The Talent-Competitor Ranking Standard: Tiering Rivals by Public Movement

Every strategy team needs a defensible answer to one question: which companies actually compete with me for the same people, and how much should each one weigh in my planning? This guide is the rubric for that judgement. It is for talent-intelligence analysts, strategy teams, and operators who have public profile movement data but no paid analytics license, and it gives you the dimensions, the weights, the sample-size floors, and the tier thresholds to score a company by hand.

Most sources that rank today explain talent-flow mapping and then route you into a platform. They also conflate two different lists: the companies you compete with for revenue and the companies you compete with for talent. Those lists overlap, but the gap between them is where hiring plans go wrong. This standard keeps them separate and gives you a method you can run on public records.

Why talent competitors are not your revenue competitors

A talent competitor hires the same people you do; a revenue competitor sells to the same buyers. The two lists diverge, and the divergence is documented. As one workforce advisory puts it, organizations tend to assume their talent competitors are the same as their revenue competitors, "but we know this is not always the case," because competition for technology and digital roles "has become more industry agnostic, moving well beyond traditional competitor groups."

The concrete examples are blunt: a healthcare provider may compete with tech giants for data scientists, and a bank may lose engineers to a fast-growing startup. Refolk's own index corroborates the pattern. In the German Machine Learning Engineer sample, the top employer is not a tech company at all - it is the chemicals maker BASF. If you had pre-seeded your competitor list with product rivals, you would never have found it.

This matters because failing to distinguish the two lists carries a cost. Industry commentary warns that companies which do not separate talent competitors from industry competitors "risk missing top talent and lengthening time-to-fill." The rest of this guide assumes you will let movement data surface non-obvious employers first, and filter to a competitive set second.

The four dimensions and what each one proves

A talent competitor is scored on four dimensions: net talent flow, skill and role overlap, geographic co-presence, and hiring velocity. Each proves something different, and each has a way it lies.

• Net talent flow. Inbound minus outbound moves between you and the company over a fixed window, plus the hires-to-departures ratio. This proves the company is actually exchanging people with you. It lies when you count only departures to a rival and ignore the hires you take back, which mislabels a net source as a drain.
• Skill and role overlap. How much of the company's hiring targets the same roles and skills you compete for. This proves you are fishing in the same pool. It lies when title noise merges non-competing roles - "engineer" alone spans dozens of real jobs.
• Geographic co-presence. Whether the company hires the segment in the same metros you do. This proves the pools physically overlap. It lies when two firms hire the same role in different cities and look like competitors but never contest a single candidate.
• Hiring velocity. How aggressively the company is filling the segment and how fast it is growing. This proves future pressure, not just past exchange. It lies through reporting lag, which understates the newest and fastest-growing entrants.

No public source publishes a numeric overlap threshold, and none publishes a canonical weighting for these four dimensions. That is the honest state of the evidence. The dimensions are documented; the thresholds and weights below are my contribution, offered as a defensible default you should tune and write down.

The scoring rubric: weights, scales, and the tier map

Score each company 0 to 5 on all four dimensions, multiply by the weight, and sum to a single number out of 100. Net flow carries the most weight because it is the only dimension that measures an actual exchange of people rather than a proxy for one.

| Dimension | Weight | 0 means | 5 means | | Net talent flow | 40% | No moves either way | Heavy two-way exchange, ratio near or below 1.0 (a drain) | | Skill and role overlap | 25% | Different roles entirely | Same roles and skills you hire | | Geographic co-presence | 20% | Different metros | Same metros, head to head | | Hiring velocity | 15% | Flat or shrinking | Fast-growing, hiring the segment hard |

The single score then maps to one of four tiers, and each tier drives a specific action. A tier with no action is a leaderboard, not a rubric.

| Tier | Score band | What it is | Action | | Core rival | 70 to 100 | Direct, two-way, same metro | Name in comp and retention planning; monitor monthly | | Adjacent poacher | 45 to 69 | Strong on flow or velocity, cross-industry | Benchmark pay against them; watch for surges | | Watchlist entrant | 25 to 44 | Fast-growing, thin history, high lag risk | Re-check quarterly; do not rank on current data alone | | Out of set | 0 to 24 | Weak overlap or no exchange | Exclude; revisit only if the segment changes |

How to compute net talent flow without a platform

Net talent flow is inbound moves minus outbound moves between you and one company over a fixed window, expressed alongside the hires-to-departures ratio. A ratio above 1.0 means the company is a net source of talent for you; below 1.0 means a net drain. No named source publishes an exact numeric cutoff separating drain from source, so 1.0 is the natural pivot, not a rule handed down from a vendor.

To resolve a single job change into a directional edge you need five fields: previous employer, new employer, the role at each end, timing, and location. Public profiles carry all five - job titles, employers, skills, and a self-selected location - and the workforce-data schema formalises it: each transition row records the previous and new roles, location, and seniority, with the month recorded as "YYYY-MM." Normalise every title and roll company names up to their parents before you count, or title noise will merge pools that do not compete.

For attrition, if you need it as a hiring-pressure input, use the standard formula and hold the denominator fixed across every company: departures divided by average headcount, times 100, where average headcount is start plus end divided by two. A worked example: 5 exits over an average headcount of 75 is a 6.67% annual attrition rate. For context, U.S. voluntary attrition averages about 13%, ranging from 8.2% in insurance to 26.7% in retail and wholesale, so a segment number only means something against its own industry band.

The workflow below turns these dimensions into a repeatable half-day-to-day procedure. Refolk removes the heaviest friction in Step 2 and Step 3: pulling the move records and resolving them into clean edges. Instead of scraping profiles and hand-normalising titles, you ask for the segment in plain English and get back the people and their prior employers already resolved.

The procedure, end to end

Run this in order. The first three steps build the edge list, the middle two score it, and the last two turn scores into tiers and actions. Timings assume one analyst and a strategy lead for the final call.

The sample-size floor and why pool depth sets your ceiling

Before you trust any pair's net flow, enforce a floor of roughly 30 moves. No talent-flow vendor publishes a move-count minimum, so this borrows the statistical n=30 rule of thumb: at about 30 observations, the distribution of sample averages approximates a normal curve, which is what most tests assume. It is explicitly a heuristic, not a universal answer, but it stops a two-move "flow" from being ranked as a top competitor.

Pool depth compounds the risk. In a thin market a few competitor moves swing your whole hiring plan, which is exactly where the floor matters most. Refolk's index makes the point across two countries for the same role.

| Role | Country | Matching profiles | Top sample employer | | Machine Learning Engineer | United States | 10,571 | Meta | | Machine Learning Engineer | Germany | 1,243 | BASF | | Ratio US:DE (derived) | - | 8.5x | - |

The same index shows a second pattern worth scoring for: one large employer can dominate several adjacent pools at once. Meta is the top sample employer for both U.S. Machine Learning Engineers and U.S. Data Scientists, so a single retention action can cover two segments.

| Role | Country | Matching profiles | Top sample employer | | Machine Learning Engineer | United States | 10,571 | Meta | | Data Scientist | United States | 28,741 | Meta | | Ratio DS:MLE (derived) | - | 2.72x | - |

One public methodology encodes the same caution differently: it includes a source company only when its share of the target's total inflow is at least 3.5%, and it nudges sparse signals toward the median "so a company cannot jump to the top off a single lucky data point." That median-nudge is the difference between a rubric and a leaderboard. If you have thin data, pull the score toward the middle rather than letting one viral hire crown a rival.

Sparse-signal smoothing is what separates a rubric from a leaderboard.

How this goes wrong: the failure modes

Most bad talent-competitor rankings fail for one of seven reasons. Each has a check you can run before you publish a tier.

• Title noise inflates false edges. "Engineer" spans dozens of real roles, and without normalisation you merge non-competing pools. Check: confirm titles were standardised before counting.
• Small-n mistaken for signal. A single high-profile move looks like a trend. Check: enforce the n=30 convention per pair and push thin signals toward the median.
• Gross flow read as net. Counting only departures to a rival ignores the hires you take back. Check: always compute both directions and the ratio.
• Reporting lag. Public profiles update late, so recent flows are understated. Check: use a lagged window, not the latest month, and give fast-growing entrants a velocity bonus.
• Revenue-competitor bias. Analysts pre-seed the list with product rivals and miss cross-industry poachers like the BASF and Meta cases. Check: let the data surface non-obvious employers before filtering.
• Geographic false overlap. Two companies hiring the same role in different metros are not competing for the same people. Check: score location co-presence separately, not just role match.
• Denominator drift. Mixing start-count, end-count, and average-count across companies makes attrition non-comparable. Check: fix one denominator rule for every company, and note that opening headcount understates attrition during growth.

The lag failure is worth its own weight. Public data understates your newest rivals structurally, which is why workforce-data vendors model the inflows and outflows "that will be revealed once all public profiles have been updated" rather than counting raw ones. If a rubric ranks only what is already visible, it will always be a step behind the market.

Before you call the ranking done

Run this checklist against every company before it earns a tier. It catches the failure modes above in the order they usually bite.

Keeping the ranking current

A talent-competitor ranking decays because the underlying data decays. Re-run the full procedure on a fixed cadence rather than reacting to individual moves, and treat the tiers as a living document. Core rivals warrant monthly monitoring; watchlist entrants warrant a quarterly re-check because their thin history and high lag risk make any single reading unreliable.

Two triggers should force an off-cycle refresh. First, a segment change: if you open hiring in a new metro or a new role family, the geography and overlap scores you set no longer hold, and the candidate list has to be rebuilt from Step 1. Second, a velocity spike: if a watchlist entrant starts filling the segment fast, its lag-understated inflow will surface soon, so re-score it before it becomes a core rival you failed to plan for.

Keep the thresholds you invented visible. Because no source publishes canonical overlap thresholds or weights, your rubric is only defensible if the next analyst can read exactly what you decided and why. Record the weights, the sample floor, the denominator rule, and the tier bands alongside the ranking itself. The template below is the minimum you should carry from one run to the next.

SEGMENT: [function / seniority / metro radius]
WINDOW: 12 months, lagged one month
SAMPLE FLOOR: 30 moves per pair
DENOMINATOR: average headcount = (start + end) / 2
WEIGHTS: net flow 40 / overlap 25 / geo 20 / velocity 15

Company | Net flow (in-out) | H:D ratio | Overlap 0-5 | Geo 0-5 | Velocity 0-5 | Weighted total | Tier | Action

Done well, this ranking does something the paid platforms describe but rarely hand you: it names the companies contesting your people, weighs them by how much they actually move talent with you, and turns each one into a decision. The method is portable, the thresholds are yours to defend, and the whole thing runs on public movement data you can pull yourself.