment, promotion, termination, task allocation and performance monitoring as high-risk, and its rules for organizations deploying those systems require human oversight by people with the competence, training and authority to exercise it (obligations currently slated to apply from December 2027). That’s the law asking, more or less, for a job description.
Finally, time saved needs to be spent, well, somewhere else. If you’re a recruiter, not getting back to candidates probably isn’t the best answer. Instead, separate the cash savings your CFO can actually book from the capacity your teams can redeploy, since nobody has ever made payroll with “capacity.”
Count the full operating cost, including integration, administration, change management and the ongoing review that responsible use requires, and set pilot success criteria before launch instead of after the renewal conversation starts.
Measure not just speed, but also quality and the fallout for employees. Hours saved might look impressive on a vendor’s sales deck, but unless you can prove your team is doing something genuinely valuable with those hours, there is no real business case. And remember: time savings alone is not ROI—claims that it is should raise eyebrows, because it’s what you actually achieve with that time that matters.
Speaking of, if you made it this far, you should probably get back to work.