Building a hiring function from nothing
A profitable trading firm came to us to fill two roles. What they were missing was everything behind the hire: the strategy, the screening, the market read, and someone who could run all of it. This is what we built, and what they kept.
Start
Two roles, no apparatus behind them
A four-person algorithmic trading firm, three years in and profitable, working a niche, physically-driven market. The core team came from physics and applied math and had reached the work sideways, through curiosity rather than a conventional path. They had built something that worked. What they had never built was any way to hire for it. No recruiter, no process, no defined roles, no screen. The founder described himself, accurately, as building the ship as he went.
They came to fill two roles at different levels of the team. That framing is almost always incomplete at this stage, and it was here. A founder asking for two hires is usually missing the entire apparatus that produces good hires, and doesn't know it yet, because the gap shows up as "I just need to fill this seat," not "I have no hiring function." The search itself is what exposes the truth, and this one exposed it fast, because it was hard in several directions at once. Each kind of hard pointed at a capability that wasn't there.
Missing
No target. No screen. No way to recognize the failures.
The role was undefined, because no one had defined it. There was no description that matched what the firm needed, because the firm was still working out what these hires would do. The pool was thin before anyone added a constraint: a quantitative foundation, real curiosity, a beginner's mindset, and the ability to communicate uncertainty rather than oversell a result is a rare intersection. Then hard requirements narrowed it further, on-site on a weekly cadence, native-level communication for a small context-heavy team, and US work authorization with no sponsorship driven by an incoming government contract. And they had been burned before, by people who looked right and failed in practice, but the lessons lived in the founder's head as instinct, not criteria you can screen against.
A recruiter executes against a defined target. There was no target. So the work was to build the whole apparatus, from the strategy down to the shortlist, and leave it behind.
An undefined role.
A pool of about eighteen.
Failure patterns no one had named.
Build
The apparatus, not just the hires
The role, defined in terms of the business
We started by turning instinct into structure. The output was a placemat, but calling it a one-page brief undersells what it does. It defined what the hire was actually for: not a list of skills, but a researcher whose work would eventually inform capital decisions, which meant the entire profile was built around trust and intellectual honesty rather than credentials. It set what good looks like at six months and twelve. It translated the founder's named past failures into specific, screenable anti-patterns. And it carried a behavioral question set where every question was tied to a trait and to the exact failure it was meant to catch, with what a strong answer sounds like beside what a weak one sounds like.
The role was a job post and a feeling.
The team measured every candidate against one written bar.
The screening, built from their own scars
Culture had caused the most expensive mistakes, but it lived as a vibe: worries about ego, coachability, and people who already knew the right answer. We ran calibration sessions, not intake calls, and the distinction is the whole job. An intake call collects a wishlist. A calibration session pulls out what the founder doesn't know he knows. We asked him to run hindsight on every hire that had failed, not "what skills do you need," but "tell me about the ones that didn't work, and what it looked like day to day." The patterns came out in specifics: the combative hire who fought details that didn't matter and wouldn't hear out how the team worked, the one who went silent for weeks then surfaced with code only he understood, the capable and well-liked hire who repeated the same fundamental mistake after correction, the researcher who kept producing results that looked spectacular and got defensive when asked to check his own work. None were skill failures. Every one was a pattern, and until it was named it was invisible. We turned those patterns into a layered screening methodology where each signal was tied to a real scar, with concrete examples of good and bad answers so anyone on the panel could hear an anti-pattern coming.
Bad fit was discovered months after the hire.
The discovery moved to before the offer.
The interview architecture
The fragments existed, an async test, a technical screen, a panel, all loosely defined and duplicative. We designed a staged process where each stage had an owner, a focus, and a reason to exist, with realistic pass-rate expectations at every step so the funnel was a plan, not a hope. For the more senior role, the technical exercise was built on messy real-world data and the judgment the work actually demands: validation, debugging, and decisions about scope and fidelity, with a panel designed to watch the candidate think live rather than recite.
Founder time leaked across redundant conversations.
Founder time went only where founder judgment was the point.
The market read, modeled like a market
This is the piece a recruiter never delivers. The hard constraint was work authorization, and rather than guess at it, we sized the local talent market the way the founder sizes the markets he trades. We sourced 151 profiles and classified every one. 54 went through full evaluation, profile, code, portfolio, and verified history, with the remaining 97 cataloged at the surface and labeled as such, so the founder always knew which numbers were verified and which were directional. From the fully evaluated set we measured a real conversion rate, roughly a third reaching top tier, and applied it to project the realistic pool: about eighteen genuinely strong candidates in the entire metro. On the policy question, the analysis was blunt and quantified. Roughly four in five of the local pool were likely authorized to work without sponsorship, probabilistic rather than certain, and relaxing the requirement would add volume at the unreviewed tier but zero confirmed top-tier candidates to the existing pipeline, against real logistics and timeline cost. Before, the sponsorship decision was a gut call. After, it was a tradeoff the founder could see and own. For someone who trades on base rates and confidence intervals, this was hiring spoken in his own language.
Sponsorship was a gut call.
It became a quantified tradeoff.
The decision-ready shortlist
Most agencies forward links into a founder's inbox and call it a slate. We did the first round of judgment. Candidates arrived tiered, each with why they fit, what to verify, and where they might stretch or fall short, plus a clear action for the founder on each one.
Evaluating candidates was triage the founder did at midnight.
A short set of high-value conversations with the judgment already done.
Anonymized snapshots of the actual deliverables
Anonymized snapshots of the actual deliverables. Generic content, real method.
Local talent market: addressable pool
MeasuredTop-tier conversion ran about 1 in 3 of those fully evaluated.
ProbabilisticRoughly 4 in 5 of the local pool were likely authorized without sponsorship. Relaxing the requirement added volume but zero confirmed top-tier candidates.
Their hardest constraint was work authorization, and they were treating it as a gut call. We sized the local talent market the way they size the markets they trade, so sponsorship became a tradeoff they could see. The call got made with eyes open.
Quantitative Researcher
Build and validate the models that inform capital allocation. The work moves money, so the profile is built around trust and intellectual honesty, not credentials.
Owns one model end to end and surfaces its failure modes before they cost money.
Trusted to size positions. Others calibrate against this person's judgment.
Question"Tell me about a result you were unsure of. How did you present it?"
TraitCommunicates uncertainty honestly
CatchesDressing up a guess as a hard number
The role didn't exist yet, just a posting and a feeling. We defined what the hire was actually for and set a measurable bar at six and twelve months. Every candidate measured against one written standard instead of instinct.
Signal: intellectual honesty
A capable past hire produced results that looked excellent, then grew defensive when asked to validate them. The error reached production before anyone caught it.
Tell me about a result you were confident in that turned out to be wrong. How did you find out?
They caught it themselves, through a check they run by habit. The story is about the process, not the embarrassment.
Someone else caught it. Framed as bad luck. Nothing changed in how they work afterward.
Goes quiet for weeks then surfaces with code only they understand. Repeats a corrected mistake. Fights trivial details and won't hear how the team works.
They'd been burned by people who looked excellent and failed in practice, but the lessons lived in the founder's head as instinct. We turned each failure into a signal and the question that catches it. Bad fit now surfaces in the interview, not six months into the job.
Candidate 07
Tier 1Quantitative foundation from an applied-math background. Portfolio shows self-initiated validation work. Communicates uncertainty plainly, without overselling.
Depth of production experience beyond research. Whether a weekly on-site cadence works.
Limited exposure to messy, real-world data.
Recommended action: advance to technical exercise.
Evaluating candidates had been midnight triage. Slates arrived tiered, judgment already done, a clear action on each one. Founder time went only where founder judgment was the point.
Results
The numbers that mattered
realistic strong candidates in the entire metro
hires, still in seat at 5 months
This search was hard by design. The point was never to manufacture the illusion of abundance. It was to find the few who genuinely fit and build enough structure to be right about them.
Kept
The apparatus stayed with them
A contingency recruiter would have taken the first-draft description and started sending resumes, and the same patterns that burned this team before would have walked back through the door, because nothing in the process was built to catch them.
Both roles were filled. Both hires are still with the company five months on, from a search where the realistic pool of genuinely strong candidates in the entire metro was around eighteen people. That is what the apparatus was for: not finding more candidates, but finding the few who fit and could do the work, and being right about them.
And the firm kept the apparatus. The role definition, the screening methodology, the interview architecture, the market model, and the shortlisting method, all built from their own hard-won knowledge, all theirs to rerun for the next wave. Most founders at this stage think they have a hiring problem. They have a hiring-capability problem, and it hides as a single open role. The instinct that made you a good judge of your first few hires lives in your head as feel, and it works right up until you have to hand it to someone else, and then it becomes the one thing you can't delegate, and you're back doing it yourself at midnight. The work is turning that instinct into something a process can hold: searchable, screenable, repeatable. Do it once and the next role is easier to define than the last. That is the entire point. Hiring built this way compounds.
Build the system behind the hire.
If your next hires matter but the process still lives in your head, we can build the role, the screen, the market read, and the search, in a way your team keeps.