Refolk
FrameworkRecruiting and sourcing

The Inbox-Crowding Score: Six Proxies to Lead, Differentiate, or Rank Down

Score any candidate on six observable crowding proxies in under five minutes and reach the same Lead, Differentiate, or Rank Down verdict a colleague would.

21 min readLast reviewed November 5, 2026Read as Markdown

Key takeaways

  • In the one published saturation dataset, all-channel first-touch reply fell from 14.5% when one organisation reached a candidate to 5.6% when 10 or more did within 90 days.
  • The crowding penalty is about 2.6x on LinkedIn and on email alike, so switching channel raises the base reply rate but does not get you out from under the crowding.
  • About four in five first touches go to candidates no other team has reached, so crowding is a minority problem concentrated on the people every competitor's obvious query returns first.
  • Openness is not crowding: candidates with no current role replied 56% to 76% more and were under-contacted, so openness should pull a verdict toward Lead and never add a crowding point.
  • Your own history outweighs market signals: past responders re-contacted 30+ days later replied 3.6x as often as first-time contacts, while people who never replied before replied 85% less often.
  • A crowding score is only repeatable if the obvious query, its filters, and the Recruiter seat are fixed in writing, because default rank changes with seat, filters, and network distance.

Before you spend a first touch on someone, you should know whether every other recruiter running the same search has already been in their inbox. This guide is for in-house recruiters, sourcers, talent leaders, and founders who do their own hiring. It gives you six observable proxies, a crowding score from 0 to 9, three verdicts, and the change in sequence order or approach that each verdict calls for. It is built so two sourcers scoring the same profile reach the same answer.

What the inbox-crowding read decides

The read answers one question: should this candidate get your standard first touch now, a differentiated first touch, or a later slot behind an uncrowded candidate of similar fit? It does not judge fit. It does not judge whether the person is ready to move. Those are separate reads, and you should finish them before this one.

Crowding is the number of distinct organisations that contact one person within a 90-day window. That is how the one published saturation dataset defines it, and I use the same definition here. You can't see crowding directly, because nobody can see other recruiters' outreach. So the model scores things you can see that tend to go along with it.

The three verdicts:

  • Lead: send your standard first touch, in your normal sequence position.
  • Differentiate: contact the candidate, but change the channel, the sender, the timing, or the hook so your message is not one of many identical InMails.
  • Rank Down: put the candidate in the sequence below an uncrowded candidate of similar fit, and score them again at the next touch window. Rank Down changes the order. It never removes anyone from the slate.
4 in 5
First touches that went to candidates no other team on the platform had reached
From an outreach vendor's saturation report covering 3.1M+ first-touch events over twelve months; crowding is the minority case, concentrated on the visible few.

This is why the usual advice on candidates contacted by too many recruiters is only half right. Most of the market is not picked over. The likely reason crowding concentrates on a small group is that everyone runs the same obvious query, so the same people come up at the top for everyone. The read is for spotting that small group one profile at a time, without assuming the whole slate is burned.

How much does crowding cut reply rates?

In the one published dataset, all-channel first-touch reply falls from 14.5% when one organisation reaches a candidate in 90 days to 5.6% when 10 or more do. That is roughly a 2.6x drop, and it shows up at about the same ratio on every channel.

Channel1 team10+ teamsMultiple (derived)
LinkedIn30.3%11.7%2.6x
All channels14.5%5.6%2.6x
Email12.1%4.7%2.6x

I took the per-channel values from the chart's data labels in the source report. Check the order of the series on the original chart before you quote the LinkedIn and email rows separately. I calculated the multiples myself.

2.6x
Drop in first-touch reply from one contacting team to ten or more within 90 days
The ratio holds on LinkedIn and email alike, so changing channel lifts the base rate without escaping the penalty.

The fact that every channel drops by the same ratio is the most useful thing in the table. Moving a crowded candidate from email to LinkedIn raises the base rate you start from, but the crowding penalty comes with you, because the same people get crowded on every channel. Differentiating helps. Changing channel alone does not reverse crowding.

Metro averages show how much competition varies by market. Here is the same report's software-engineer data:

MetroTeams per candidate (90d)First-touch reply
Dallas-Fort Worth1.2816.5%
Atlanta1.2614.7%
Chicago1.3615.2%
Austin1.4415.6%
New York~46.3%

A typical New York software engineer heard from nearly four teams in 90 days and replied to 6.3% of first touches. Metros near 1.3 teams per candidate got 15% to 16.5%. That gap follows how many teams compete in the metro, not a different kind of candidate. The data has noise: Atlanta has fewer teams per candidate than Dallas-Fort Worth and still a lower reply rate. Use these figures as reference points, not as thresholds.

Limits of this evidence:

  • The report comes from a company that sells outreach software, which is a conflict of interest. Its denominators are first touches over a 30-day reply window. Keep that in mind when you compare it with other figures.
  • Nobody has published how much any differentiation tactic lifts replies on crowded candidates specifically. Every lever in this guide was measured across whole candidate pools.
  • The shares in the middle crowding buckets are only partly readable in the source, so I don't use them.

The six crowding proxies: what each proves and how it lies

You can observe six things that stand in for crowding: default-query rank, employer raid pressure, role-and-metro posting density, profile fatigue markers, channel scarcity, and your own CRM touch history. The first five add up to a crowding score from 0 to 9. CRM history and openness flags then move the verdict.

ProxyPointsWhat it provesHow it lies
Default-query rank0-2Competitors see this person firstRank shifts by seat and filters
Employer raid pressure0-2A public event sent recruiters hereCandidate may be re-employed
Role-and-metro density0-2Many teams chase a small poolRelative; moves with region set
Profile fatigue marker0-1Person says they are over-approachedNo published data; easy to over-read
Channel scarcity0-2Everyone queues in one inboxA listed route may be stale
Own-CRM historymodifierYour relationship, or its absence"Never replied" read as "uncontacted"

Default-query rank

The obvious query is the title, metro, and core-skill search a typical competitor would run. If your candidate is on its first two pages, every recruiter who ran that search has seen them. LinkedIn Recruiter shows at most 1,000 results, which is 40 pages of 25.

Scoring: ranks 1 to 50 score 2, ranks 51 to 250 score 1, and anything deeper or absent scores 0. These cut points are my convention. LinkedIn does not publish its ranking formula. The common claim that most recruiters only read the first two or three pages has no stated method behind it. The bands work only because everyone on the team applies the same ones.

How it lies: two sourcers running "the same" search get different ranks, because results depend on the seat, the filters, and network distance.

Obvious-query spec (one per req)
Req:
Shared seat used for rank tests:
Title string (exact, with OR variants):
Metro and radius:
Core skill keywords (exact):
Filters applied (list every one, or write "none"):
Date run:
Result count:
Comparison metros for density band (fixed set, same query):
Density band for this req: High / Medium / Low

Fill this once per req, paste it into the req notes, and rank every candidate on the slate against it.

Employer raid pressure

A layoff or other public event at a candidate's employer sends recruiters straight to that company's people. One study, cited by a secondary source, found 37% of laid-off tech workers re-employed within one month and 79% within three, and 45% of those re-employed were recruited into the job. That last figure means recruiters actively go after these people.

Scoring: a public event 0 to 30 days old scores 2, 31 to 90 days scores 1, and nothing in 90 days scores 0. Check Layoffs.fyi and the news. Nobody has established how long recruiter attention stays high after an event, and I found no data on acquisitions or lockups. The 30-day cut is my convention.

How it lies: the candidate may have been re-employed already. Before you score, check the profile's current role and its start date. If a new role started after the event, score the new employer instead.

Role-and-metro density

This proxy measures how many teams are hiring for this role in this metro compared with how many people could fill it. Score it once per req, not once per candidate. Count open postings for the role against the pool size from your obvious query, for your metro and a fixed set of comparison metros. Band your metro High (2), Medium (1), or Low (0) relative to that set.

There are no published thresholds. The metro table above gives rough reference points: a market that behaves like New York for software engineers, at about four teams per candidate, is High, and one near 1.3 is Low. LinkedIn Talent Insights reports "hiring demand" as the average number of Recruiter InMails sent to a pool over 12 months, relative to other pools. Use it if you have it, but it is paywalled and relative.

How it lies: change the location filter and the relative figure changes. Keep the comparison set the same for the life of the req.

Profile fatigue markers and openness flags

A fatigue marker is a public statement that the person is over-approached. Examples are a headline or about section asking recruiters not to message, or one stating firmly that they are not looking. It scores 1. I cap it at 1 because no published data ties these markers to reply rates, which makes this the most judgement-dependent point on the card.

Openness flags are different, and they work in the opposite direction. These include a seeking headline, no current role, the Open to Work badge, and a recently updated profile. One outreach dataset found candidates with no current role replied 76% more by email and 56% more on LinkedIn, and they got less contact than their share of the market would predict. LinkedIn's own published figure for Open to Work candidates has moved over time, from 75% in an older post to 37% and then about 35% more likely to respond. Every source agrees on the direction. Openness is a modifier, never a crowding point.

Channel scarcity

If the only way to reach someone is InMail, you are in the same queue as everyone else. Scoring: InMail only scores 2, one other verified channel scores 1, and two or more score 0.

This matters for your own account health too. Secondary sources report that LinkedIn can pause bulk InMail for Recruiter seats that fall below a 13% response rate across 100+ InMails in 14 days. Check LinkedIn's current terms. Spending InMails on crowded people only reachable by InMail lowers your account-wide response rate.

How it lies: a channel you listed may be stale. Count only routes you have actually verified.

Own-CRM touch history

This proxy is the one that moves reply rates most, and it is in your own system. Record one of three values: prior reply, never replied, or never contacted. Past responders re-contacted 30+ days later replied 3.6x as often as first-time contacts, across 15,000+ candidates. People who never replied before replied 85% less often. "Never replied" is a negative signal. It does not mean the same thing as "never contacted".

How to score a candidate and assign the verdict

Add the five crowding proxies for a total from 0 to 9. Map 0 to 3 to Lead, 4 to 6 to Differentiate, and 7 to 9 to Rank Down. Then apply the modifiers once each, in a fixed order.

Modifiers, applied in this order:

  1. Prior reply in your CRM: the verdict becomes Lead, and the first touch is a warm re-contact. Stop here.
  2. Openness flag present: move the verdict one step toward Lead.
  3. Never replied to you before: move the verdict one step toward Rank Down.
  4. Rank Down check: Rank Down stands only if the slate has, or you can find, an uncrowded candidate of similar fit. If not, the verdict becomes Differentiate.

Worked example: a candidate ranks 12th on the obvious query (2), their employer had a layoff 20 days ago (2), the req is Medium density (1), there is no fatigue marker (0), and you have one verified email (1). Their crowding score is 6, which is Differentiate. They have no current role, so the openness modifier moves them to Lead. Because the score is driven by raid pressure, Lead here means this week, not next.

Ordering the slate by fit and crowding

Strong fitAdequate fit
Lead first
Standard first touch at the top of today's sequence
Differentiate
Change channel, sender, or timing before the first touch
Lead next
Standard first touch after the strong uncrowded names
Rank Down
Sequence behind uncrowded equivalents and re-score later
Crowding score low (0-3)Crowding score high (7-9)
Crowding changes when and how you contact a strong candidate, and only pushes adequate-fit candidates behind uncrowded equivalents.
Inbox-crowding score card (paste into the ATS note)
Candidate:
Scored by / date:
Obvious-query spec: (link to req note)
1. Default-query rank: ___ (1-50 = 2 | 51-250 = 1 | deeper or absent = 0)
2. Employer raid pressure: ___ (event 0-30 days = 2 | 31-90 days = 1 | none = 0)  Event and date:
3. Role-and-metro density (from req): ___ (High = 2 | Medium = 1 | Low = 0)
4. Fatigue marker: ___ (explicit public signal = 1 | none = 0)
5. Channel scarcity: ___ (InMail only = 2 | one other channel = 1 | two or more = 0)
Crowding score (0-9): ___
Band: Lead (0-3) | Differentiate (4-6) | Rank Down (7-9)
Modifier, CRM history: Prior reply / Never replied / Never contacted
Modifier, openness flag: Yes / No  Which flag:
Final verdict:
Sequence change:
Uncrowded equivalent (required for Rank Down):

One card per candidate. Every line takes one value, so two scorers can compare their cards line by line.

The hard part of Rank Down is finding the uncrowded equivalent. The obvious query is the tool that surfaces crowded people, so running it again won't find you anyone else. You need a different description of the same fit: someone without a listed employer, in a lower-density metro, or found through public work instead of a title match. You can describe that person in plain English to Refolk, which searches GitHub, LinkedIn, and the open web, and it returns people the obvious query doesn't put on page one.

Step-by-step: scoring one candidate in under five minutes

Once the obvious query and density band are set for the req, each candidate takes about a minute per remaining proxy. That comes to under five minutes per profile.

The inbox-crowding read, one candidate

  1. Define the obvious query
    Write the title, metro, and core-skill search a typical competitor would run, with every filter, and save it from one shared seat with its result count. Once per req.
  2. Run the rank test
    Record the candidate's rank in the saved query: 1 to 50 scores 2, 51 to 250 scores 1, deeper or absent scores 0.
  3. Check employer raid pressure
    Confirm the current role and start date, then check the employer on Layoffs.fyi and in the news for events in the last 90 days, and log the event and date.
  4. Assign the role-and-metro density band
    Once per req, count postings against pool size across a fixed set of metros and band yours High, Medium, or Low.
  5. Read fatigue markers and openness flags
    Score a public over-approach signal as 1, and record any openness flag separately as a modifier.
  6. Pull own-CRM touch history
    Record prior reply, never replied, or never contacted from your own CRM or ATS.
  7. Check channel scarcity
    List verified non-InMail routes: InMail only scores 2, one route scores 1, two or more score 0.
  8. Record the verdict and sequence change
    Sum the score, map it to a band, apply the modifiers in order, and write the verdict, the change, and any uncrowded equivalent into the ATS.

Practitioners disagree about the order. One outreach vendor treats multi-channel reach as the first priority. Some layoff-timing advice says to fix timing first. Market-intelligence advice says to choose the market (step 4) before doing any candidate work. I keep the order above because it puts the per-req work first and the cheapest per-candidate checks next. After a fresh layoff at a target employer, though, do step 3 first for that employer's people, because they are being re-employed quickly.

What to change for each verdict

Each verdict maps to a concrete change: Lead keeps the standard sequence, Differentiate swaps in the strongest lever you have, and Rank Down changes the order and sets a date to score again. Here are the levers with their published effects. All were measured across whole pools, not on crowded candidates specifically.

LeverEffectSample
Past responder re-contact3.6x first-time15,000+ candidates
Email plus LinkedIn vs email only+27% within-team280,000+ sequences
Two-step email plus LinkedIn vs email only45.76% vs 19.73%4M+ messages
Send-on-behalf-of (SOBO)+50% or morenot stated
Leader- vs recruiter-signed email5.27% vs 5.53%1.5M+ emails

Send-on-behalf-of (SOBO) means sending the message from a hiring manager's or leader's account rather than the recruiter's.

What moves a reply rate, strongest first

  1. Own history
    Past responder re-contacted 30+ days later replies 3.6x as often
  2. Crowding
    One team vs ten or more in 90 days is a 2.6x gap
  3. Channel stacking
    Email paired with LinkedIn replies 27% more within-team
  4. Send mode
    Individually sent InMails get roughly 20% more replies than bulk
  5. Sender identity
    Contested: one source shows +50%, another shows no lift
Your own relationship history moves reply rates more than any market signal, and sender identity is the least settled lever.

These figures come from different studies with different denominators, so read the order roughly, not as exact rankings.

Lead

Use your standard first touch at its normal place in the sequence. A multi-step sequence still beats a single message: one benchmark found a 4-step sequence earns 2x the replies of a single email. If the score includes raid points, Lead means contacting them now. One practitioner suggests reaching out on days 3 to 10 after a layoff. One compensation analysis put median severance at 14 weeks, so a laid-off candidate may not be in a hurry.

Differentiate

Pick the strongest lever this candidate allows, in this order:

  1. Warm re-contact if your CRM shows a past reply. The modifier already makes this Lead, but the message itself should refer back to that earlier conversation.
  2. A second channel. Pair email with a LinkedIn touch. Channel scarcity points tell you whether you have a second channel at all.
  3. Individual sends. Send the InMail one at a time, never as part of a bulk send.
  4. A different hook. No public data measures hook lift on crowded candidates. Use it, but don't expect a known effect size.
  5. A different sender, as a test. The evidence conflicts. One recruiting-CRM vendor reports a 50%+ lift from SOBO and says only 22% of recruiters use it. A separate test of 1.5M+ emails found leader-signed email slightly below recruiter-signed, and on LinkedIn 9.38% against 14.60%. That vendor's explanation is that hiring managers usually write to hand-picked candidates, which makes SOBO look better than it is.

When a candidate scores 2 on channel scarcity, the useful fix is usually finding a second route, not rewriting the message. Refolk searches GitHub and the open web as well as LinkedIn, and public work such as commits or community activity can give you something to reference besides an InMail. A commit-email or community-DM touch has no published lift figure, so track it yourself.

Rank Down

Move the candidate below the uncrowded equivalent you named on the score card, and set a date to score them again at the next touch window. Raid points fade after 90 days, and density bands change with the req. Today's Rank Down can be next month's Lead.

Rank Down changes the order, not the slate. A crowded candidate is a later touch, not a lost one.

How the crowding read goes wrong

Most scoring errors come from two places: treating a positive signal as crowding, and treating a single number as more precise than it is. Check each failure mode below before you trust a verdict.

  1. The rank test doesn't reproduce. Results depend on the seat, the filters, and network distance, so two sourcers get different ranks for the same query. Check: use the written query spec and score from one shared seat.
  2. Field-weight myths. Guides that say the headline carries 30% or 60% of the ranking weight have no primary source. LinkedIn has never published field weights. Check: don't adjust your obvious query to fit a weighting nobody has documented.
  3. Openness scored as crowding. Marking open candidates as picked over is the most expensive mistake on this list. The evidence says they are under-contacted and reply 56% to 76% more. Check: openness is always a separate modifier.
  4. SOBO treated as the fix. One source shows +50%. Another shows no lift on email and a drop on LinkedIn, and suggests hand-picking inflates the apparent effect. Check: A/B test sender within the same slate before making it standard.
  5. Layoff timing misread. A recent layoff can look like crowding when the candidate is already re-employed: 37% are back in work within a month. Check: read the current role and its start date before you score raid pressure.
  6. Market density drift. Talent Insights hiring demand is relative, and changing the location filter changes it. Check: keep the region set the same for the whole req.
  7. CRM "never replied" read as neutral. Never-repliers reply 85% less often. Check: score never replied as a step toward Rank Down, and never contacted as neutral.
  8. Vendor bias. Every large reply-rate dataset here comes from a company selling outreach volume. Check: when you cite a figure, cite its method and denominator too, for example first touches on a 30-day reply window.

One more trap applies to the whole method. When a sourcer's reply rate drops, it is tempting to blame crowding for all of it. A low response rate on the same candidates every recruiter messages can also mean the message is weak or the fit is wrong. Look at the score cards. If most of the slate scored Lead and replies are still low, crowding is not the problem.

Keeping the read calibrated

The thresholds in this model are conventions, not published standards, so treat them as hypotheses and test them against your own reply data each quarter. That turns a borrowed model into one calibrated to your market.

How to keep it current:

  • Tag every first touch with its verdict in the ATS, then compare reply rates by verdict after a quarter. If Lead and Differentiate reply at the same rate, the cut points are too loose.
  • Re-run the density band per req. It is relative and goes stale as postings open and close.
  • Re-check the source data mechanisms. The saturation report may be updated, and LinkedIn's openness figure has already changed three times. When a source changes, update the reference figures, not the method.
  • Re-check LinkedIn's InMail policy and result cap on the Recruiter help pages before you rely on the 13% threshold or the 1,000-result limit.
  • Keep a log of calibration disagreements. When two scorers disagree, record which line they split on. Repeated splits on the same line mean that line needs a tighter definition.

Before the first touch goes out

  • The req has a written obvious-query spec with seat, query string, filters, date, and result count.
  • The candidate's rank was taken from the shared seat named in the spec.
  • The profile's current role and start date were checked before raid pressure was scored.
  • The density band comes from the req notes, not from a fresh per-candidate guess.
  • Openness flags are recorded as modifiers and added no crowding points.
  • CRM history is recorded as prior reply, never replied, or never contacted, not left blank.
  • Every channel on the scarcity list has been verified, not assumed.
  • Every Rank Down verdict names a specific uncrowded equivalent of similar fit.
  • The verdict and the specific sequence change are written in the ATS where the next scorer can see them.

Questions practitioners ask

Should I skip the top of search results when sourcing?

No, not on rank alone. Default-query rank is one of six crowding proxies, worth at most 2 of 9 points. A top-ranked candidate in a low-density metro with two reachable channels still scores Lead. "Start on page five" ignores openness, your own reply history, and channel access, and those can move reply rates more than rank does. Score the candidate, then decide where they go in the order.

Are Open to Work candidates already picked over by other recruiters?

The published evidence says no. In one outreach dataset, candidates with no current role replied 76% more by email and 56% more on LinkedIn, and they got less contact than their share of the market would predict. LinkedIn has also published that Open to Work candidates are about 35% more likely to respond. Treat openness as a pull toward Lead, never as a crowding point.

Does sending from the hiring manager fix low response rates on crowded candidates?

That is not established. One recruiting-CRM vendor reported a lift of 50% or more from send-on-behalf-of. A separate test across 1.5M+ emails found leader-signed email at 5.27% against 5.53% for recruiter-signed, and a lower rate on LinkedIn. Neither measured crowded candidates specifically. If you want to use it, A/B test it within one slate before you rely on it.

How soon should I contact candidates after a layoff?

Fast, but check the profile first. One study found 37% of laid-off tech workers re-employed within a month and 79% within three, with 45% of those recruited into the new job. One practitioner suggests reaching out on days 3 to 10. How long recruiter attention stays high after a layoff has not been established. If the profile shows a new role dated after the event, score the new employer instead.

What should I do when a strong candidate scores Rank Down?

Rank Down only applies if your slate has an uncrowded candidate of similar fit. If it does, sequence that person first and re-score the crowded one at the next touch window. If no equivalent exists, downgrade the verdict to Differentiate and use the strongest lever you have, which is usually warm re-contact or a second channel. Rank Down changes the order. It never drops the candidate.

How do I prioritize candidates for outreach when everyone runs the same search?

Score each candidate on the six crowding proxies against one written obvious query. Then order the slate: strong fit with low crowding goes first, strong fit with high crowding gets a differentiated touch, and adequate fit with high crowding goes behind uncrowded equivalents. Most first touches across the market go to uncontested people, so a slate built from the obvious query is where crowding concentrates.

Try it on the search you came here for

Stop building boolean strings. Just describe the person.

Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.

  1. 01Describe them

    One plain sentence. Role, city, stack, stage, whatever matters to you.

  2. 02I read the web live

    GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.

  3. 03You read the shortlist

    Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.

  • No boolean, no filters, no seat to buy. One box.
  • Read at search time, so a profile updated yesterday counts today.
  • Every step visible as it runs, every name with its reason.

500 free credits on sign-up. No card, no demo call. See real searches.

Read next