Few labour markets are transparent, but there is one where it is possible to get a complete picture - academia. When a university department appoints a professor, every step of the move is public: where the person earned their doctorate, and where they were hired. No confidentiality agreements, no survivorship bias, no sampling. In 2015 Aaron Clauset, Samuel Arbesman and Daniel Larremore hand-assembled this record for nearly 19,000 faculty across three very different disciplines - computer science, business, history - and drew the picture nobody had seen: the complete map of who hires from whom.
The picture confirmed what many of us would expect. Faculty hiring turned out to be steeply, almost feudally hierarchical. Roughly a quarter of institutions produced the large majority of all professors. The prestige ranking hidden in the flows predicted where a graduate would end up better than any published league table. And the flows ran overwhelmingly downhill: the typical graduate placed at an institution ranked well below the one that trained them, while moves up the hierarchy were rare. A follow-up study by Hall Wapman, Sam Zhang, Clauset and Larremore in 2022 extended the picture to essentially all of American academia - about 300,000 professors - and found the same shape everywhere: one fifth of departments supply four fifths of the faculty, and in most disciplines a handful of elite departments seed the entire field.
What is interesting, at least to me, is that nobody designed this hierarchy. There is no committee that decides Berkeley outranks a regional university (though league tables might, one could argue, contribute). The ranking is not written down anywhere the hiring departments can consult. It emerges - from hundreds of thousands of individual, locally sensible decisions - and it is steeper, more stable and more consequential than anything a committee would dare to design.
Academia is unusual only in being visible. Every labour market has this structure. The one in your organisation’s market sector does. Your firm has a position in it - an address - and that address was determining the quality of your candidate pool long before anyone in the building started calling things “talent strategy”. This issue is about learning to see the map.
HR language is full of water. Talent pools, candidate pipelines, talent streams, hiring funnels. The image behind all of it is a reservoir: a homogeneous mass of candidates from which a firm with a good enough brand and a fast enough process can draw at will.
The research of the last decade describes something quite different. Picture instead a dot for every firm in the economy. Now draw an arrow between two dots every time a person moves jobs from one firm to the other. Network researchers often call the dots nodes and the arrows edges. Because each arrow has a direction - from the old employer to the new one - the whole object is a directed network. This particular type has acquired a name: the labour flow network.
The idea has a long history. Alfred Marshall, writing in 1890, described industrial districts where “the mysteries of the trade become no mysteries; but are as it were in the air” - his explanation ran on dense local movements of workers between firms. AnnaLee Saxenian’s 1994 book Regional Advantage told the modern version as a natural experiment. Silicon Valley and Boston’s Route 128 entered the 1970s with the same industries, comparable firms and equivalent university feeders. The Valley developed a culture of relentless job-hopping - helped, materially, by California’s refusal to enforce non-compete agreements - while Route 128’s proud, vertically integrated firms prized loyalty and punished departure. The same type of nodes, but with very different edges. The dense network won - ideas recombined faster, failed ventures returned their people to circulation instead of stranding them, and the Valley pulled decisively ahead. The topology of the network, not the quality of the firms, decided the outcome.
In an important 2013 paper, Omar Guerrero and Robert Axtell built the first labour flow network for a whole economy. By combining, as they put it, “the emerging science of networks with newly available employment micro-data, comprehensive at the level of whole countries”, they were “able to broadly characterize the process through which workers move between firms.” It was nothing like the matching-market of the economics textbook: a few firms are enormous hubs of churn, most connect to only a few others, and well-worn “highways” of repeated movement link particular pairs of firms year after year. Then in 2019 Jaehyuk Park, Ian Wood, Elise Jing and Yong-Yeol Ahn did it for the planet, using the employment histories of 500 million LinkedIn members - 130 million job moves across more than four million firms. The world economy, seen this way, organises itself into nested clusters that are simultaneously geographic and industrial, and movement between clusters is rare compared with movement within them. Whether or not you have ever thought about which cluster your firm belongs to, your firm is in one, and that is where nearly all of your candidates, and nearly all of your leavers’ destinations, sit.
The network provides a clear message: a firm does not draw from a pool. It occupies a position in a structured network, and the position fixes what it can reach. The word “ecosystem”, which Peter Cappelliand JR Keller used in their 2014 review to describe modern talent management, is the right one - and the point of an ecosystem is that you are in it whether you manage it or not.
A recurring theme of this newsletter is that the interesting truths about workforces tend to get discovered several times, by fields that do not reference each other - most recently in Issue 13, where five fields proved the same impossibility result in five sealed rooms. Labour flow networks are no different: at least four disciplines arrived at essentially the same object but, for different reasons, in mutual near-silence.
Complexity science mapped it. Guerrero, Axtell, Park and colleagues - physicists and computational social scientists by training - treated the economy as a graph and characterised its shape, its clusters, its dynamics. Their tools came from studying the internet and protein interactions, not from labour economics.
Economics valued it. We will come to Isaac Sorkin’s remarkable result shortly however the short version is that economists realised the direction of the arrows contains information about the value of the nodes (firms), and worked out how to extract it.
Jan Sebastian Nimczik went further and asked the graph to answer a question economists usually answer by assumption: what even is a labour market? Instead of defining markets as regions or industries, he let the pattern of actual job moves reveal which firms exchange people. The data-driven markets he finds cut across geography and sector - and they predict how shocks like plant closures spread better than the official definitions do.
The practitioner translation which your recruitment team have probably already told you - your competitors for talent are not your competitors for customers. The set of firms you actually trade people with is an empirical question, and most organisations have never asked it. Some of the consultancies have. Willis Towers Watson sells an engagement called Talent Flow Analysis whose opening line is “talent competition can often be different from revenue competition” - Nimczik’s finding cited to nobody.
Strategy traversed it. Management scholars spent two decades studying the edges one at a time, under the name “employee mobility” - what moves with a person when they cross from one firm to another. The answer, per John Mawdsley and Deepak Somaya’s 2016 review: knowledge, relationships, client ties, routines, legitimacy. When Martin Ganco, Rosemarie Ziedonis and Rajshree Agarwal studied inventors leaving firms known for aggressively enforcing their patents, they found the legal deterrent worked on average. The very best inventors left anyway. Barriers retain the average employee and leak the exceptional, which is worth thinking about before reaching for non-competes and gardening leave. These deterrents often become a cost of hiring, paid by the new employer.
Geography explained the neighbourhoods. Marshall and Saxenian, as above, plus Enrico Moretti’s finding in The New Geography of Jobs that each additional innovation job in a city generates roughly five further local jobs - talent concentration compounds, which is why network position, once gained, is so durable.
The true picture, I believe, is a mixture of all four. The complexity scientists have the map but few prescriptions. The economists have the interpretation but often aggregate away your firm. The strategists have firm-level advice but rarely see the whole graph. The geographers know why the clusters exist but not how to move within one. A reader trained in any single discipline - and most of us are that - inherits a piece of the picture and, usually, a quiet conviction that it is the whole.
So how should People Analysts think about this?
We economists have a framing that we call revealed preference: trusting what people do over what they say. Isaac Sorkin’s 2018 paper “Ranking Firms Using Revealed Preference” applies it to job moves. The reasoning is disarmingly simple. When a person voluntarily leaves firm A to join firm B, they have told you something: their choice reflects their interpretation of pay, but also the manager, the commute, the hours, the meaning, the prospects. Considered together they judged B better than A. One move tells you little; the person might be mistaken, or unusual, or like me in earlier days - following my partner to Geneva. But millions of moves are millions of votes, and the labour flow network records every one.
Extracting the verdict from the votes takes care, because the votes are relative - B beat A says nothing about B against C. Sorkin’s solution was, computationally, the same trick Google used to rank web pages: a firm’s value score depends on the scores of the firms it wins people from, calculated across the whole graph simultaneously until the rankings settle. Applied to US administrative data covering essentially the whole economy, this yields a ranking of employers by what workers demonstrably value - not what they tell surveys, not what awards juries decide. The flows are the employer-value measure.
And the ranking contained a surprise. Comparing it with what firms pay, Sorkin found that more than half of the firm-level variation in pay is compensation for something else being wrong - firms that pay above the market largely do so because, on the bundle of everything else, workers judge them worse places to be.
Economists call this a compensating differential, and the idea is as old as economics itself: Adam Smith wrote in 1776 that the whole of the advantages and disadvantages of different employments must be “either perfectly equal or continually tending to equality”. High pay, in other words, is as often a symptom as a strength. Think of it as the money needed to pay for the pain.
Consider what this does to the standard toolkit. League tables of pay measure the compensation, not the attraction. “Employer of choice” awards measure the marketing (and from my experience are easily gamed - I know, I’ve done it). Engagement surveys measure what people say. The flow network measures what they do - silently, continuously, and potentially (we’ll come to this in a bit) free, because half of it is already sitting in your own systems. If your firm shows a sustained net outflow of people you regret losing to a competitor that does not outpay you, the market has already returned its verdict on your employee value proposition. No focus group required.
The prestige hierarchy tempts every ambitious firm in the same direction: hire up the chain. Poach the star from the better bank, the partner from the stronger practice, the leader from the more admired brand. However, the evidence suggests more caution than the instinct does.
Boris Groysberg spent years following one of the few professions where individual performance is public and portable in principle: equity analysts, ranked annually by name. In his brilliant book, Chasing Stars, he tracked over a thousand of them, and found that star analysts who switched banks suffered an immediate performance decline that persisted for years. The star, it turned out, was partly the platform - the colleagues, the systems, the internal information networks, the brand that opened doors. Performance travelled better when analysts moved with their teams, or between firms of comparable capability. The “star” was a property of a person attached to a node, and part of it stayed at the node.
Matthew Bidwell’s 2011 study made the same point from inside a single US investment bank. Comparing external hires against internal promotions into the very same jobs, he found the external hires arrived with stronger paper credentials, were paid about 18% more - and then underperformed the promoted insiders for roughly two years while exiting at higher rates. What the insiders had was firm-specific human capital: the unwritten knowledge of how this particular place actually works, who to call, what the systems will and won’t do - an asset that doesn’t appear on a CV precisely because it cannot leave the building.
Neither finding says never hire externally, and Bidwell’s external hires did bring broader experience that eventually told - they were promoted faster once they survived. The point is narrower and more useful: when you recruit people from a higher position in the network, part of what you are paying for stays behind. The premium is real, the portability is partial, and the further the status gap, the worse the exchange rate. Prestige hierarchies are informationally efficient - position does compress real signals about quality - but they are also partly self-fulfilling, and recruiters often place more value on the signal than the hierarchy justifies.
If the labour market is a network and your firm has an address in it, the practical question is what you can do about it. The encouraging answer is that the first steps need no external data at all.
Draw your ego network. An ego network is simply the map centred on you: your node, plus every node you exchange people with. Take, say, five years of it: a ranked list of sources, a ranked list of destinations, and the net flow with each. Most organisations that do this for the first time find genuine surprises - the “competitor” they obsess over barely features, while some unregarded firm quietly harvests their best analysts. I say more below about how hard the data actually is; the short version is that the inbound half is recoverable with effort and the outbound half is much harder, and neither difficulty is a reason to skip the exercise.
Ask which flows are votes. Weight the map by regret. Losing people to firms you admire, at flat pay, is Sorkin’s signal at your own scale: a revealed-preference verdict on your proposition. Winning people from firms you admire is the opposite. Track the direction of quality, not just headcount. Your comp team probably has the pay data for this.
Identify your true talent market. Nimczik’s lesson: the firms you exchange people with define your real labour market, and it will not match your industry classification. Benchmark pay, notice periods and counter-offer behaviour against that set.
Treat leavers as dormant edges, not dead ones. An alumnus at another firm is a live connection: a referral source, a client, a possible returner. JR Keller, Rebecca Kehoe, Bidwell, David Collings and Adam Myer studied around two thousand “boomerang” rehires and found they outperformed comparable new hires precisely where coordination and internal know-how mattered most - their firm-specific capital had survived the absence. The scale of this flow is larger than most firms assume: ADP’s payroll data puts returners at around 31% of all US hires since 2018, rising to 35% in early 2025 and to nearly two thirds of hires in the information sector in one recent month. That measure is generous - it counts anyone who was once on the payroll, went inactive and came back, which sweeps in a good deal of seasonal and contract churn - but even discounted heavily, the rehire is not the rare event the exit process treats it as. McKinsey has run its alumni network as deliberate strategy for decades - it is a sales channel and a talent channel at once. Most firms, by contrast, still treat resignation as mild treason and the leaver as lost.
Recognise pipelines as bets with a tax. Repeated hiring from the same source - a university, a rival, a consultancy - is what Rhett Brymer, Janice Molloy and Brett Gilbert call a human capital pipeline, and it works: information flows both ways along the channel, onboarding routines specialise, and you can partially block rivals from the source. But Brymer’s more recent work adds the bill: pipelines over-weight the familiar, and an intake drawn down the same channels converges - the firm-level version of the homophily cost we met with referrals in Issue 12. The problem compounds when your competitors are drawing on the same pipeline. Being competitive means not behaving like your peers.
Understand that change is slow. Your network address moves over years, not quarters. Employer brand, spin-off policy, alumni management, non-compete posture, even office location are all, on this view, network repositioning levers - and all operate with long lags. That is an argument for starting, not waiting: the compounding runs regardless, in whichever direction you are pointed.
I have made this sound easier than it is, so let me be straight about the state of the tooling.
The nearest thing I can find in the market is talent competitor analysis. TalentNeuron, Aura, LinkedIn Talent Insights and others all ship a module under roughly that name, but it tells you something slightly different: it will tell you which firms hire your people and which firms supply them. If you have never looked, get a trial and look. But notice what it returns. It returns a summary matrix - your top ten destinations, your top ten sources - not your position in the network. Sorkin’s calculation needs the whole graph at once, because your rank depends on the rank of the firms you win people from, which depends on theirs. No HR vendor I can find ships a labour flow network. The distance between “here are your top ten destinations” and “here is your place” is the point of this essay, and the products don’t close it.
One exception seems worth naming. Revelio Labs sells transitions as a data feed - company-to-company moves, segmented by role, seniority and geography, monthly, back to 2008 - which is an edge list rather than a summary, so the graph calculations become yours to run. It comes with a caveat which seems natural to an economist used to official statistics like GDP but is worth highlighting to any potential buyer: as people update their profiles late, Revelio runs a nowcasting model that estimates recent inflows and outflows from historical underreporting patterns. A talent-flow chart covering the last few months is a model output, not a count. That is a reasonable approach, I think, to an unavoidable problem, but it is not what most people think they are looking at.
Getting your own data is harder than it sounds too. The inbound half is nearly there - your applicant tracking system holds every joiner’s employment history, because the candidate typed it in or the system parsed it from a CV - however, it is usually stored as an attachment and the parsing is of variable quality. Compounding this, employer names are stored as free text. Resolving “IBM”, “I.B.M.”, “IBM UK Ltd” and “International Business Machines” to one node is entity resolution, a genuine technical problem with its own research literature - CompanyDepot, built inside CareerBuilder, is a good worked example of what it takes. Here I’d turn to domain-specific embeddings: TechWolf publish theirs openly on Hugging Face, and while their public models cover job titles and skills rather than employer names, they are the right species of tool for the job. This is exactly the sort of problem those embeddings are for, and it is the reason we built our own domain-specific embedding models at OrganizationView.
Don’t discount the work of extracting the information from your ATS. Modern recruitment systems now have decent APIs, but they were never designed for this use case, and you will probably need a good data engineer to help get the information out in the shape you want.
Getting the outbound edges is worse, and worse in an interesting way. Mainstream HR systems record why someone left and whether you minded; they do not record where they went. Where a destination field exists it is free text, filled in during an exit interview or taken from a reference request, and leavers have every incentive to withhold it - the advice circulating on employee forums is explicitly not to name the firm, particularly when it is a direct competitor. So the edges least likely to be recorded are the edges to your closest talent competitors, which are the highest-information edges on the whole map. Any picture you build from your own exit data is biased towards the harmless flows. I do think you can design a process that minimises this, but it will require human effort and like the situation Revelio finds itself in, won’t be ‘real time’.
None of this argues for waiting. It argues for knowing what you are holding: an inbound map that costs effort, an outbound map that requires resources or stays partial, and a market that will sell you a list while implying it has sold you a position.
One boundary to respect: everything here concerns the ladder between firms. The ladder inside the firm - why promotions and pay rise the way they do, and what tournaments and deferred pay actually purchase - has its own strange and rigorous economics, and it gets an issue of its own later in the series.
I started in People Analytics analysing recruitment data - the reason for this recruitment series - and I have watched the field acquire ever more elaborate tools for finding out what people think, mostly through surveys. There is nothing wrong with asking. But it is worth remembering that the whole time we have been surveying, the labour market has been running a census of what people actually choose - move by move, vote by vote, written into the very administrative systems we stare past every day.
Academia’s map showed a hierarchy nobody designed, steeper than anyone would admit to wanting. Your organisation’s map has been drawn the same way, over decades, by every joiner, every leaver, every founder who once worked somewhere else. These flows contain valuable information. The question isn’t whether they are telling you something important - it’s whether you’re listening.
Clauset, Arbesman & Larremore (2015) — Systematic Inequality and Hierarchy in Faculty Hiring Networks. Science Advances
Park, Wood, Jing & Ahn (2019) — Global Labor Flow Network Reveals the Hierarchical Organization and Dynamics of Geo-Industrial Clusters. Nature Communications
Guerrero & Axtell (2013) — Employment Growth through Labor Flow Networks. PLoS ONE
Sorkin (2018) — Ranking Firms Using Revealed Preference. Quarterly Journal of Economics
Nimczik (2023) — Job Mobility Networks and Endogenous Labor Markets. Working paper
Bidwell (2011) — Paying More to Get Less: The Effects of External Hiring versus Internal Mobility. Administrative Science Quarterly
Groysberg (2010) — Chasing Stars: The Myth of Talent and the Portability of Performance. Princeton University Press
Ganco, Ziedonis & Agarwal (2015) — More Stars Stay, but the Brightest Ones Still Leave: Firm and Inventor-Level Antecedents of Patent Litigation. Strategic Management Journal
Mawdsley & Somaya (2016) — Employee Mobility and Organizational Outcomes: An Integrative Conceptual Framework and Research Agenda. Journal of Management
Keller, Kehoe, Bidwell, Collings & Myer (2021) — In with the Old? Examining When Boomerang Employees Outperform New Hires. Academy of Management Journal
Brymer, Molloy & Gilbert (2014) — Human Capital Pipelines: Competitive Implications of Repeated Interorganizational Hiring. Journal of Management
Cappelli & Keller (2014) — Talent Management: Conceptual Approaches and Practical Challenges. Annual Review of Organizational Psychology and Organizational Behavior
del Rio-Chanona, Mealy, Beguerisse-Díaz, Lafond & Farmer (2021) — Occupational Mobility and Automation: A Data-Driven Network Model. Journal of the Royal Society Interface
Saxenian (1994) — Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Harvard University Press
Moretti (2012) — The New Geography of Jobs. Houghton Mifflin Harcourt
ADP Research - Boomerang hiring makes a comeback. Payroll-data analysis of returning employees
I use various AI tools to help me get my ideas into type. You can read my full process here.