People Analytics Isn't Dying. It's Being Decided.

Before I built people analytics teams I spent eight years in intelligence. That background is why this debate makes me twitch. Three essays are going around about the future of the field, and they agree on almost nothing except the diagnosis. Napper says people analytics collapses into one AI-native function he calls people intelligence. Sun and Nesbitt say it has to die and come back as work analytics. Starbuck says the value moves to the interpretive layer and the unit of analysis shifts from headcount to social capital. Three confident futures, each one sold as a reckoning.

They're all answering the wrong question, and in intelligence, answering the wrong question confidently is the classic way to miss the thing that actually mattered. The real question isn't whether people analytics ends, gets renamed, or gets reborn. It's who owns the workforce-intelligence mandate when the dust settles. When I look hard at the evidence, the answer is uncomfortable: it's being decided right now, the default is that HR loses it, and most teams are losing without noticing.

That's the argument. The rest is why.

The diagnosis is right, and it's old

The three essays are correct about one thing. The descriptive layer is collapsing toward the cost of inference. Claude in Excel builds in a minute what used to take a team a quarter. Once production is basically free, the bottleneck moves to selection: knowing what's worth doing, what to refuse, which question actually matters.

True. Also twenty years old. Cascio and Boudreau told us to speak the language of financial returns back in 2007. The descriptive-to-predictive-to-prescriptive curve has been conference wallpaper for a decade. Everyone already agrees the value sits above the artifact. Agreeing on the diagnosis isn't a position. Where the three essays go wrong is the prescription, and each one mistakes its own pitch for a forecast.

The default is dismemberment, not death

The part nobody wants to say at a people analytics conference: the most likely future isn't survival or rebirth. It's a quiet dismemberment.

I need to be precise here, because two things are happening and only one of them is a problem. Some commoditization is good. When the routine descriptive work becomes a baseline skill every HRBP has, the way typing stopped being a job and became something everyone does, that's maturity, not loss. Conversational BI is pushing this along fast. A leader types a question into the people-data lake and gets an answer in minutes, no analyst in the loop. Fine. Let that go. It was never the moat.

The problem is the high end, and who takes it. Dedicated practitioners peaked near 12,000 in late 2022 and have been shrinking since. When rates rose, people analytics roles dropped more than eight times harder than payroll roles, which is what happens to a function priced as a luxury. Of the people who left a dedicated seat after 2021, 83% never took another one; they went to HRBP, recruiting, generalist roles. Staying loyal to the title cost you money. That's the cleanest evidence in the whole debate, and it points one way. The cheerful counter-story, that the field is consolidating into an AI-native powerhouse, is mostly told by the companies selling the platform that story requires.

And the conversational BI that commoditizes the bottom cuts the other way at the top. If a leader can self-serve the answer, the translation work HR has been banking on erodes too, and Finance can pull people data without asking permission. The productive-capacity work that Sun and Nesbitt care about already has an owner, and it isn't us. FP&A holds the purse, and it's now running headcount as a live, driver-based model wired into the HRIS, the static tracker bypassed entirely. In high-volume industries, Operations built workforce planning before HR noticed it existed. When the prize is measuring what the work system costs and returns, the function that already speaks P&L is standing closer to it.

So the path of least resistance looks like this. Talent intelligence floats up and out. Productive capacity goes to Finance. Agent measurement goes to Engineering and Ops. What's left of "people analytics" is the harmless commodity layer, spread thin across HR. Nothing dies. It just gets divided up, and the part that mattered is the part that goes. The version I keep coming back to: what survives might not be a dedicated function at all. It might be an HRBP, a chatbot, and a platform.

The bar doesn't rise. The field barbells.

I want to push on the comforting version of my own side's story. "The bar rises" makes it sound like everyone gets pulled up to a higher baseline. That's not what cheap production does. It splits the field.

A thin tier doing real judgment and framing work pulls away. The broad remainder gets folded into generalist HR, where analytics is an assumed skill and analytics-as-a-job disappears. The middle thins out. The labor data already shows the shape of it.

The optimists have one good rebuttal: a dead job title isn't a dead capability. Typing stopped being a profession and became a baseline skill, and that was progress. Maybe analytics diffusing into every HRBP is the same thing.

But that analogy gives away the game. The typist job did vanish. It was fine because there was no high end to typing, no "typing strategy" that resisted commoditization. People analytics has a high end. The trouble is that high end doesn't diffuse to a generalist with a chatbot, and it doesn't stay put either. It either concentrates in a few hands or gets taken by Finance. That's the barbell. Their best objection describes it instead of dissolving it.

This is also why I'm careful about "it's literally what I do every day." That's not proof the panic is overblown. It's evidence I'm in the tier pulling away, not the median getting absorbed. The essays are describing the gap. I happen to be sitting at the far end of it. Both things are true, and pretending the bar rises evenly just flatters everyone who isn't.

The fight is the operating model, not the unit of analysis

The three essays burn most of their energy arguing the unit of analysis: social capital versus productive capacity versus none. That fight decides nothing, because the unit is downstream of whether you've already won a different fight.

The teams that keep a strong function don't win by picking the right denominator or the right name. They change what they are. They stop being a research group that studies psychology and become an analytics function that speaks finance. They hire businesspeople into HR. They run reviews that look like the CFO's QBR. They get into the comp, planning, and succession cycles before the budget is set, because an insight delivered after the decision is a dashboard with extra steps.

There's now hard evidence for why most teams can't do this, and it isn't a skills gap. A two-year ethnography inside a multinational found that analytics stays "loosely coupled" to real decisions through three moves. People endorse the dashboard and then decide the old way, so they're never on the hook. Other functions stall by fighting over who owns the data. And nobody can say whether the analytics team is a service, a change function, or an advisor, so it gets treated as none of them. These are political and identity failures, not modeling failures. Which tells you the real ranking of constraints. First, getting into the decision cycle. Second, fluency and standing. Third, data access. A distant fourth, proving ROI, which is what you get for solving the first three, not a lever you pull. I used to think access and standing were the whole game. Access is third.

As for the name: it follows the power, not the other way around. Rename a siloed reporting team to "people intelligence" without changing how it works and you've done nothing but repaint the thing finance cuts first in a downturn. Win the operating model and the reporting line and you can call it whatever you want. Lose them and no name saves you. The renaming debate is the deck chairs.

The two units aren't rivals. They're cause and ledger.

On the question the essays treat as the main event: social capital and productive capacity aren't competing futures. One explains, the other measures, and you need both.

Productive capacity is the ledger, what the work system costs and returns, and it gets the executive attention because it ties straight to the P&L. Social capital is the cause. It tells you why the number moved: where the connector everyone routes through is about to burn out, where heavy collaboration is buying innovation and paying for it in turnover. A network map that never cashes out into a cost-and-return number is a diagram nobody funds. A capacity number with no causal story is accounting that can't tell you what to change.

But the easy way to get the network data, passively scraping Slack and email and calendar exhaust, is fragile on both ends. Technically the models curdle; benchmark accuracy near 98% drops to 50-60% on real, fragmented HR data. Culturally it triggers revolt. When the Daily Telegraph put sensors under desks, staff forced them out in a day, and something like 81% of people analytics projects now stall or die on ethics and privacy. Scrape harder and you get surveillance theater, people performing for the metric, and the data is worse than useless. Worth being clear about what that is. It's a constraint, not an advantage. The same backlash hits whoever holds the pipes; Finance scraping the same exhaust gets the same revolt. Restraint is the cost of doing the work at all, not a moat anyone owns. The edge, if there is one, is in what you do with the data once you have it. That's the rest of this.

The moat is tradecraft, not judgment

This is the claim I changed my mind on, and the one easiest to get backwards.

The instinct is to say the edge is human judgment, the tacit stuff that never made it into a dashboard. I think that's the weakest version. Tacit expertise is more extractable than we like to admit; model an expert's actual decisions and you'll describe them better than they describe themselves. Worse, intuition is least reliable exactly when it matters most. In a real paradigm shift, the people who know the most have the most to unlearn, and the veteran is often the last to see the turn. Intuition is a fast lookup in a stable world, and this is not a stable world.

So when someone says "let the AI do the analysis, the human edge is the moat," I half agree, and the half matters. There are two things people lump together as analytical skill.

One is production: running the regression, cleaning the data, building the model, making the chart. That's gone. AI does it now, faster, without a PhD. Concede it and move on; defending it is how you lose.

The other is tradecraft, and it isn't production. It's the discipline of reasoning under uncertainty. Making the competing hypotheses fight instead of defending your first guess. Telling which evidence actually discriminates between explanations and which is consistent with all of them. Hunting for what would disconfirm you. Noticing the dog that didn't bark. Putting confidence in numbers instead of words like "possible" that every reader fills in with their own bias. Naming, in advance, what would change your mind. Auditing the model for the bias it inherited from history.

That last one isn't housekeeping. Train a model on data where men got promoted more, and it learns men succeed, then launders that into a clean-looking score. This is the legal third rail the optimists skip: disparate impact, masking, discrimination wearing the costume of objectivity. Calibration, bias auditing, and disconfirmation are what catch it. The rigor is the difference between an insight and a lawsuit.

Here's the part that makes tradecraft the moat instead of the casualty. The LLM is a satisficer. It grabs the most plausible story, confirms it, ignores what's missing, can't tell a diagnostic fact from a decorative one, and stays confident while it's wrong. That's the exact failure the intelligence-analysis world spent fifty years documenting in humans. AI didn't kill tradecraft. It industrialized the error tradecraft exists to catch. So the discipline is worth more now, not less. It also reclaims the "human edge" the optimists keep gesturing at. They're right that the edge is human. But empathy and judgment, as usually invoked, is the romantic version that already failed the test. Judgment without method is just confident intuition, and confident intuition is what the machine now mass-produces. Tradecraft is the human edge with the method put back in.

Rigor over the wall gets ignored

The field research caught me on something. Cold tradecraft, handed over as a black-box output, doesn't win. It gets routed around, rejected as numbers that miss the human context. The same ethnography that documented the loose coupling also documented what breaks it: analytics got coupled to real decisions only when it was recast in the language the decision-maker already runs on. Cost, risk, consequence. In that case the predictive models had stalled for years, then gained traction the week a new CEO reframed them as labor-cost governance. The methods didn't change. The framing did.

This part stings, because it cuts against the HR reflex. The framing that lands isn't warmth or human context. That's the relational logic, and the relational logic is exactly what loses in the room where budgets get set. What wins is the quantification logic: calibrated, falsifiable, denominated in dollars and risk. Heuer's discipline already points there, with its odds in numbers and its named failure conditions. The moat isn't the analysis and it isn't the bedside manner. It's rigor that arrives in the CFO's language at the moment she's deciding.

Tradecraft is a function, not a hero

The barbell worry is that all this leaves a few stars at the top and everyone else absorbed. It would, if tradecraft were a personal gift. It isn't. The whole contribution of the intelligence world was making it institutional, and that's the part HR never imported.

You make it a property of the team instead of the person with a few moves. Set standards a manager can check before the function puts its name on a judgment. Require, on the high-stakes calls, that the rejected alternatives get named and the reasons written down. Run review by someone outside the team, precisely because they don't share its blind spots. And run post-mortems against what actually happened, because without that loop nobody's model ever updates and "experience" becomes a polite word for entrenched error. Do that and the edge stops depending on whether you got lucky with a hire. It becomes how an ordinary team produces trustworthy work on purpose, which is also what separates a function that survives the barbell from a single person who does.

The comfort that finishes the job

When the ground gives way, there's a tempting place to land, and I'm naming it because I almost built this essay on it. The story goes: HR's real moat is trust. We're the ones employees confide in, so we'll own the voluntary, "zero-party" data nobody else can ethically get. It's the most dangerous idea in the field right now, and it fails four ways.

It's a vendor's optimism wearing a halo. The "recognition is the most human signal" line is sold by the company that sells recognition software, and the widely shared manifesto on ethical voluntary data was written by that vendor's CEO. Trading the surveillance pitch for the recognition pitch is still buying a pitch.

The voluntary data isn't clean either. People won't volunteer the bad news, because they believe, correctly, that it can cost them a promotion. Voluntary disclosure gets performed the same way surveillance gets gamed.

HR isn't the trusted party. Workers read HR as an agent of management, and they're right. Trust runs sideways, to peers, and real advocacy sits with unions and works councils, not the function holding the layoff list.

And even if all of that were false, trust doesn't convert. The binding constraint is standing in the room where budgets are set, and "employees confide in us" buys nothing there. In the one case I've seen documented well, stalled analytics finally won executive traction by stripping the relational justification out and reframing the whole thing as cost control.

So trust isn't the moat. It's the consolation prize, and an un-auditable one. A wrong attrition model can be caught and fixed. "Our value is empathy" can't be, which is exactly why it's where a scared function hides while Finance and Engineering split the mandate. Claiming it doesn't prevent the dismemberment. It is the dismemberment.

Governance is the same trap in a better suit. The EU AI Act mandates a human in the loop but never says that human is HR, and a person babysitting a black box they can't actually evaluate becomes the liability sponge: blamed for following a biased model, blamed for overriding it. Taking that seat without the tradecraft to use it isn't power. It's volunteering to take the fall.

How it loses, and the harder truth

A position that can't lose isn't a position, so here's the test. I'm wrong if, five years out, teams that only bought new tools and changed their name, without changing their reporting line, their finance fluency, or their place in the decision cycle, are thriving anyway. If cosmetic change turns out to be enough, I'm wrong. It's falsifiable, and I don't think it'll happen.

There are two ways to lose and they pull in opposite directions. One is to measure harder: double down on passive ONA and trigger the revolt, claim the governance mandate and eat the liability, push cold numbers over the wall and get ignored. The other, more seductive, is to retreat into comfort: decide the real value was trust and empathy all along and quietly cede the numbers to Finance. That one feels like wisdom and works like surrender.

But I have to be honest about something the optimists and I both glossed. When I went looking for proof that the window is still open, I mostly found the opposite. Finance already owns the P&L, the planning tools, and the line to the CEO. The "10% of teams transforming into a strategic powerhouse" is mostly a vendor line; the independent maturity data is going backward, not forward. And the identity barrier is deep enough that "just adopt the finance logic" is a little like telling a fish to walk. So I don't think the honest claim is that HR can still win the mandate. I think the mandate is mostly already lost.

What's left is narrower and I won't dress it up. It isn't owning workforce intelligence. It's becoming the check that keeps everyone else's version of it from blowing up. Finance models capacity as a cost line and misses the humans. Engineering measures throughput and misses the tail: the burnout in the connectors, the trust quietly draining, the skills rotting. The models they ship will encode old bias, fail on messy data, and walk into legal exposure the moment they touch a real hiring or firing decision. The only durable seat I can find for our discipline is the one that catches all of that, before it becomes a headline or a lawsuit. Not the author of the people-models. The auditor of them. That seat exists only because the failures are expensive and legally radioactive, and it's open only to people who actually have the tradecraft to sit in it.

The last thing, and maybe the worst, is that the tradecraft is learned in exactly the work that's vanishing. You build judgment by pulling the messy data yourself, building the bad model, sitting with the hard problem for an afternoon instead of taking the plausible answer in five seconds. That entry rung is the one this whole transition is sawing off. We could win the narrow seat and still starve the pipeline that produces anyone able to hold it. I don't have a clean answer to that one. It might be the most important sentence in here.

Where it lands

People analytics doesn't end and it doesn't get reborn. It gets divided, and most of it is already being handed out while we argue about names. There's no fortress here, no structural moat the field gets to sit behind. Take the comforting candidates one at a time, the irreplaceable judgment, the special read on people, the governance mandate, the trust of employees, and each one comes apart in your hands.

What's left isn't a moat. It's a discipline, and a closing door. Rigor the machine can't fake because it only ever satisfices. Spoken in the language of cost and risk. Delivered inside the decision instead of after it. Run as a team habit instead of a lone talent. And pointed, increasingly, not at producing the people-models but at being the one who can tell when they're wrong. That's a smaller job than the one the field thought it was getting. It's also the only one I can defend.

The cost of taking it is giving up the thing the field has used to feel safe: the idea that being trusted and humane is the same as being indispensable. It isn't. The default is dismemberment, and the only version of survival I can find takes more nerve than comfort. Stop defending the label. Stop hiding in empathy. Go be the check on the numbers everyone else is now producing, while there's still a door open to do it. Manage that, and the name was never the thing you were going to lose.