Do employee referrals help you get a job, and by how much

Somebody has told you to stop applying cold and get a referral instead. The advice is almost never accompanied by a number, and it is never accompanied by an answer to the two questions that actually decide whether you act on it: how much a referral is worth, and which of the people you know are worth approaching.

There are numbers for the first. One American corporation handed researchers its entire applicant pool, which is a rare thing to be able to see, and nine other firms handed over their personnel records. For the second there is no direct evidence at all: nobody has run an experiment on whom to ask. What exists sits next to the question. LinkedIn spent years randomising which connections its members formed, and researchers counted the job moves that followed. That is not advice about whom to approach, but it does contradict the assumption most networking advice rests on.

Six per cent of the applications, twenty-nine per cent of the hires

Meta Brown, Elizabeth Setren and Giorgio Topa obtained the hiring records of a single US corporation in financial services: a few thousand employees, an urban labour market, decades old, hiring across everything from support roles to executives. Their estimation sample is 62,127 applications for 315 advertised positions, which produced 340 hires.¹ The applications run from 2006 to 2010.

Two figures from the paper's summary statistics set the scale: the average posting drew 185.2 applicants and interviewed 6.7 of them.¹ Those averages are computed over a slightly different set of postings from the estimation counts above, so read them as the order of magnitude rather than as arithmetic to divide into the totals. Either way, only a small fraction of the people who applied were spoken to.

The corporation recorded how each applicant arrived: its own website, campus recruiting, internet job boards, an employee referral, their own initiative, or something else. Watch what happens to two of those groups as the process narrows. Job-board applicants were 60% of the applicant pool, 40% of the people interviewed, and 24% of the people who received an offer and were hired. Referred applicants were 6% of the pool, 21% of the interviewees, 27% of offer recipients and 29% of hires.¹

That gap is the case for referrals in one line. The group that supplied six applicants in every hundred supplied twenty-nine hires in every hundred.

Where in the process the advantage appears

Shares like that could be an artefact of referred people applying to different jobs. So the authors modelled the probability of being interviewed, of being offered the job, and of being offered it having been interviewed, with controls for the observable characteristics of the posting: its staff level, its education and experience requirements, the number of applicants it drew and the share of them who were referred, plus a calendar-year effect.¹ Those are the features of a vacancy the data records, not the vacancy itself, so a posting that happened to attract both referrals and stronger applicants is not ruled out.

Against a job-board applicant, a referred applicant was 7.3 percentage points more likely to be interviewed and 2.4 percentage points more likely to receive an offer. Having reached the interview, they were 14.0 percentage points more likely to be offered the job. All three estimates are significant at the 1% level.¹

The first number is the big one in practical terms, because it moves a probability that starts out small. The third is the one that changes the shape of the story. If the gap were only about getting past the pile, it would close once everybody being compared had reached an interview. It did not close. That estimate comes from the 1,811 people in the sample who were interviewed, 428 of whom were offered the job.¹

The honest caveat is large and the authors are clear about it. This is one company, in one industry, in one city, and it is not a randomised comparison. Employees choose whom to refer, and they can be expected to choose people they think will do well. Nothing in the data separates "the referral helped" from "the referred applicants were better bets", so read these as the size of the gap between two groups rather than the size of what a referral would do for you.

One procedural detail is worth knowing mainly for what it does not do. At this employer the latest an employee could claim a referral was the interview stage, when the recruiter reviewed the application, so nobody could attach their name after watching someone get hired.¹ That rules out credit claimed on a finished outcome. It says nothing about who was chosen to be referred in the first place, and because a claim could still land while an application was being read, it does not fully insulate the interview estimate either.

The employer is not doing you a favour

The reason asking feels like begging is that it looks like a one-sided transaction. From the employer's side it is not.

Stephen Burks, Bo Cowgill, Mitchell Hoffman and Michael Housman assembled productivity and survey data from nine large firms across call centres, trucking and high-tech, covering hundreds of thousands of workers, to find out what firms actually get from referrals.² Their answer is specific. Referred workers had similar measured characteristics to everyone else, including cognitive and non-cognitive skills, and similar performance on ordinary metrics. What they did differently was stay: they were 10–30% less likely to quit. And they were better on rare events with large consequences, such as producing patents and not crashing trucks. The authors' conclusion is that referrals let firms select people better matched to that particular job, rather than better in general.²

Some firms pay for it. At the corporation Brown and colleagues studied, an employee whose referral was hired received a bonus of between $500 and $4,000 in nominal terms, most commonly $1,000, with a median around $2,000, payable only if the new hire was still there after six months.¹ That six-month condition is one employer's rule and not a general one.

Whether your own contact is paid anything is a fact about their employer that you cannot look up and they can. Greenhouse's FAQ pushes exactly that question back to the company: "Referral programs are determined by your company. For questions about your company's referral program, including incentives, bonuses, and payout schedules, reach out to your in-house contacts."⁶ Some pay nothing at all.

The money is the variable part. What is steadier is the reason firms bother to run these programmes, which is the subject of the nine-firm study: a person forwarding your name is doing something their employer has set up a process to encourage. That is a better footing to ask from than gratitude, whether or not there is a cheque attached.

What a referral is inside the software

The word "referral" in advice columns covers everything from a formal submission to a friend mentioning your name at lunch. Those are different objects, and at a company whose recruiting software distinguishes them, only one of them is a referral.

Greenhouse, one of the widely used applicant tracking systems, documents its own version plainly. An employee submits a referral through a form on their dashboard, and must attach it to a specific job: "It is not possible to refer a prospect, so choose a job for your referral."⁵ They can only pick a job with a live external posting, unless their permissions extend further.⁵ The form has a Details section asking for background on their relationship with you, and they can tick a box to follow your progress through the stages.⁵ For credit to attach at all, "the candidate's source must be Referral and you must be credited as the referrer".⁶ Once submitted, a referral cannot be edited or deleted.⁶

That is Greenhouse's workflow today, not a description of every employer, and it is not how the referrals counted in the study above were necessarily recorded: that employer is unidentified, its applications date from 2006 to 2010, and what the paper does tell us is that a referral there could be claimed by either party and was verified by the HR department once the candidate reached the interview stage.¹ The two systems are not the same thing. What they share is the part that matters here, which is that a referral was a recorded attribute of an application to a particular job rather than a conversation.

So it is worth making the ask that fits that shape, because it is also the easier one. Find the actual posting first, check it is live, send the person the link and two or three sentences about how you know each other and why this role in particular. If their employer runs a formal programme, you have filled in their form for them; if it does not, you have given them something concrete to pass on. If your problem is that you do not yet know which live vacancies are worth spending a favour on, JobCraftly will search on your terms and show you the ones that genuinely match, which is the step that has to come first anyway.

There is a sequencing point in this, and it runs against the instinct to get your application in first. Where the referrer is credited only when the candidate's source is recorded as Referral,⁶ applying on your own can leave your application filed under a different source before anyone has referred you. So if you are going to ask, ask before you apply rather than after. If you are not going to ask, or you get no answer within a few days, apply anyway: an introduction that never becomes a formal referral is still worth having, and nothing here says an application counts for less because it arrived on its own.

What randomised network experiments actually showed

Everything above assumes you already know someone inside. The received wisdom about which contacts matter comes from a 1973 sociology paper on the strength of weak ties, and the first large-scale experimental test of it arrived in 2022.³

Karthik Rajkumar, Guillaume Saint-Jacques, Iavor Bojinov, Erik Brynjolfsson and Sinan Aral tested it using LinkedIn's own A/B tests. Between 2015 and 2019 the platform ran experiments on its People You May Know algorithm which, as a side effect, randomly varied how many weak and strong ties different members formed. The authors analysed that randomisation across more than twenty million members, two billion new connections and 600,000 new jobs.³⁴

Be precise about what they counted, because it is not a referral. A job transmission is recorded when one member worked at a company, another member later joined that same company at least a year afterwards, and the two had been connected for at least a full year before the move.³ Nobody observes a request being made, a form being filled in, or a referral bonus being paid. So this evidence is about the ties along which job moves travel, not about who says yes when you ask.

Within that, their headline result is a shape rather than a direction. Measuring tie strength by the number of connections two people have in common, the relationship with job transmission is an inverted U: moderately weak ties produced the most job mobility, and the strongest ties the least.³ Measuring it instead by how much two people interact, the weakest ties did the most.³ Plain correlations in the same data said the opposite, which is precisely the point of running the experiment.³

Those are two separately analysed definitions of tie strength, and the paper puts them together in one place: its edge-level analysis found that adding "new moderately structurally diverse ties with weak interaction intensity created the greatest marginal increases in the likelihood of job transmissions".³ In ordinary terms, the connections that did the most work were at a middling structural distance and low on contact: someone you would share a professional world and some acquaintances with, and would not talk to much. That is a statement about what adding a connection caused, not a census of where jobs come from. The experiment nudged people's networks at the margin and counted what followed.

And it cannot tell you what to do about it. It randomised which new connections people formed, not whom anybody approached, so it supports no outreach strategy at all: not asking a moderately distant contact rather than a close one, and not reviving a dormant tie rather than forming a new one. Where it earns its place here is narrower than that. The intuition that a job is most likely to reach you through the people you know best is the one the plain correlations supported and the experiment reversed.

There is one large exception, and which side of it you fall on depends on your industry. The effect varied by how digital the sector was. Weak ties produced more job applications in sectors with higher IT and software intensity, more suitability for machine learning and remote work, and more robotisation. In sectors low on software use and automation, it was strong ties that produced more applications.³ So the weak-tie result is a result about digital industries. Outside them the sign flipped, which is a reason to treat the standard advice as unproven where you work rather than as a rule to turn upside down.

What none of this settles

The firm-level numbers describe one financial-services employer's applications from 2006 to 2010, and are not causal. The nine-firm study covers call centres, trucking and high-tech, and its findings are about what happens after the hire rather than about your odds of getting one. The LinkedIn experiments are causal but bounded: they randomised which connections people formed rather than which contacts people approached, they measure job transmissions and job applications on a single platform, they could not compel anyone to accept a recommendation and so are analysed as intent-to-treat, and the members exposed to the treatments were slightly younger and more active job seekers than average.³

There is also a cost that belongs in the same paragraph as the benefits. Burks and colleagues found that people refer others like themselves, in characteristics and in behaviour, down to unsafe workers referring other unsafe workers.² A hiring channel that works by similarity reproduces whoever is already inside, and if you are not already inside, that is the channel working against you.

None of this makes a referral a substitute for fitting the job. What it does is put a measured gap where a vague instruction used to be. Six per cent of applications, twenty-nine per cent of hires, at the one employer that let anyone count — and a request that amounts to asking one person to pass on a vacancy you have already found for them.