Our current human capital infrastructure is not serving employees well. That’s the opinion of Ben Zweig, CEO of workforce intelligence company Revelio Labs and author of ‘Job Architecture: Building a language for workforce intelligence’ – a book that explains how companies can transform human capital management through the power of taxonomies. In this interview, ai:sight unpacks the details of what this means.

Ben Zweig is one of today’s leading voices on the future of labor – and it’s easy to see why. His background has seen him experience the labor market through a variety of lenses: as a PhD student, where his research explored occupational transformation and social mobility; as a quantitative strategist at an emerging markets hedge fund; and as a managing data scientist at IBM. Today, he not only teaches data science and the future of work at NYU Stern, but he’s also the CEO of Revelio Labs, a workforce intelligence company that leverages the latest advances in AI to create a universal HR database from public sources.

Could you tell us about your background and the decision to write your new book, ‘Job Architecture: Building a language for workforce intelligence’?

My career began in economics – I did a PhD in economics and studied the labor market, focusing on social mobility and occupational transformation.

After that, I spent some time in finance, working as a quantitative strategist at a small hedge fund. Finance is interesting because it’s a very mature field where data is completely ubiquitous. Everyone has a Bloomberg Terminal where everything is categorized neatly – it’s easy to analyze.

Then I went to my first corporate job at IBM, where I eventually ran the workforce analytics group. That was very different. We were analyzing internal data, but it was a mess – nothing was categorized, everything was inconsistent, and cleaning and preparing the data for analysis was a considerable effort.

As I became more integrated into the people analytics community – people doing similar work at other companies – I found they were struggling with the same issues. They were also trying to analyze internal data to improve business performance, but the data wasn’t clean.

That’s why I started Revelio Labs. We’re striving to be the Bloomberg Terminal for labor markets – standardizing and structuring all this information so we can provide more clarity and help businesses improve their performance.

Tell us a bit more about job architecture and why it is so important.
In every large company, the HR departments are partly analytical. They’re trying to answer questions like: ‘Do we have the people we need?’, ‘Should we hire more of one type and fewer of another?’, ‘Should we promote more, or less?’ and ‘Should we source from different talent pools?’ Employees represent about 70% of costs at most companies, so these are huge decisions that need to be made rigorously and analytically. To do that, you need categories – you need to understand what types of employees you have. That’s where job architecture comes in.

A job architecture is a set of categories for employees – by occupation (what they do), by seniority (their level, like manager or director), by skills (what they bring to the job), and by work activities (the actual tasks they perform).

It’s helpful to define these terms clearly: A job is what you’re hired to do – a bundle of work activities. A cluster of similar jobs forms an occupation. Skills are the individual attributes people use to perform their work activities.

With the right categories – occupations, levels, skills, and activities – companies can start asking useful questions: ‘Where are we growing or shrinking?’, ‘What’s our attrition rate for key roles?’, ‘What’s the salary premium for a skill?’ and ‘When new technology arrives, what activities might it displace?’

All of this is only possible with a solid job architecture.

How is the difference between capital markets and labor markets particularly apparent when it comes to job architecture?

We all know what accountants do. Their job, at its core, is to categorize capital market data. They’re supported by systems like the Financial Accounting Standards Board (FASB) and Generally Accepted Accounting Principles (GAAP), which define how things should be categorized. But there’s no equivalent for labor markets. And because there’s no global standard, every company must invent this from scratch. The US government’s taxonomy – ONET – exists, but virtually no companies use it because it’s not well-suited for business applications.

So, companies either build their own job architectures or hire consulting firms to do it for them – which can cost millions of dollars. And because these systems aren’t maintained, they quickly become obsolete after mergers, technological changes or market shifts. Ultimately, we’re now in a place where every company is spending too much time and money on manual systems that don’t hold up – it’s not a good place to be.

What’s the biggest mistake organizations make when establishing a job architecture?

It’s thinking skills are the building blocks of jobs. All too often, companies invest heavily in so-called skills-based workforce planning – trying to manage everything purely through skills as the unit of analysis.

That’s a huge mistake, because skills aren’t the building blocks of jobs. With generative AI becoming so prominent, it’s especially clear how flawed that thinking is. AI doesn’t have ‘skills’. What it does is automate tasks. So, if a company’s entire workforce strategy is built around skills, it can’t adapt to technological change.

You mentioned AI is transforming how we work. In what other ways is AI proving transformational when it comes to job architecture?

It’s having an impact in two major ways – both equally powerful.

First, at Revelio Labs, our core technology is built on AI. Large language models (LLMs) are incredibly good at taking unstructured human capital data – job titles, skills, and work activities – and creating meaningful categories. They can cluster related roles, map skills, and identify similarities across jobs.

Second, once you have a strong job architecture, AI lets us observe how work is changing. When people ask me, ‘How many jobs can AI automate?’ I tell them that’s the wrong question. It’s not about how many jobs – it’s about how much work can be automated.

In truth, 100% of jobs can be partially automated. Some might have 10% of their tasks automated; others 50%. But there’s always a remainder – tasks that aren’t automatable yet.

As automation changes job composition, we need to understand how tasks reconfigure – which tasks co-occur and complement each other, which can stand alone, and what skill profiles we need for the new combinations of work.

By quantitatively measuring the changing mix of tasks and skills, we can design new, more efficient job structures. And it doesn’t take much to get real, tangible organizational benefits from that. 

For companies just starting to create their own architecture, what next steps do you recommend – and what benefits can they expect?

Honestly, most organizations need to start. They need to get a handle on how they categorize people – once you have structure, the other benefits follow naturally.

For example, with structure, you can bring in benchmarks. Once you know your categories – how they’re growing, attrition rates, etc., you can compare against peers.

At Revelio Labs, we bring in a lot of external data to show clients how their workforce compares. Are peers paying more? Are they recruiting from different places? Are they retaining people better?

You can learn what to emulate – or what mistakes to avoid. It’s about bringing in external data and developing real competitive intelligence.

And what about the employees themselves? What benefits can be realized at an individual level?

I think most people have a poor understanding of the occupational landscape they’re navigating.

We grow up hearing about a few familiar jobs – doctor, lawyer, firefighter, teacher – but there are countless others we never hear about, like program managers or DevOps engineers.

Having better data gives people visibility into what’s out there – what jobs exist, what paths people with similar backgrounds take, what satisfaction levels look like, and where work-life balance or promotion opportunities are better.

One last question: what can we expect from reading your book?

The book is split into three parts. First, the philosophy of work – what are jobs, tasks, and skills? How do they interact with technology? It’s a conceptual foundation.

Next is a bit of a ‘how-to’: how to build a job architecture, including the steps, clustering, labeling, maintenance, versioning, and managing change. It’s very practical.

Finally, we delve into what it all means – for organizations, for individuals, and for the economy. If we reached a world where job categorization was as efficient and scientific as capital markets, what could that unlock? That’s the fascinating part.

Ben’s book: ‘Job Architecture: Building a language for workforce intelligence’ is available to buy now.