Where Your Candidates Actually Come From

TL;DR

• The core decision: which channels you keep running, and which ones you keep running only because nobody has ever proposed stopping them.
• When doing nothing is right: when you hire rarely, fill from referral without asking, and can name the origin of every hire from memory.
• What has to be true: somebody records where a candidate came from at the moment they arrive, not reconstructed later from a calendar and a guess.
• How the options split: by who does the work of finding the person, you, your employees, a supplier, or the candidate themselves.
• Decision rule: judge a channel on hires it produced for roles like this one, never on applications it produced in total.
• Outcome to expect: fewer channels, each one you can defend, and a straight answer when someone asks why a role has been open for months.

The Spreadsheet That Answers the Wrong Question

A talent lead is asked, in a Tuesday leadership meeting, where the last five engineering hires came from. She has a spreadsheet. It tells her how many applications arrived from each source last quarter, which is the number the system produces by default, and it's not the number anyone in the room wants. The board with the most applications is the one that has produced no hire in over a year. She knows this and cannot prove it, because nothing in the pipeline records the difference between a person who applied and a person who started.

This is the ordinary condition of sourcing at most companies. Channels accumulate. A job board gets used because it was used for the last role, an agency gets called because someone has a relationship with them, a referral scheme runs because it launched two years ago and nobody has authority to end it. Each was chosen for a reason. Almost none of those reasons has been revisited.

The real problem isn't that you're using the wrong channels. It's that the cost of a channel is visible on an invoice and its yield isn't visible anywhere, so the two never get compared. A channel that produces nothing is indistinguishable from a channel that produces quietly, and in the absence of evidence the loudest one wins. This piece is about making the yield visible enough to act on.

Recommended Tools

Best tools for Recruitment & Hiring

G Greenhouse Applicant tracking system L Lever Recruiting & ATS platform L LinkedIn Recruiter Talent sourcing tool

View all recommended tools

When You Genuinely Do Not Need to Act Yet

Your current setup is genuinely fine. You hire a handful of people a year, the roles repeat, and the people you hire arrive through someone who already works with you. Nobody is confused about where candidates come from because there's only one route and everybody uses it. Formalising this would add a process to a thing that is working. Leave it alone and spend the attention elsewhere.

Friction is starting to show. You're hiring into two or three shapes of role now rather than one, and the route that filled the first shape isn't filling the second. You notice this as a slow role rather than as a channel problem: the vacancy stays open, people say the market is tight, and nobody has checked whether the market is tight or whether you're only looking in one place. This is the moment when the cheapest useful thing is a single question asked at offer stage, recorded somewhere durable.

It has become a real cost. Hiring is now a named part of somebody's job. You're paying for at least one channel, probably more, and cannot say which of them produced your last several hires. Roles stay open long enough that a manager has started doing the work themselves, which is a cost that never appears in a recruiting budget and is usually the largest one. At this point the absence of attribution isn't a reporting gap. It's the reason you can't fix the slow roles.

The edge case that forces it. You open a role in a function or a location you have never hired for. Every existing channel was chosen for the shape of hiring you were doing before, and none of them is calibrated for this. Companies discover here that they never had a sourcing strategy, only a set of habits that happened to fit one kind of role, and habits don't transfer. The same thing happens after an acquisition, when two sets of inherited channels arrive in one function and nobody can compare them.

Five Questions This Reader Asks at 11pm

Where did our last five hires actually come from? If you can't answer without opening a system, you don't have attribution, you have a report. The answer matters because everything downstream depends on it: which channels to keep, where to spend, which supplier to renew. Get this wrong and every subsequent decision is guesswork wearing a spreadsheet.

Are we confusing volume with yield? A channel that sends hundreds of applications and no hires is costing you screening time, which is real money paid in your team's attention rather than on an invoice. The reason this persists is that application volume is the number every system reports by default and hire origin is the number none of them reports without being configured to. Volume feels like activity. It's often the opposite.

Would this role be filled faster somewhere else, or is it just hard? These feel identical from the inside and have opposite fixes. A hard role needs patience and a wider brief. A badly sourced role needs a different channel and will stay open indefinitely without one. Telling them apart requires knowing whether comparable roles have been filled from your existing channels before.

What happens if we turn this off? Most channels have never been tested against their own absence. If you can't describe what you would lose, you can't justify what you're paying, and you will keep paying it. The honest version of this question is whether anyone would notice within a quarter.

Who is doing the sourcing work we aren't paying for? Somebody is. It might be a hiring manager messaging people privately, a team member passing a role to a friend, or a recruiter working a list outside the system. That work is invisible, unmeasured, and usually productive. If you don't know it is happening, you can't support it, and you will lose it the moment that person changes role.

Three Honest Categories the Approaches Split Into

Inbound, where the candidate finds you. You publish a role and wait. This covers general and niche boards, your own careers page, and anything else where the first move is theirs. It's right when the role is common enough that qualified people are actively looking, when you can absorb the screening load, and when your employer reputation does something for you. It fails when the people you want are employed and content, because those people aren't reading job adverts. It also fails quietly at volume: inbound produces applications regardless of fit, so a badly specified role generates more work rather than less.

Outbound, where you find the candidate. You identify people and approach them, either yourself or through somebody you pay. This covers direct outreach, headhunting, and agency search. It's right when the population is small and identifiable, when the role is senior enough to justify the effort, and when the people who could do it aren't looking. It fails when nobody has decided what the target looks like, because outbound without a specification is just expensive noise. It also fails when the approach is generic: a message that could have been sent to a thousand people gets treated as though it was.

Networked, where somebody in between does the finding. Referral, alumni, returning applicants, and the people your team already knows. It's right almost always, which is why it's the one channel nearly every company has, and it's the one most likely to be running without any deliberate support. It fails in two specific ways. It scales with the size and shape of your existing team, so a small or homogeneous team produces a small and homogeneous set of referrals. And it goes quiet when people are busy, which tends to be exactly when hiring pressure is highest.

Five Diagnostic Questions You Can Self-Assess Against

Can you name the origin of your last five hires without looking? Do this now, from memory, then check. The gap between what you remember and what the record says is the size of your attribution problem. If there's no record to check against, that's the answer. Most people find they can name two confidently, guess at two, and have no idea about one, which is a worse position than it sounds because the confident answers aren't necessarily the correct ones.

Which channel produced your longest-tenured hires? Not your fastest hires or your cheapest. Look back at the people still with you after several years and ask where each came from. This is slower to measure and far more useful, because it's the closest available proxy for quality. If one route dominates, it deserves more of your attention than the reporting currently gives it.

What proportion of your screening time goes to your highest-volume channel? You won't have this number. Estimate it by asking whoever screens: which source generates the applications you reject fastest? If one channel dominates the rejections, you're paying for it in attention rather than fees, and that cost is invisible in every budget you have.

When did you last add a channel, and when did you last remove one? If you have added several and removed none, you're accumulating rather than choosing. Channels don't expire on their own. Somebody has to end them, and in most organisations nobody has the authority or the evidence to do it, so the set only grows.

Could you fill this role if your main channel disappeared tomorrow? Answer for your most common role, not your hardest. If the honest answer is no, you have a single point of failure in your hiring, and it's usually one supplier relationship or one person's personal network. That's a risk worth naming out loud before it resolves itself.

Six Sourcing Channels, Reviewed

The general job boards

The broad, horizontal places where anyone can post and anyone can apply. They earn their place through reach: for common roles in populated markets, this is where the largest number of actively looking people are, and getting in front of them takes very little effort. Posting is fast, the mechanics are familiar to everyone involved, and no relationship needs to exist before you start. For a role that many people can do and many people are currently seeking, this is often the shortest distance between a vacancy and a shortlist.

Where it genuinely falls short is precision. A general board optimises for matching keywords, not for matching work, so the applications arrive in proportion to how findable the advert is rather than how suitable the applicant is. That cost lands on whoever screens, which makes it invisible in the budget and painful in practice. It also reaches almost exclusively people who are actively looking, which for senior or scarce roles is a small and self-selecting slice of the people who could do the job.

Niche and industry boards

Narrower places tied to a profession, an industry, a technology or a community. They earn their place by trading reach for relevance: fewer people see the advert, but the ones who do are closer to the work, so the screening burden falls sharply. For a specialised role, a niche board can outperform a general one on hires while producing a fraction of the applications, which is the shape you want and the shape your reporting will describe as failure if it only counts volume.

The honest limitation is fragility. A niche board depends on a community remaining active, and communities move. A board that filled roles reliably two years ago can quietly become a place where adverts sit unread, and the symptom is indistinguishable from a hard role. They also tend to be concentrated: one board per discipline, sometimes one per region, which gives you very little to compare against and makes it hard to tell a weak channel from a weak advert.

Employee referral

Your own people recommending someone they know. It earns its place on two dimensions at once. Referred candidates arrive with an implicit endorsement from somebody who understands both the person and the work, which is a stronger signal than anything a screening call produces. And they arrive pre-informed: somebody has already told them what the job is really like, so the conversation starts further along.

It falls short in ways that are structural rather than fixable by effort. Referrals reach the shape of network your existing team has, so a team that is narrow in background produces referrals that are narrow in the same direction, and doing more of it deepens that rather than correcting it. It's also unreliable exactly when you need it: referrals depend on people having spare attention, and hiring pressure usually coincides with your team having none. And a referral that doesn't work out costs more than an ordinary bad hire, because the person who made it is still there afterwards.

Direct outreach and headhunting

Identifying specific people and approaching them yourself. It earns its place by reaching the population no other channel touches: people who are doing the job well somewhere else and aren't looking. For a role where the qualified population is small and identifiable, this is frequently the only route that produces candidates at all, and it gives you control over exactly who is approached.

Where it falls short is effort and skill, both of which are usually underestimated. Outreach done properly means researching individuals and writing to them specifically, which is slow and doesn't parallelise. Done at volume it becomes generic, and generic outreach damages the thing it depends on: people who receive an obviously mass-sent message remember the company that sent it. It also has a long lag. Someone contacted while content may respond months later, which makes it a poor answer to a role that is already urgent.

Agencies and search firms

Paying a supplier to do the finding. It earns its place by converting a fixed internal cost into a variable external one, and by giving you access to a network and a working method you don't have to build. For an occasional senior hire, or a first hire in a market you have never recruited in, this is often the fastest route to a credible shortlist, and the fee buys real work rather than just introductions.

The limitation is alignment, and it's a genuine one rather than a criticism of suppliers. A search firm is paid to fill the role, and you want the role filled well, which are close but not identical. That gap shows up as pressure toward the candidates who are available now rather than the ones who fit best. It also means the market knowledge accumulates outside your organisation: use agencies exclusively for long enough and you lose the ability to hire without them, which is a dependency that is easy to create and slow to unwind.

Returning applicants and silver medallists

The people who reached your final stages before and were not selected, and the people who left on good terms. It earns its place on efficiency that no other channel matches. They know the company, you know them, and you have real assessment evidence rather than a CV. The single most common source of a fast, confident hire is a person who came second for a similar role recently.

It falls short because it requires maintenance nobody owns. A list of strong near-misses decays: people move, their circumstances change, and a note written a year ago tells you less than it appears to. Keeping candidate records also carries real obligations that differ considerably by jurisdiction, and the constraints on how long you may hold that information and what you must tell people aren't the same everywhere. Get local advice on retention before you build this into a system. And it only works if somebody actually maintains the list, which in practice means it works at companies where one person has decided to care about it.

The Decision Table

Most organisations sit in two or three of these rows at once, and the temptation is to solve the loudest one by adding a channel. Start instead with the row that describes your slowest role, and find out whether it is slow because the work is scarce or because you're only looking where you always looked.

Measuring a Channel Honestly

Attribution fails for a mundane reason: nobody records the source at a moment when the answer is still known. A candidate applies through one route, is contacted through another, and by the time they accept, the origin is a matter of reconstruction. The fix isn't a better report. It's deciding when the question gets asked and who is responsible for the answer, then asking it once at that moment.

The moment that works is the offer. It's late enough that the candidate is real rather than a name in a funnel, and early enough that everyone still remembers how the conversation began. Asking earlier produces answers about the first click rather than the actual origin. Asking later produces guesses. And ask the candidate, not the system: the person who applied knows how they heard about you, and the system only knows which link they last touched.

Two records, not one, is the point of the table. The channel an application arrives through tells you where your screening time goes. The channel a hire originated in tells you where your hires come from. They are different numbers and conflating them is the single most common attribution error, because the system reports the first one by default and the second one not at all.

Retiring a Channel Without Losing the Pipeline

The reason dead channels persist is that turning one off feels like a risk with no upside. The saving is modest and visible, the downside is unbounded and hypothetical, so the rational move for any individual is to leave it running. That logic is correct for each decision and wrong in aggregate, which is why it needs a process rather than a judgement.

Run the replacement in parallel for one full hiring cycle before ending anything. A cycle means from opening a comparable role to a signed offer, not a calendar month. Ending a channel between cycles hides the effect until the next role opens, by which time nobody connects the slow role to the decision made two months earlier.

The step people skip is the fourth. A channel that is quietly stopped tends to be quietly restarted by a manager who liked it, and because nobody announced either decision, the spend reappears without anyone choosing it. Say out loud that it is ending, when, and what replaces it.

What to Put in Writing

Sourcing decisions are made in conversations and remembered by whoever was in them. When that person changes role, the reasoning leaves with them and the next person inherits a set of channels with no explanation, which is precisely how habits become permanent. A small amount of written record turns an inherited set into a set somebody chose.

None of this is heavy. The channel list is a page. The point isn't documentation for its own sake, it is that each of these answers a question that otherwise gets answered by whoever happens to be in the room. Note that the last row carries obligations about holding personal data that vary by jurisdiction, so check locally what you may keep and for how long before you formalise it.

Questions to Ask Before You Commit

On the channel itself. What specific role shape is this for? Which of our roles has it filled, and when? If the answer is a category rather than a list of hires, you're being sold reach. A bad answer sounds like a description of the audience size.

On attribution. How will we know this worked? What number will we look at, and when? If nobody can say what evidence would end the arrangement, the arrangement won't end. A bad answer is that you will review it in due course.

On the screening cost. How many applications should we expect, and who is absorbing them? Volume is a cost paid in your team's time. A bad answer treats high application numbers as a benefit without naming who reads them.

On suppliers. Who owns this relationship internally, and what happens if they leave? Which of our roles would we struggle to fill without it? A bad answer is that the relationship is with the company rather than a person, when in practice it is always with a person.

On the trial. What would we run alongside it, and for how long? A channel adopted without a comparison cannot be judged later. A bad answer is that you will just see how it goes.

On exit. What is the notice, and what do we lose access to when it ends? Some arrangements take candidate records with them. A bad answer is vague about what remains yours.

What Getting This Wrong Costs

The visible cost of poor sourcing is a role that stays open, and that's the smallest part of it. The larger cost is that an open role redistributes the work to the team around it, which is invisible in every budget and is where the real money goes. A team covering a vacancy for a quarter delivers less, and some of them start looking, which turns one vacancy into two.

The second cost is that unmeasured channels make every subsequent decision worse. Without attribution, the response to a slow role is to add another channel, because adding is the only move available when you can't tell which existing one is failing. So the set grows, the screening burden grows with it, and the function gets busier while getting less effective. This is why sourcing budgets tend to rise smoothly and results don't.

The third is dependency. Fill enough roles through one supplier or one person's network and you lose the capability to do it yourself, quietly, over about two years. Nobody decides this. It happens because each individual decision to use the easy route is correct, and the accumulated effect is that the easy route becomes the only route.

So before you add anything: are you solving a reach problem, a targeting problem, or an attribution problem? Reach means the right people aren't seeing it. Targeting means they're seeing it and it's not for them. Attribution means you can't tell which. The answers point at completely different actions, and only one of them costs money.

When You Are Ready to Go Further

The work above is the work the function can do on its own, with a page of notes and a question asked at offer. It doesn't need a system and it doesn't need a budget. What it needs is somebody deciding that the question is worth asking every time, and then asking it.

The next step, for a reader who wants to test their channel set against what comparable organisations are actually using, is comparison work. That's the part the function cannot do from the inside, because you only ever see your own results.

HROpsLab publishes independent comparison work across HR tooling, applicant tracking and payroll. We sell nothing, we take no vendor money, and we publish no paid placements. If the next step is to test your assumptions against the wider market, our comparison work is one place to start.

Frequently Asked Questions

Know where your candidates come from before you spend more on finding them.