The Talent-Competitor Set: Ranking Who You Fight for Hires

This is the method for turning one skill, level, and location into a ranked list of the organizations you actually compete with for those people. It is written for strategy and research teams, talent-intelligence analysts, and operators who need the offer-competition list a talent team briefs against when it loses candidates. By the end you can build that list from public data, separate true talent rivals from mere business competitors, surface the cross-industry bidders you never see coming, and attach a refresh cadence so the set stays valid.

Most published research either maps one company's talent inflows and outflows or scores a market's hiring difficulty. Neither produces the list you want when a candidate takes a competing offer and you need to know who wrote it. This playbook anchors on a single skill and level rather than a company, which is exactly why it catches the sectors that win your candidates but never appear on a product-competitor slide.

What is a talent competitor, and why it is not your business rival

A talent competitor is any organization that recruits for the same skill, level, and location you do, regardless of what it sells. That definition, not your revenue-competitor list, is the one that predicts where your offers go to die.

The distinction is easy to state and easy to forget. Business competitors compete on services and products; talent competitors compete for your ideal hires, often from very different industries. Amazon and JPMorgan Chase may not compete on services, but both target top software engineers, operations experts, and data analysts. When a retailer began building payments capability, the exact skills it needed were also sought by fintech startups and payment processors, so it found itself competing with companies like Stripe, PayPal, and Square - none of which sell what a retailer sells.

The reason this matters is structural, not seasonal. In past cycles, tech hired from tech and healthcare hired from healthcare. That is no longer how it works. More than 40% of US software engineer and developer hires were made by nontech companies in 2019, up from about a third in 2010. Skills are transferable across every digitizing sector, so the buyers for your talent multiply faster than your product rivals do.

The practical consequence: if you only benchmark against your own vertical, you are blind to where candidates go when they reject your offers. The whole point of this method is to force those hidden bidders onto the page.

Which public signals reveal talent competition, and how each one lies

Three public signals indicate that two organizations compete for the same people. Each is useful, and each misleads in a specific way you must correct for.

The first is posting overlap: employers advertising the same skill, title, and location. Posting data shows where demand for a role is highest, which locations are rising or falling, and where competitors concentrate their hiring. It lies because job postings are not the same as vacancies. Postings measure recruitment marketing by employers purportedly looking to fill roles, and many recruitment practices break the link between an ad and a real opening. A staffing agency posting one title ten thousand times will top your ranking without competing for a single of your candidates.

The second is talent flow: the prior-employer histories visible in profiles, aggregated into where a company gains and loses people. Flow data shows where employees came from and where they go after leaving, with attrition calculated from observed profile transitions. It lies because self-reported titles are noisy and must be normalized, and machine mapping folds many member-entered titles into one standardized label, which can hide sub-specialties.

The third is offer competition: candidate-reported losses, the record of which companies actually beat you to a hire. This is the strongest signal and the least public. No vendor supplies it, because every public tool returns aggregates only.

There is no published numeric threshold - no "X% posting overlap equals a talent competitor." Vendors describe the distinction qualitatively. If you want defensible overlap math, labor-economics work quantifies concentration with the Herfindahl-Hirschman Index computed from postings by occupation and commuting zone, but the label itself is not standardized. Treat your ranking as ordinal and confirm the top of it against real losses.

Define the atomic unit: skill, level, and location

The unit of analysis is a single skill at a single level in a single location, not a company. Get this wrong and every downstream number describes the wrong pool.

Anchoring on a company gives you a product-rival list. Anchoring on a skill-level-location tuple gives you the offer-competition list, because it captures everyone hiring for that exact profile no matter what they build. The tuple also sharpens the pool dramatically, which changes who you actually fight.

Refolk's index shows how much the axes matter. Geography narrows the pool: the same skill and level splits to 1,188 Senior Rust engineers in the US versus 325 in Germany. Seniority narrows it far more sharply than geography does.

Here are the three cuts, drawn straight from Refolk's index of professional profiles. Read them as the reason the tuple must be fixed before anything else.

Table A - One skill, two markets (Senior Rust engineers)

Table B - One market, two skills (US, Senior level)

Table B carries the most important lesson about competitive intensity. Senior ML talent outnumbers Senior Rust talent 85.7x in the same US band. A scarce-skill set will be dominated by a few determined bidders; an abundant-skill set spreads across many employers. Concentration rises as the number of qualified employers per available candidate rises against a thin supply. Which skill you anchor on, not how many people you plan to hire, sets how brutal the competition is.

The step-by-step procedure

The method runs eight steps from a defined pool to a published, owned, and self-refreshing set. Budget roughly four to five working days for the first pass; later refreshes are faster because the frame is fixed.

Two ordering notes. Sources disagree on whether to map company flows first or skills first; TalentNeuron and Lightcast start from role and skill postings, and I recommend skill-first for this playbook because it is what surfaces the cross-industry bidders. And the validation step is not optional decoration - it is the one thing no vendor does for you.

Surface the cross-industry bidders you never see coming

Run the cross-industry sweep as a named step, because the sectors that win your candidates rarely appear on any product-competitor list. This is where the skill-first frame earns its keep.

Refolk's index makes the point concrete. The top current employers of Senior Rust talent in the US span aerospace, robotics, and biotech: Cloudflare, SpaceX, Figure, OpenAI, Biohub, and F5. A typical software-company rival list would name almost none of these. The German cut echoes the pattern - automotive supplier ZF Group sits among Rust employers alongside DeepL, Capgemini, and Matter Labs.

Product-competitor lists never name the aerospace, robotics, and biotech firms quietly outbidding you for the same engineers.

The sweep is deliberate work, not a hope that a tool surfaces these on its own. Force at least three non-obvious sectors into the frame each cycle. The anchor expectation is that non-tech employers now hire tech skills at scale, so a sweep that returns only tech companies has failed by construction.

Where a public tool tells you a sector is active, Refolk lets you read the specific people and employers inside it in plain English, which is what turns an abstract "aerospace hires Rust" into a concrete list of companies to add to the set.

Tier the ranked set and attach evidence to each entry

Rank the set into three tiers, and record the evidence type behind every placement so the ranking survives scrutiny. No single canonical tiering is published, but the components are well documented, and the discipline is to never place a company in a tier without naming the signal that put it there.

• Core rivals: high posting volume and skill overlap in your geography. Evidence type: postings and skill-overlap data. These are the companies advertising hardest for your exact tuple.
• Feeder or academy companies: organizations your pool consistently comes from. Evidence type: talent-flow inflow data. A survey of 853 executives named McKinsey, Google, Microsoft, Unilever, and P&G as the most-mentioned academy companies, and in 2018 there were 31 companies on the S&P 1500 led by former GE employees - but a generic list is a starting hypothesis, not your answer.
• Adjacent and emerging entrants: fast-growing companies and new market entrants moving into your skill. Evidence type: posting growth rate. Emerging-entrant analysis exists to reveal new entrants and fast-growing companies to inform retention strategy.

Choose your sources and know what each omits

Pick sources by what they can prove, and hold each one's blind spot in view. Three public source types cover posting volume, workforce composition, and work histories; none covers the offer outcome.

A postings-and-profiles labor-market source de-duplicates job postings from about 40,000 websites, which constitutes most of US online vacancies, and covers 165 countries representing 99% of world GDP; its database spans over 2.5 billion job postings and 800 million career profiles. Its blind spot: it measures new postings, and it standardizes and estimates employer names, so its de-duplication can filter a lower share of duplicates as scraped sites rise, inflating volume trends over time.

A government baseline gives you the reality check. The BLS Quarterly Census of Employment and Wages covers 95% of the employed US workforce broken out by industry, but omits skills and company-level movement. Use it to sanity-check posting trends, not to build the set.

A talent-flow source shows how a company gains and loses talent over the last 6, 12, or 24 months plus the current month, which is what identifies feeders. But it returns aggregate figures only; it does not show individual candidate profiles or let you message anyone, and it depends on self-reported profiles.

How this goes wrong: the failure modes

The failure modes below are where a talent-competitor set turns into confident nonsense. Each has a false positive or a blind spot, and each has a check that catches it. Treat this section as the part of the method you return to when a list looks wrong.

• Posting overlap is not offer competition. A staffing agency or aggregator posts the same title thousands of times and tops your list. Check: filter to standardized or verified employer names and cross-check against talent-flow data. Remember that postings measure recruitment marketing, not vacancies.
• Title normalization noise. "SWE," "Software Eng.," and "Software Developer" fragment or over-merge the same pool, and machine mapping can fold distinct sub-specialties into one label. Check: inspect the skill layer, not just the title.
• Industry-only benchmarking blind spot. Benchmarking against your own vertical hides where candidates actually go when they reject you. Check: force the cross-industry sweep every cycle, and reject any set whose sectors all match your own.
• De-duplication drift. As the number of scraped sites rises, a postings source may filter a lower share of duplicates, inflating volume trends and making a company look like it is scaling hiring when the data is. Check: compare against a government baseline such as JOLTS or QCEW.
• Stale set. A quarter-old comp number loses live candidates. A role posted at a salary approved three quarters ago draws thin applications, ghosting, and lost offers. Check: enforce the quarterly plus 90-day refresh trigger.
• Academy list is not your feeder list. The famous incubator names may not feed your specific skill and geo. Check: validate feeders against your own inflow data, not a generic survey.
• Aggregate-only tools cannot confirm who won a candidate. Every public platform shows aggregates, not individual outcomes. Check: pair the data-derived set with a candidate-reported offer-competition log.

The last of these is the load-bearing point. The offer-competition signal is private, which is why the validation step is the guide's true differentiator, and it is worth restating as a rule.

Keep the set current

A talent-competitor set decays, so bind it to a refresh cadence at the moment you publish it. Quarterly is the practitioner consensus for active roles; niche and executive searches get re-researched at the start of each new search.

The rhythm that works is monthly data collection with quarterly analysis and reporting to leadership. The most successful organizations refresh their market data quarterly. For niche or executive-level searches, conduct fresh research at the start of every new search in a role type you have not placed in 90 days or more. Compensation-only benchmarks can move slower - a comp refresh every two to three years is a floor and annual is better - but the competitor set itself moves with hiring, not with pay cycles.

Publishing is itself part of keeping the work alive. Distribute findings to product, sales, finance, and executive teams; the single biggest mistake is keeping talent data inside HR. A set that only recruiters see cannot inform the retention and comp decisions that actually change who wins your candidates.

Before you call the first pass done, run this check.

POOL: <skill> / <level> / <location> - supply: <n> people
DATE BUILT: <date>   OWNER: <name>   NEXT REFRESH: <date>
CORE RIVALS (postings + skill overlap):
  1. <company> - evidence: <posting count / top overlapping skills>
  2. <company> - evidence: <...>
FEEDERS (confirmed inflow):
  1. <company> - evidence: <inflow count from your flow data>
ADJACENT / EMERGING (posting growth):
  1. <company> - evidence: <posting growth rate, sector>
OFFER-LOSS VALIDATION:
  - Losses last 2 quarters: <n>, mapped to set: <n>, exceptions logged: <n>
DISTRIBUTION: shared with <product / sales / finance / exec>

Adapt the template to your own stack, but do not drop the offer-loss validation row. It is the difference between a set that predicts where you lose candidates and one that merely lists who advertises loudly.