One hire is 10% of the company.

The seventh employee of a young company is often chosen in a coffee shop, in an hour, by a founder working from questions half-remembered from their own last interview and a strong feeling about the person across the table. Nobody involved would sign a supplier contract this way. Yet the decision being made is who one-seventh of the company will be: a seventh of its output, its payroll, its habits, and its bench. Startup hiring is at its most informal at exactly the size where each decision weighs most, and the weight is not a mood but a fraction. Call it the arithmetic of headcount: employee #10 is 10% of the firm, #25 is 4%, #50 is 2%, shares computed by nothing more than dividing one by the number of seats.

The informality has understandable origins. There is no recruiter, no interview training, no time; the pipeline is a dozen warm introductions; and founders tend to trust their own read of people, or they would not be founders. So hiring your first employees runs on gut feel and questions borrowed from whatever company the founder left. Meanwhile the organizations for which any one hire is a rounding share of headcount, the thousand-person firms, run structured processes tuned by decades of selection research. The allocation is backwards: the least evidence and the least structure are applied precisely where each decision weighs most.

Every result the selection literature settled on enterprise samples applies with more force where errors cannot average out. A ten-person company cannot run a validation study of its own, and does not need to: the field's cumulative evidence is available to inherit, instrument by instrument, practice by practice. What follows is the smallest set of hiring practices that claims that inheritance — a process sized for a company that is too small for any other kind.

Early-stage hiring mistakes concentrate instead of averaging out

A large employer's selection errors behave like a portfolio. Hire a thousand people a year with an imperfect process and the misses are diluted by the hits; no single decision moves the institution, and the cost of imperfect prediction becomes a line item, smoothed across divisions and quarters. Most hiring research was conducted in that world, and most hiring advice quietly assumes it. A company of ten is not a smaller version of that world; it is a different regime, one in which every decision is visible in the aggregate because the aggregate is ten people.

Figure 1 computes the regime's defining quantity: the share of the company that one seat represents, at every headcount from five to one hundred. The values need no study and carry no error bars: each is one divided by the number of employees. At ten people, the next hire is 10% of the firm (computed: one seat in ten). By the fiftieth hire the share has fallen to 2%, computed the same way, and the curve has flattened into something an enterprise would recognize. Everything this playbook recommends happens to the left of that flattening, in the range where a single decision is a visible percentage of everything the company is.

Concentration does not stop at the seat itself. The seventh employee interviews the twelfth, onboards the fifteenth, and sets the tone the twenty-fifth walks into; in a small company, early hires are the process by which later hires happen. A weak early decision therefore repeats, panel by panel, in every decision it touches afterward. What a bad hire costs in money and drag is the cost-of-a-bad-hire briefing's subject; the point this article adds is about denominators: whatever the bill, it lands on a company with no parallel teams to absorb the work and no slack to carry the seat while it is re-filled.

What the field settled is already yours

The evidence for structure was not produced by companies like yours, and that is its virtue. Selection research spent the better part of a century measuring which hiring methods predict job performance, on cumulative samples no single employer could assemble (Schmidt & Hunter, 1998); the running scoreboard, method by method, is the flagship briefing's subject. Nothing in those findings requires an HR department to be true. They are facts about interviews, tests, and human judgment, and they hold in a coffee shop as firmly as in an assessment center. Startup hiring's founding error is treating smallness as an exemption from findings that were never about size.

Four settled results bear directly on how founders hire by default. The improvised conversation feels like evidence from the inside, and the unstructured-interview briefing assembles what it predicts from the outside, including the experiments in which adding a free-form conversation made forecasts worse. Weighing everything holistically feels like diligence, and the comparison literature finds that combining the parts by a pre-set rule predicts better than synthesizing them in the head (Kuncel, Klieger, Connelly, & Ones, 2013), a record the mechanical-versus-holistic briefing unpacks. Hiring for fit without measuring it becomes hiring for resemblance to the people already in the room (Rivera, 2012), and the culture-fit briefing traces where resemblance takes a small team. And because a new person joins a team rather than an org chart, who the first nine are changes what the tenth adds, a question the team-composition briefing treats as measurable in its own right.

Nor is borrowing the field's evidence a workaround. The profession's validation principles recognize cumulative findings and transportability (a demonstrated case that evidence established elsewhere applies to your role) as accepted grounds for adopting a selection procedure without running a study of your own (Society for Industrial and Organizational Psychology, 2018). What the standards ask in return is modest and portable: know what each measure is for, administer it the way its evidence assumes, and keep a record of both (American Educational Research Association, American Psychological Association, & National Council on Measurement in Education, 2014). That duty reads as bureaucratic until the first disputed decision, and it does not shrink with headcount.

The seat you are filling is already changing shape

Below fifty employees, the seat actually being filled is usually a compound noun: someone to "own growth," "run operations," "do product and support until we split them." These are accurate descriptions of real work. The job description, though, is often copied from a company where the role has been a specialism for a decade, so the page advertises a specialist while the seat demands a generalist. The correction is free: define the next twelve months of the seat, not its job title — what must exist a year from now that does not exist today, and which parts of that this person owns.

The definition will not survive contact with growth, and should not be expected to. Figure 2 sketches the usual trajectory: the first marketing hire owns every strand of the function on day one; within the first year, one strand splits off to a dedicated hire; by the second, another has been absorbed by a newer team, and the seat has narrowed to the specialist core its occupant proved strongest in. That narrowing is drift rather than failure: the role was defined against a company that no longer exists, partly because the seat helped change it.

The practical consequence is for measurement. A measured profile (ability, personality at working resolution, motivations) read against the role as defined today is pre-hire evidence a small company can actually collect, and it keeps a working life after the yes: when a strand splits off and the seat narrows, the move is to re-read the existing profile against the new definition, a practice the assessment-data-after-hiring briefing builds out in full. The re-read is cheap, the drift is predictable, and the alternative is judging a specialist by the ghost of a generalist role.

Specialization also changes which evidence applies. The moment a strand becomes a seat of its own (your first dedicated sales hire, say), the role stops being generic and the literature gets specific; the sales-hiring playbook is this series' worked example of reading the evidence for a single role family. The generalist-to-specialist drift is, among other things, a schedule: it tells you when to stop hiring on the generic evidence and start hiring on the specific kind.

An SMB hiring process sized to the company that runs it

An SMB hiring process has to run without a recruiter, inside founder calendars, at the pace of candidates who hold competing offers; whatever it asks for must justify itself the first time it is used. The stack that survives those constraints fits on one page. It borrows its validity, spends founder time only where judgment genuinely helps, and leaves a record behind it.

It begins with the page from the previous section: the role, written down before anyone is met. One page is enough — outputs for the first year, the strands the seat owns, the strand most likely to split off first. That page does double duty: it is the target the assessment scores against, and it is the artifact that makes every later disagreement adjudicable.

The measurement layer is one validated multi-construct battery, sat by every finalist under identical conditions. Its whole point is provenance: the instrument arrives already validated on samples your company could never assemble, which is precisely the evidence a small company cannot generate for itself. Administered identically, it makes the third candidate comparable to the first in a way that three differently improvised conversations never are; scored against the written role (role-referenced, in the trade's term), it answers the question the coffee shop was trying to answer by feel. What the output reads like is a sample report away.

The conversation stays, and changes genre. Questions chosen in advance, asked of everyone, notes taken against anchors rather than adjectives, scores recorded before anyone in the room says what they think. The mechanics (question design, anchoring, panel discipline) belong to the running-structured-interviews briefing; the small-company version needs nothing a founder cannot set up this week. The conversation and the battery are then combined by a rule agreed in advance (weights first, candidates second), and the rule's output is the default, overturned only in writing. References close the loop as verification: specific questions about specific claims, decided before the call is placed.

Figure 3 stacks the whole of it on one page, beside the dimmed column of what is deliberately missing. No local validation study: at feasible headcounts such a study mostly measures sampling noise, and the disciplined alternative, adopting transportable evidence now while local records accumulate for later, is the local-validation playbook's subject. No assessment center, no home-made situational test, no bespoke scoring formula: each is machinery whose upkeep exceeds this company's scale, and each has a validated, borrowable counterpart already in the stack. The absences are load-bearing. A short stack the founder actually runs beats a fuller one that slides back to coffee-shop defaults by the third hire.

The objections that keep startup hiring in the coffee shop

Founders rarely dispute the evidence for structure. They dispute its relevance, on grounds that feel specific to their situation, and the grounds deserve answers in their own terms instead of an appeal to what large employers do.

"We are too small for process." The comparison being drawn is between process and nothing, and nothing is not the alternative on offer. An improvised hire still consumes a founder's week: the introductions, the coffees, the second conversations, the deliberation afterward that reopens every question because no criterion was ever fixed. Against that, the minimal stack adds a page and an agreed rule, both set up once and reused by every hire that follows. Weigh the setup against what a single wrong yes takes out of a ten-person company and the comparison is not close. The informal alternative is an undisclosed process: the same demands on the calendar, none of the comparability, and no record of why anyone was chosen.

"We hire for fit, and fit cannot be tested." Half of that is right, and it is the half worth protecting. Whether a person will do well in this particular company is a genuine question, and a small company has more riding on the answer than a large one does. But fit judged across a table degrades reliably into fit by resemblance, and a mirror makes a poor criterion. The distinction the culture-fit briefing draws separates the version of fit that can be stated as a requirement (how this company decides, argues, and ships, and what a new person will have to do inside all that) from the version that records how much the founder enjoyed the hour. The first belongs on the role page, where it can be asked about and scored. The second is what fills the vacuum when nothing has been put on the page at all.

"Speed matters more than rigor right now." Speed is a real constraint, and the stack in Figure 3 is shaped by it: the battery runs while the founder does something else, and a structured conversation is shorter than an unstructured one because it knows in advance what it is for. In any case, assessment is seldom the slow part of hiring at this size. The slow parts are the week after the interviews, when two founders discover they were weighing different things and reopen every conversation, and the quarter after a wrong yes, when the search is run a second time and the seat sits empty throughout. Fixing the criteria in advance shortens the first and makes the second less likely, which is speed by any definition a calendar recognizes.

Beneath the objections sits one assumption: that structure is a luxury good, sensible once a company can afford it. The history runs the other way. Structured selection exists because unaided judgment about other people proved unreliable, and that finding is about how human beings read human beings. Highhouse (2008) traced why the finding fails to change behavior even among professionals who know it: the barrier is not ignorance of the evidence but confidence in one's own reading of people, which no amount of contrary data seems to dent. A founder has that confidence in its purest form, having been right about hard things often enough to trust the instrument. Owning equity confers no exemption from it, and neither does being employee number one.

Fix the process while the company is small enough to change it

Every move below can be set up in an afternoon and then used for years, which is the property that makes them worth taking seriously at this size.

Put the role on a page before the first conversation. Outputs for the first year, the strands the seat owns, and the strand most likely to split off first. Everything downstream points at that page: it is what the battery is scored against, what the interview asks about, and what a stalled debate is settled by. A role nobody can describe on a page has not been thought through yet, and the candidates will be the ones who find that out.

Every finalist sits the battery, under conditions that do not vary. Exceptions are what destroy the comparison, and they arrive wearing good manners: the warm introduction it seems rude to test, the candidate everyone already likes, the founder's former colleague. Each waiver excuses precisely the person the comparison most needs to cover, because a strong prior about someone is the thing measurement is there to check.

Scores are written before anyone speaks. In a room this small, whoever speaks first sets the anchor, and the founder's voice is the loudest one in it. Independent scores, written against the anchors and recorded before anyone says what they thought, take a few minutes and preserve the independence that makes two opinions worth more than one. Discussion then does the work it is good at: surfacing what one person saw and another missed.

Combine by rule. Agree what each component is worth while the candidates are still hypothetical, apply the weights when they are real, and let the result stand as the decision unless someone can state in writing why it should not. The override is the useful part of the discipline, because a reason committed to writing can be examined later against how the hire actually worked out.

The profile is re-read whenever the seat changes shape. The hand-offs in Figure 2 are the schedule: each time a strand leaves the role, the person is holding a job the original definition no longer describes. Re-reading the existing profile against the new definition takes a conversation, and it is the difference between a development plan aimed at the seat as it is and one aimed at a role that stopped existing a year ago.

The paperwork is worth keeping. The role pages, the scores, the rule, the decisions, and eventually how each hire worked out: at this volume they document nothing, but they are the only raw material from which a company can later ask whether its own process predicts anything for its own roles. That question belongs to a larger company than this one, and the records it will need are being generated, or not, today.

Startup hiring gets one real advantage from being small: nothing has to be rolled out. There is no applicant-tracking configuration, no panel to retrain, no policy to socialize across offices and regions. A decision to hire on evidence can be taken in a single conversation and be in force by the next candidate, and it will never again be this easy to take.

The hires are not the only thing that compounds. The seventh employee is chosen once; the method that chose them is chosen once as well, and it is still running at the fiftieth hire, in panels the founder no longer sits on, against seats the founder never defined. Every company already has a hiring process. The only decision left is whether the one in use was picked on purpose.