The Ideal Candidates Are Not In The Funnel

Where a role has real leverage and the people who could fill it have real options, companies want the person who would change the outcome, then write a document that can't reach them. Every stage after inherits the omission, and recruiting data cannot reveal it, because the evidence never enters. The change is one of order: the company goes first, with real information, before it asks for anything. Obvious, and almost nobody does it.

The hiring authority tells me they can't fill the role that matters most to them this year: the one where they can name, precisely, what doesn't happen until someone is in the seat. "Our head of engineering needs to move into a chief architect role. They've lost the confidence of some of our top devs. We need an engineering lead with strong people skills who can recover it." The market is thin, they say. Everyone good is locked up.

I ask to see the pipeline. It isn't thin at all. Plenty of applicants, a dozen screened, four finalists, one offer, one decline. Four months on the role. They point at the numbers as evidence about the market.

That pipeline isn't evidence about the market. It's evidence about the willing: the people prepared to be processed by the company on the strength of a job description that omitted what was good about the role.

This is not a claim about hiring in general. For plenty of roles there is genuine applicant surplus, the cost of a false positive is high, and rigorous screening is correct.

It applies where two conditions come together: leverage and optionality.

The first is leverage. The outcome of the role varies enormously with who fills it: an impact hire, in the plain sense that one person materially moves what happens. The founding engineer, the first designer, the person who opens a market, the person who re-energises a team. These are genuine properties of some roles, not job titles. Plenty of senior-titled positions in saturated functions have modest leverage, and plenty of high-leverage roles carry unremarkable titles.

The second is optionality. The people capable of filling it have alternatives, and a current situation they're not desperate to leave. The most common source of optionality is career equity: accumulated, portable reputation, a track record that precedes them, a network that surfaces opportunities unasked. But it isn't the only source. Scarce capability produces optionality even before reputation catches up: the unknown high-performer is usually being courted internally, and has a trajectory they would be giving up to leave.

Both conditions must be true together. Leverage without optionality is ordinary hiring with higher stakes, and screening hard is correct. Optionality without leverage means the role probably doesn't justify the effort this argument requires. Where the two meet, the rest of this explains almost everything.

The failure that cannot be seen

Start with the thing that makes this so persistent: for high performers with options, the failure is invisible in the company's own data.

The standard critique of over-engineered hiring processes is that good candidates drop out. Six rounds, three take-homes, a two-week silence between stages, and the strong ones, who have alternatives, leave the process partway through. That does happen, and it is visible while it happens. Stage-by-stage drop-off is a number any ATS will happily produce.

But high performers with options don't drop out at stage four. They never enter. There is no row in the system. They didn't read past the first few lines of the job description, or they read it and closed the tab, and no artefact of that decision exists anywhere anyone can look.

Which means the recruiting data is survivorship data. Every metric available (time to hire, pass-through rate, offer acceptance, quality of hire) is computed exclusively over the willing. High performers with options are absent from the system that would report the problem. So the report comes back clean. The funnel looks healthy. The conclusion is that the market is thin, when the honest conclusion is that the company surveyed the willing and generalised to everyone.

And where outcome measures do exist, they are self-assessed by the people whose judgment is under assessment. Regretted and non-regretted attrition is the clearest case: the company decides, after the fact, which of its own losses it minds about. That is not a measurement. It is a position, recorded later by someone with an interest in it. The same is true of most quality-of-hire scoring, which resolves to a manager rating a decision they made themselves.

This is why the problem survives contact with competent people. It isn't that the TA function is ignoring the evidence. It's that the evidence of this particular failure cannot appear in the reporting they're using to look for it.

The trade nobody with options will take

So why don't they enter?

The usual answer is friction, the process is too long, too slow, too demanding of unpaid work. That's true but it undersells it, because it frames the cost as time. Time is the smaller half.

The larger half is that entry is conditional. A candidate cannot decide to take a job. They can only ask to be considered for one, and then be assessed, ranked and possibly refused, on a timetable somebody else sets. What that condition costs depends on what the person asking already has. Someone in a position they value is putting a settled thing against a refusal they cannot appeal. Someone who wants out is risking nothing they mind losing. The process is therefore expensive for one reader and cheap for the other, before either has judged the role. The only thing that offsets the cost is the value of what is on the other side.

Take the proven first, the ones whose reputation the market has already priced. Career equity is an asset accumulated over time: portable reputation that means someone can text three people and have two conversations by Friday. It's what makes them worth hiring in the first place. And a standard application process asks them to set that asset down at the door and re-prove themselves from zero, to enter as an unknown quantity, evaluated and ranked among strangers.

That's not a scheduling inconvenience. It's a status claim that contradicts their actual position in the market. Framed as a trade, it's bad, and it stays bad no matter how good the job is.

The unknown high-performer, whose reputation hasn't caught up to their ability, faces a different version of the same wall and arrives at the same place. Nobody is asking them to disclaim an asset, because the market hasn't priced it yet. Their problem is that a standardised process is structurally incapable of seeing what makes them outstanding. It sorts on legible proxies (titles, tenure, brand-name employers, keyword overlap) and by construction they don't index well on any of those. So entering means joining a long queue for a lottery they have poor odds in, while the internal trajectory they'd be walking away from is one where people already know what they can do. That's not an appealing expected value either, and the people best placed to judge it are exactly the ones who decline.

Both routes, the proven and the unknown high-performer, lead to non-entry. The population that will accept the trade is the willing: low optionality, disliking their current role, nothing better running in parallel. That correlates with a lot of things, but ability isn't reliably one of them. And no amount of employer branding changes the arithmetic of what's being asked.

The document is the source code

There is a step between those two things that is omitted, and it is what makes the failure so difficult to correct.

The job description is not only the first thing candidates see. It is the origin artefact for everything that follows. The recruiter's outreach is written from it. The screening questions are built on it. The interviewers form their sense of the role by reading it. Whatever is absent at that first point is absent everywhere downstream, because nothing downstream generates the missing information independently. The failure doesn't get corrected as the process runs. It gets replicated.

The obvious defence: told that high performers with options don't apply, a hiring team will reasonably answer that they don't rely on applications: they have recruiters who approach people directly, they go and find them. That would settle it if the outreach carried something the posting didn't. It almost always doesn't. It is derived from the same document and inherits the same omissions.

I have read hundreds of these messages and the pattern is consistent enough to be predictable. The outreach is the JD made shorter. It is never better than the posting, because compressing it drops what little specificity there was and keeps the adjectives. All it does is add more applicants without optionality. The requirements survive the edit. The substance doesn't.

Consider what the approach actually contains. A title. A company. A description of the team as collaborative and the business as growing. An invitation to a conversation. What it cannot contain, because the source material never held it, is the thing the recipient needs in order to answer: whether this is a career move. Scope, measured against what they hold now. Scale, against what they run now. Where it leads. What doing it well would open up for them afterwards that they cannot do today.

Go back to the role I opened with. The hiring authority described a team whose confidence had been destroyed by the person still in the seat, and a job that consists of rebuilding it. That is a real problem, and it is unusually well understood by the person describing it. None of it reaches the posting. What reaches the posting is "strong people skills", which is the same fact with everything informative removed. The candidate who could do that work reads a competency where a problem should be, and has no way to tell this role from the many others using the same phrase.

Notice that nothing was concealed. Nobody decided to leave it out. The document simply has no place to put it, so it converts into the nearest available requirement and the specificity is gone before anyone writes a word of outreach.

The space is not empty

Something fills it, and it is usually one of two things. Young companies lead with mission: the problem in the world, the values, the belief that this matters more than other work. Mature companies lead with brand: the name, the scale, the logos, the implied prestige of association. Different registers, identical omission. Neither says what is on the desk on the second Tuesday.

The mission version carries a particular hazard. It can shade into something like false vocational awe, the suggestion that the work is a calling and that treating it as a job is a small failure of character. That framing is not neutral. Its practical effect is to make ordinary questions feel impolite: the band, the reporting line, the hours, what happened to the last person. Someone with options notices they are being pre-emptively shamed out of asking things they fully intend to ask. They don't argue with it. They close the tab.

And it produces a signal nobody intends. The louder the mission and the vaguer the role, the more a strong reader assumes the first is covering for the second. Nobody writes three paragraphs about purpose ahead of a job with an obvious, exciting, well-defined problem in it. The problem would be the pitch.

And a problem is what this population is scanning for. Not mission, which they have. A problem worth two years: something broken, unbuilt or at stake, sized to what they can do. That is the only scarce good in the exchange, and it is the thing most reliably left out, because it is the only part that takes work to write.

Mission is also the one claim this reader is equipped to price. Anyone worth approaching at this level already works somewhere with a credible one, so what gets quoted at them is usually close to what they have, and they have watched from the inside what it amounts to: a sentence they have never seen decide anything. It doesn't land as empty. It lands as evidence that this company writes the same sentences as the one they are already in, which is an argument for staying.

That doesn't make mission language a mistake. It works, and it works on people who don't have one already: those early in a career, those choosing an industry, those looking for something to belong to. Those are the willing as well, and the willing is a position rather than a quality. It describes what someone's alternatives look like on the day they read the advert, not how good they are. Plenty of them are excellent and they are often the right hire. It inverts only on the population that already has a credible mission, where it stops being an attraction and becomes a reason to stay put. And it inverts hardest on the ones with the longest experience of hearing it.

No way to check

There is a further problem, and this one sits in what the document does contain rather than in what it leaves out. What it contains is a snapshot. A static description of a dynamic job is inaccurate the day it is published, and more inaccurate every week after. The gap isn't in the document, it's between the document and the job.

Which sorts the readers. The willing don't price that gap, because they aren't buying the future state, they're buying an exit, and an exit is worth the same whatever the job becomes. High performers with options are the only readers for whom the gap is expensive, because trajectory is the entire question they are asking. They won't bet on it, and they can't check it. So they pass.

The willing take the call regardless, because an hour is cheap and the alternative is staying put. High performers with options need to know before they spend the hour.

The obvious move is to ask. But the only channel is the person who sent the message, and they were handed a requirements list and asked to find people. They can answer the questions that decide nothing: the band, the process, the size of the team. The ones that decide it, what the scope is against what this person already holds, where the role leads, what doing it well would open up, were never in the document and were never in the briefing either. What the sender can do is move them to the next stage, because that is what they are measured on. So the offer is: enter the process, and find out whether it was worth entering. They don't reply.

It would be wrong to conclude the sender couldn't tell. Plenty of them can. A recruiter who has run a dozen searches in a function knows a thin brief when one arrives, and can usually name what is missing from it. Knowing isn't the constraint. Going back is.

Because going back means telling a hiring manager, who is busy and senior and believes the specification is finished, that it isn't, and that the search cannot properly start until they have sat down and worked out what the role is for. That is the right conversation and it is occasionally had. What makes it rare is that it cannot be had every time. A recruiter who challenges one brief is doing the job well. A recruiter who challenges every brief becomes an obstacle, and obstacles get worked around: the next requisition goes to an agency, or straight to a job board, or to whoever returns CVs by Thursday without the seminar. The working relationship is the recruiter's capital, and it is not spendable at that rate.

The person best placed to see that the document is thin is the person least able to say so. The cost of saying it lands on them immediately. The cost of not saying it lands on somebody else, much later, in a form nobody traces back.

And the non-reply leaves no trace either. This is the same omission one layer out. An unanswered message produces nothing to review: no record, no rejection, no stage in a funnel.

What can be counted

None of which makes the defence insincere. By the measures the TA function is held to, it worked. Messages went out, replies came back, a shortlist formed, the role was filled. Those are outputs, and outputs are what the TA function is assessed on. Whether high performers with options ever seriously considered the message, and whether the person hired turned out to be the right one, are outcomes, and almost nobody is measured on those.

I have run talent acquisition at four scale-ups, and this is the part of the argument I would defend hardest. A function assessed on outputs will keep producing them, and will keep reporting success, and will be right to on its own terms. The applicants who arrived were real. That they were the willing is not a fact the reporting can contain.

This is not a management failure that better metrics would correct. Outcomes in hiring are close to unmeasurable. Whether a hire worked depends on the manager, the brief, the team, the market, and a dozen other variables the TA function cannot control. The verdict arrives a year or more later, usually as an opinion held by the person who made the decision and is invested in having been right. There is no control group, because the hire not made leaves no data, for the same reason high performers with options leave no row.

So the function is assessed on what can be counted, because the thing that matters cannot be counted, and the measures offered in its place are the self-assessed ones already described, opinions wearing the clothes of data. That isn't a defect in the reporting. It is the same absence, surfacing one more time.

The applicants are the willing, and the interviewers often sense it. What they can't tell is why: the pool in front of them is the only evidence they have, so the weakness reads as a screening problem, and the answer to a screening problem is another round. Stages get added when a hire comes apart and never removed when one works, so the process only ratchets. The document is never a suspect, because it isn't in the room when the failure shows up.

Which leaves exactly one thing the function can answer for. Not whether the hire succeeded, for all the reasons above. What it can answer for is the composition of the pool. Who was in it. Who wasn't. Whether high performers with options ever considered it at all.

That is the part the TA function genuinely controls, and it is the part nobody examines, because the reporting counts who arrived and has no way of representing who didn't. It is also the only part of this that can be checked without waiting a year and without asking anyone to grade a decision they made themselves.

Nobody writes for the pool

Which raises the obvious question. If pool composition is the one thing the function can answer for, and the document is what sets it, then the document ought to be the most carefully written artefact in the process. It is reliably the least.

The explanation offered earlier was that the document has nowhere to put the relevant information, so the specificity converts into the nearest available requirement and is gone before anyone notices. That is true, and it is not all of it. Even where there is somewhere to put it, something takes it back out, and it is not carelessness or a shortage of time.

The clearest version is the one where the incentive is printed on the page. A contingency agency is paid when a placement is made and paid nothing for anything else. What its advert has to achieve is a curriculum vitae arriving in the agency's system. It does not have to describe this particular seat, because the sender holds several, and a reader who is wrong for this one may be right for the next. A specific advert narrows that. A vague one does not. Which is why those postings read the way they do: a confidential client, a leading firm in the sector, a competitive package. The usual reading is that somebody was lazy, or that the client insisted on discretion. Neither is needed. The document is written correctly for what it has been asked to do.

The newer version of the same arrangement states it out loud. A recruitment platform advertises a client's role, withholds the client's name at the client's request, and closes by instructing the reader that the next step is to speak to the platform's own agent. Someone who wants that job cannot apply for it. They can join a network, from which they may later be introduced. The advert is a gate, and behind the gate is a twenty-minute conversation with the platform. That document never had to persuade anyone to take the job. It had to persuade them to take the call, and it is written to precisely that standard: nine sentences about the work, four hundred words about the platform.

All of which would be a complaint about intermediaries if it stopped there. It does not. The same arrangement runs inside companies that use no agency at all, and it is harder to see because no fee changes hands and nobody involved is selling anything.

A TA function is measured on applications, pipeline, time to fill, cost per hire. Every one of those improves when the advert is wide, and several of them worsen when it is narrow. A document that names the problem precisely is legible to a smaller number of people and produces fewer applications, later. A document that asks for strong people skills produces more, sooner, and no report in use distinguishes between them. So the vague version does not merely survive. It wins on the measures, every time, and goes on winning for as long as those are the measures.

That is a harder claim than the one made a few paragraphs ago, which is why it is worth separating. The earlier point was that the reporting cannot represent a good document, so it fails to reward one. This one is stronger. The reporting rewards the opposite. Someone competent, optimising honestly against the targets they have been given, will produce a thinner advert on purpose, and will be right to, and will be able to explain why in a quarterly review. Nothing has to go wrong for this to happen and nobody has to be careless. It is what the instrument is for.

One party in all this has an interest pointing the other way. The hiring authority owns the outcome. They are the only person who gains nothing from a large pool of the wrong people, because they are the one who will spend the next two years with whoever comes out of it. Their interest and the document's purpose are aligned, and they are almost never the one writing it.

This is not a claim that a hiring authority left alone with the document would produce a good one. Most would not, and for an ordinary reason: the writing is a specialist skill, and the material sits in their head in a form nobody has ever asked them to organise. The claim is narrower than that, and it is about alignment rather than competence. Theirs is the only seat in the process where the incentives do not argue for a thinner document, and it is the seat that does not write it.

Everyone is hiring for the mean

Everything so far has depended on high performers with options never entering. The argument doesn't need that to hold. Some of them enter anyway, by referral, or on a good day, or because the job was interesting enough to override the reluctance, and the process loses them a second time by an entirely different route. This is the same mechanism seen from the inside.

A scorecard is an averaging instrument. It has to be, its purpose is comparability, and comparability requires that every candidate be measured on the same axes. So the axes are chosen for what can be assessed across many people: communication, structured thinking, relevant experience, stakeholder management. All of them reasonable dimensions. Then the scores are averaged, across dimensions and again across interviewers, by consensus, and the shortlist ranks on the aggregate.

Every step of that compresses a person into a single number. And single numbers reward the absence of weakness, not the presence of anything remarkable.

Consider what that does to someone whose value is concentrated. They are not merely good, they are extraordinary at one or two things and unremarkable at the rest, which is what being extraordinary usually looks like up close. On a rubric with no column for the thing they are extraordinary at, that concentration is invisible. They score adequately across the board. Nobody rejects them; they simply never register a strength. And in a ranked process, adequate everywhere loses to good everywhere, every time.

The candidate who wins is the one with no low scores. That person is often excellent and frequently the right hire. But note what has been selected for: evenness. The process reliably surfaces the person who is good at everything and exceptional at nothing, because that is the only shape that scores well once the scores are averaged.

Which is entirely correct, for a seat where the outcome doesn't vary much with who fills it. Most seats are like that, most hiring is high-volume and high-uncertainty, and a risk-reducing instrument is the right tool for the job. Averaging is the right tool for reducing risk and the wrong tool for finding upside. The problem arises when the same process gets pointed at roles where the outcome swings hard on the individual, and there it selects against precisely the property the hire was meant to supply.

So the loss happens twice, by one mistake. Outside the funnel it shows up as invisibility: they never appear, so nobody learns they existed. Inside it shows up as averaging: they appear, and rank in the middle. Both are the same error: treating an outlier as a comparable, and then trusting the comparison.

Fit has a direction

There is a third failure that isn't invisible and is arguably more important, and it gets misread rather than missed.

Performance is a property of a match rather than of a person, so predicting it requires two assessments: whether the candidate fits the role, and crucially, whether the role fits the candidate. Almost all hiring processes run the first exhaustively and the second last. The company's questions are answered by the midpoint. The candidate's questions (who they would report to, what they could decide without permission, the real state of what they'd inherit, where the seat leads, the number) only get answered near offer stage, after exhaustive interview rounds all going one way: does this candidate fit the role.

So the candidate invests four weeks, performs well, and assembles a conclusion at the end: this is a sideways move dressed as a step up. That is the moment a counter-offer becomes very easy to accept.

From the company's side it looks like they used the offer as a bargaining chip, or lost their nerve. Mostly the process never gave them a reason to move beyond money, and against money alone staying wins, because staying carries no risk. The company didn't lose a negotiation. It never opened one.

There is a further reason the neglected direction is the one that matters, and it doesn't stop mattering once someone has signed. Candidate fit is fitted to a snapshot: the role as it was described on the day the specification was written. Most roles don't hold that shape. The function grows, the strategy moves, and anyone worth hiring reshapes the job around their own judgment inside the first year. So the assessment the company ran exhaustively was an assessment against a role that stops existing, and in a dynamic job it starts decaying almost immediately.

Job fit is fitted to a trajectory instead: whether this is progress for this person, given what they can already do and what they are trying to become. That question survives the role changing, because it was never tied to the shape of the work on any particular day. It is the more durable of the two assessments, and it is the one almost nobody runs.

Which gives a version of the eighteen-month bounce, the senior hire who comes apart around month eighteen, that requires nobody to have made a mistake. Someone assessed carefully against a role that no longer exists by month twelve comes loose on schedule, and there is nothing in the file to point at. The interviews were rigorous. The scores were fair. The role moved, and nobody had asked whether the person wanted the place it was moving to.

Selection presumes a pool

Here's the clearest statement of the error I can manage.

The hiring process is a selection instrument. Selection presumes a pool: it operates on candidates who are already there, and its job is to rank and filter them. Everything a good process does well, it does downstream of a pool existing.

For high performers with options, the pool doesn't exist yet. It has to be created, and the only thing that creates it is persuasion. Not filtering, not ranking, not rubrics. Persuasion, in the ordinary sense of making a specific case to a specific person about why this particular problem is worth their next two years.

Companies get this wrong not by choosing filtering over persuasion but by not noticing there was a choice. They apply a selection instrument at a stage where selection has nothing to operate on, and then read the empty output as information about supply.

Which is what happened to the role I opened with. A team that has stopped trusting the person leading it, and a job that consists of rebuilding that trust: broken, at stake, and sized to one person. It is difficult, it is visible when it works, and it is the sort of thing somebody carries with them for the rest of their career. That is a problem worth two years, and there were people who wanted it. The advert asked for strong people skills.

What can be said is what each version selects for. A document describing the actual problem is legible to the few people who have solved it before and want to again. A document asking for strong people skills is legible to anyone prepared to claim they have some. The company ran the second one and concluded the market was thin. I rewrote the advert, and the outreach that came from it. The people who replied were the ones who had solved it before. What I can't tell you is who the first version lost, because they never entered and left no trace, and that is the whole difficulty.

What follows is a change of order, not a change of method. Every failure in this piece is the same one: something done before the thing it depends on exists. The document before the thinking. Selection before the pool. The company's questions before the candidate's. So if the pool has to be created, the company must go first: the initial contact carries real information rather than requesting it, and it is aimed at particular people rather than posted at a population. Rigour doesn't disappear. It moves later, after there is someone to apply it to. None of that is a technique, and none of it is new. It is what the observation implies once there is nothing to select from until somebody has made a case.

There's also a second-order effect: candidates can't inspect the job before they take it, so they infer it from the only sample available, the way the hiring is run. The process isn't a gate standing in front of the product. It is the first chapter of the product, and it's read as one. A slow, one-sided, poorly-informed process is taken as an accurate preview of what working there is like, and it usually is accurate.

The first sample usually arrives before any person does. One of the better-funded AI companies, with enterprise customers it can name, asks four things of anyone applying to any role: a link to a LinkedIn profile, what makes them excited about the company's mission, a hard problem from their past, and a recent achievement, what they did, and what the outcome was. Three of the four want written answers. None of the four is about the job. Nothing asks what the applicant would want to do in the seat, what they would need to do it, or what would make it worth two years.

The mission question is the one that sorts. Someone buying an exit can answer it, because the answer requires nothing they do not already have: enthusiasm is available to anyone who has read the homepage. Someone assessing a trajectory has nothing to write, and not because the question is difficult. They have not been told the trajectory. They are being asked to supply the reason they would move, at the stage where the company was supposed to supply it, and before anything has come back about the seat. The willing complete it, which is what makes them the willing. Four questions from the company and none available to the candidate, in a single artefact, before either party has met. That is the whole error in the space of one page.

Not broken. Built for someone else.

It matters how this is stated, because the obvious phrasing is wrong and does damage.

"The process is broken" is an attack on the competence of whoever built it, and it invites the objection that is genuinely hard to answer: we really are the buyer, we really are paying, we really do have to choose. All of that is true. None of it is the point. The accurate version is narrower:

The process isn't broken. It's precision-built for a population the company would prefer not to hire from.

It's more correct. Most hiring processes are well-designed for the job they were built to do, which is high-volume, high-uncertainty screening, and they do it competently. The claim is about where the instrument is pointed, not about the people holding it. Which also makes it the kind of claim that can be checked, rather than a judgment that has to be accepted.

That matters more than usual here, because an argument about people who leave no trace can slide into unfalsifiability if nobody stops it. Left alone, every absence of evidence reads as confirmation, and a claim that cannot fail is worth nothing. So here is the test. Name ten people the company would genuinely want. Approach them properly, going first, with real information. If none of them will take a conversation, the constraint isn't posture and I am wrong about the situation. The experiment is the argument's own falsification test, which is the main reason I would rather someone ran it than agreed with me.

The claim, one more time, in its narrow form: where a role has real leverage and the people who could fill it have real options, the company is not running a selection process, because there is nothing to select from until a case has been made. Funnel data won't show this. It structurally can't.