People Analytics is not Dead

I’ve lost track of the number of times people have asked me if I think that AI is going to mean that People Analytics is over.

No.

The work will definitely change – historically, the work involved a lot of code writing, either to run analyses or to build tooling that took advantage of the research. That technical work was gated by the capability and capacity of the experts on the team. I’ve worked on projects with heavy technology components that took months or years and those would now be doable in weeks at most. That’s amazing.

Frequently, when people ask about the end of People Analytics, they point to the ability to explore data or build dashboards with basically zero friction. They aren’t wrong about the friction.

At least 3 issues mean that they are wrong about People Analytics no longer being relevant though:

1) Privacy and confidentiality requirements mean the most interesting data isn’t available broadly. Attrition prediction is almost universally the first ask of a People Analytics team (loss aversion is strong!). And now I’m seeing attrition prediction as an early request for citizen data scientists trying to self-solve for People Analytics.

The problem is that far and away the best predictor for exits across several organizations where I’ve been responsible for that work is each employee’s response to an “intent to stay” question. Turns out people are pretty honest about whether they plan to leave – as long as they are confident that their answers will be kept confidential. Sure, there are other variables you can use, but they are much weaker predictors.

The same thing will be true for lots of the kinds of questions that People Analytics teams tackle – the underlying data have pretty hard constraints about who can actually see and interact with them. Performance review data, pay data, sentiment data – these are held tightly, and for good reason. A People Analytics team is a crucial ethical moat that protects employees while also supporting the organization through helping leaders make informed decisions based on aggregated, privacy-protected insights.

2) Defining useful questions and identifying appropriate analyses is harder than it sounds. One of the biggest values that a good People Analytics professionals – or analysts generally for that matter – bring to an organization is the ability to scope and build a research question. Getting to an underlying question is hard. Knowledge on what data are relevant and available is surprisingly scarce. Expertise in which statistical technique is appropriate to the question at hand is surprisingly hard. Even something as simple as counting or calculating a rate can be hard – the choice of denominator can have profound implications on the inferences you make from the analysis.

This is a big enough point that a couple of examples are in order. Let’s look at a seemingly simple question about cancer rates. Are they rising, flat or falling?

Yes. Depending on how you ask the question.

Obviously, they are rising. Cases have been rising for many years! If you simply count all cases, this is true.

Obviously, they are flat. The global population is rising, so you’d expect occurrence of pretty much any human experience to also be increasing. So, we obviously need to diving the total number of cases by the global population. And cancer is pretty flat as a cause of death.

Obviously, the are falling. Cancer is also more common in older people, and life expectancies have been increasing. So, if you normalize the data by age, you can see very plainly that cancer as a cause of death has been falling.

Which of those answers is correct is not about the math – it’s about the question you are asking. If you are interested in public health, you are likely most focused on the 3^rd approach – cancer rates normalized by age. If, however, you are a company that sells chemotherapy drugs, you are much more interested in the total rates – that first calculation. None of the answers are wrong – the challenge is in figuring out which frame is appropriate.

Let’s look at another example. Imagine an organization is driving hard at making their organization more technical, and is trying to increase the proportion of engineers in its workforce. They put in place a lot of recruiting efforts to attract and hire highly qualified engineers into roles, as well as working to retain the ones they already have. The prevalence of new AI tools means functions like legal and HR are getting technical talent, when they didn’t before. Each individual leader sees their representation rise throughout the year as a result of their efforts. Celebration ensues.

But, at the end of the year, we learn that the company proportion of tech talent overall has fallen. How is this possible? Simpson’s paradox examines how subgroup trends can disappear or reverse when results are aggregated. In this example, at the same time that individual functions were increasing their efforts to increase technical talent, the company made a huge shift in its overall workforce, expanding functions like marketing, legal, government relations, while the technical functions stopped hiring junior engineers and quietly eliminated some less technical roles like product managers, leading to a modest reduction in the technical function overall. Although non-technical functions increased their hiring of engineers, and the engineering functions evolved to have a higher percentage of technical staff due to the removal some of their product managers, the mix shift meant the overall technical to non-technical ratio moved in a non-technical direction – exactly opposite what leaders intended, and opposite what each leader individually was pursuing.

Which means that both things were true – each individual leader did increase the extent to which their teams were technical. And the company overall did end the year less technical than it started. Figuring out which answer to give is highly dependent on what problem you are trying to solve. Without good business and data understanding, you can accidentally be exactly wrong in the results and insights you deliver.

3) Role of research and metrics in creating shared understanding. Think of the cancer rate and representation examples above. People Analytics exists to help organizations make better people decisions. So, one of the important jobs of a people analytics professional or team is creating a shared reality to inform decision making.

When each stakeholder is left to their own devices to explore data, not only are they taking time away from their day jobs and likely using the least relevant data for their analyses, they are all creating different versions of reality. They are essentially living in the blind men and the elephant problem. Each only understands the problem through their specific filters.

That not only means is there risk that they’ll settle on less effective approaches to their problems, it also means that they are likely to waste a good deal of time arguing over whose version of reality is right. Energy that should be productively directed at building and scaling solutions instead gets squandered in quibbles over defining the problem.

There is absolutely a place for citizen data science in the People space. And new tools will accelerate the work of People Analytics, and expand the universe of useful insights, tools and solutions we can offer.

But vibe coding and natural language based queries won’t be enough. Even with all that AI can do, there remains a critical role for an enterprise-quality analytics platform – something like One Model that can manage privacy, confidentiality and data access in an effective way. Something that can support exploration, but keeps queries on the rails of good data. Something that can focus attention on key priorities and facilitate a shared understanding of where critical gaps and differentiating upside opportunities lie.

People Analytics is more important than ever – and we have more powerful tools and more data than we’ve ever had before. Building thoughtful ecosystems that empower managers, leaders, HR professionals and People Analytics professionals to gain detailed insight is transformative – but doing so responsibly and effectively requires more than an LLM license.