We Paid Agency Fees on 11 Hires Before I Ran the 7-Step Test That Decides an In-House Recruiter

Dashboard of hiring funnel metrics on a laptop screen, representing the pipeline numbers that decide whether a Singapore company is ready for its first in-house technical recruiter
William

William

Talent Sourcing Expert · November 15, 2026 · 11 min read

TL;DR

  • •The method: 7 steps for your first in-house technical recruiter in Singapore, starting with the step most companies skip: proving the role exists before writing it.
  • •The screen that decides it: A two-hour work sample on one of your own hard roles. It is self marking, so a panel with no recruiting background can still read the result.
  • •The honest caveat: Below roughly 8 to 10 repeatable engineering openings a year, this hire is premature and buying help per search is the cheaper answer. Step 1 exists to tell you which side of that line you are on.

We did eleven engineering hires through agencies before anybody asked whether we should be doing this ourselves, and when the question finally came up it was framed entirely as a cost question. That framing is why the first attempt failed. We hired a recruiter to save fees, gave her four open roles, and watched her spend three months becoming very good at a process that ran out of work. She left in month seven and she was right to. The method below is what I use now, and step one exists because the expensive mistake is not hiring the wrong recruiter. It is hiring a perfectly good one into a job that does not exist yet.

Step 1: Run the Volume and Repeatability Test First

Two numbers decide this, and neither of them is headcount. Count the engineering roles you will genuinely open in the next twelve months, with a budget behind them rather than a wish, then count how many of those are similar enough that work on one helps with another.

Repeatability is the variable people miss. An in-house recruiter gets paid back by learning a market once and harvesting it repeatedly: the same search strings, a warm network that grows, candidates who said no in March and are available in September. Nine similar backend and platform roles compound like that. Nine wildly different roles, one firmware, one data, one security, one mobile, compound not at all, because each is a separate market and each search starts cold. In that second scenario a specialist agency genuinely beats an employee, and the difficulty of the roles is a reason against this hire rather than for it, which is the opposite of what most founders assume.

My rough line is eight to ten repeatable openings a year as the point where the arithmetic starts favouring an employee, and twelve as the point where I stop hedging. Below that, keep buying per search and keep the option to stop.

Volume alone does not decide it. Repeatability does.How similar the roles are to each otherOpenings per yearall differentnear identical1680Specialist help per searchEvery search starts cold.An employee learns four marketsbadly instead of one well.In-house recruiter wins clearlyNetwork compounds. Process gets built.Declined candidates become a pipeline.Too early, even if the roles repeatGood recruiter runs out of work by month fourand leaves. This is where we went wrong.Thresholds are our working rule of thumb from Singapore searches, not an industry standard. Funded growth plans, attrition and internal mobility all move the line.

Step 2: Write the Job Around Your Actual Bottleneck

There are three jobs hiding under the words technical recruiter, and the wrong one is a waste of a good person. Find out which you have by looking at where candidates disappear.

  • Top of funnel is empty. Few qualified applicants, outbound messages ignored, the same five candidates circulating. You need a sourcer: someone whose craft is finding and persuading people who were not looking.
  • Funnel is full and nothing converts. Plenty of candidates, lots of interviews, offers declined or candidates withdrawing mid-loop. That is not a sourcing problem and a sourcer will not fix it. You need somebody who can redesign the loop and manage hiring managers, which is a different and more senior hire. The symptom set is covered in improving developer offer acceptance rate.
  • Everything works but slowly. Scheduling takes days, debriefs arrive late, candidates go cold waiting. That is coordination, it is a cheaper hire than either of the above, and it is the one people are most embarrassed to admit they need.

Write the job description for one of those three. A posting that asks for all three gets you a generalist who is mediocre at your actual constraint, and in this market it also gets you fewer applicants, because strong recruiters read an unfocused requisition the same way strong engineers do. If the description itself is the thing you are stuck on, the mechanics transfer from writing developer job descriptions.

Where candidates disappear decides which hire you needThree jobs, one job title, and they are not interchangeableEmpty top of funnelFew qualified applicantsOutbound ignoredSame five names circulatingHire a sourcerWrong hire here: a coordinatorwho schedules an empty pipelineFull funnel, no conversionOffers declinedWithdrawals mid-loopManagers disagree at debriefHire a process ownerWrong hire here: a sourcer whoadds volume to a leaking loopWorks, but slowlyScheduling takes daysDebriefs arrive lateCandidates go cold waitingHire a coordinatorCheapest of the three, and theone people are shy about needingDiagnose from your own funnel data over the last two quarters, not from the loudest complaint. The three symptoms often coexist and one of them dominates.If two genuinely dominate at once, the sequencing question is which one blocks the other. An empty funnel blocks everything downstream of it.A first hire covers one of these columns well. Asking for all three in one requisition is how you end up with none of them.

Step 3: Screen Sourcing Craft, Not Agency Logos

Recruiter CVs are the hardest documents to read in hiring, because the people writing them are professionally skilled at making a CV look good. Brand names and placement counts tell you almost nothing: a recruiter at a company everyone wants to work for has been operating with a tailwind that will not exist at your firm.

So ask about mechanism. How did you find the last three people you placed. Walk me through the search. What channel did you try that did not work. What did the message that got a reply actually say, and what did the one before it say.

Strong candidates light up at this question, because it is the part of the job they are proud of, and the answers are concrete: a search that worked because of a niche conference attendee list, a decision to stop using a job board after two weeks of noise, a rewritten opening line that doubled replies. Weak candidates describe systems they had access to and processes somebody else designed. The tell is the grammar: craft answers are full of the first person singular and specific choices, and weak answers are full of the first person plural and tools.

Our expert take

The Singapore specific thing to probe here is how they handle work pass constraints, and probe it for honesty rather than for expertise. A recruiter who promises they can solve your pass problem is either misinformed or telling you what you want to hear, and both are disqualifying. The answer you want is boring: they ask early which roles can be filled by candidates needing sponsorship, they build separate pipelines accordingly, and they tell hiring managers the uncomfortable truth that a search restricted to one candidate pool takes longer. Recruiters who sell certainty on this cost you months, because the pipeline they build looks healthy right up to the point where it does not convert.

Step 4: Set a Work Sample on One of Your Own Hard Roles

This is the step that does the work, and it is the reason a founder with no recruiting background can run this process confidently. Take your most awkward live requisition, give the candidate two hours of paid time, and ask for a search strategy plus ten named prospects with a sentence of reasoning for each.

Three things to read in the output, none of which require you to know anything about recruiting:

  1. Did they disqualify anybody, and say why. A list of ten plausible names is easy. A list of seven names plus three people they considered and rejected for stated reasons tells you they built a model of the role rather than a keyword filter.
  2. Are there names you would not have found. If all ten are the obvious search result for your job title, you have bought a slower version of your own LinkedIn session.
  3. Did they ask you something you could not answer. This is the strongest signal available anywhere in the process. A recruiter who starts sourcing from the job description as written will fill the role you wrote. One who comes back asking which of your five requirements is genuinely a gate will fill the role you actually have.

Pay for the two hours. It is a professional exercise on live work, and the candidates you most want are the ones who decline unpaid samples.

StepWhat it testsPass signalFail signal
1. Volume testWhether the role exists yet8 or more repeatable openingsFew roles, all different
2. BottleneckWhich of three jobs you needOne named constraintA posting asking for all three
3. Craft screenSourcing ability without a tailwindSpecific searches and rejected channelsTools and brand names
4. Work sampleRole comprehensionDisqualifications with reasonsTen obvious search results
5. Pushback testInfluence with engineersAsks what the real bar isFolds, or argues to win
6. First 90 daysLeading indicators movingPipeline and response rates upOnly hires counted
7. Quarterly reviewRole fit and hire fit separatelyPre-agreed numbers reviewedA conversation about effort

Not sure the role is real yet

Send us your next twelve months of planned engineering roles. We will run the volume and repeatability test with you and say plainly whether this is a hire or a purchase.

Get started

Step 5: Test How They Handle an Engineer Who Says No

Half of this job is influence with people who did not ask for your opinion and outrank you on technical judgement. A recruiter who cannot hold that conversation will quietly become an administrator within a quarter, regardless of how well they source.

Role play it, and make it realistic rather than hostile. You are an engineering lead rejecting a strong candidate with a reason that is really a preference: not enough experience with our exact framework, or a career gap, or a background in a different industry. Then watch which of three things happens. Folding is the common failure and it produces a recruiter who stops sending anyone who is not a literal keyword match. Arguing to win is the rarer failure and it produces a hiring manager who stops reading their emails. The behaviour you want is the question: what would this person have had to show you to change your mind. That question converts a preference into a criterion, and a criterion can be screened for next time.

Candidates who have genuinely done this job recognise the exercise and often smile at it. The debrief discipline that makes those conversations possible in the first place sits in running a developer interview debrief.

Step 6: Measure the First 90 Days on Pipeline, Not on Hires

Hires in a first quarter are mostly an artefact of what was already in motion when the recruiter started, which means judging them on that number rewards luck and punishes the person who inherited an empty funnel. Use leading indicators and agree them in writing before day one.

The four I use: qualified candidates reaching first interview per open role, outbound response rate, days from application to first screen, and a soft one that matters more than it sounds, whether hiring managers now turn up to intake meetings prepared. That last item is the real test of whether the recruiter has established themselves as a colleague rather than a service desk, and you can feel the difference in the room within six weeks.

One number to deliberately exclude: total applications. It is easy to inflate, it correlates with nothing you care about, and watching it go up is how companies convince themselves a hiring problem is solved while the same three roles stay open. If you want a broader view of the funnel you are measuring against, our employer brand method covers the inbound half, and the tooling question sits in the guide to building a recruitment platform in Singapore.

Step 7: Review at the Quarter With Numbers Agreed in Advance

Set the review date and the metrics before they start, because a review designed after the fact becomes a conversation about effort, and everybody in that conversation is uncomfortable and nobody learns anything.

The reason this step is separate from step six is that it asks a different question. Step six asks whether the hire is working. Step seven asks whether the role was right, and those two can fail independently. A recruiter missing every leading indicator while hiring managers rave about them usually means the role was specified wrong in step two. Strong numbers and a recruiter who is bored by month five means the volume test in step one was optimistic, which is our own history and is a management failure rather than a hiring one.

Be willing to say the second thing out loud. We were not, the first time, and the cost of that silence was seven months of somebody good doing too little, which is a worse outcome for them than for us.

Our expert take

The pattern we see most often in Singapore is a company hiring its first recruiter at the top of a funded push, twelve roles approved, everybody urgent. Nine months later the push is over, four roles were cut, and the recruiter is maintaining a careers page. The question worth asking before you post the job is not whether you have volume now. It is whether you will still have volume in eighteen months without a new funding round, because that is the horizon on which this hire either compounds or quietly becomes something you manage around. If the honest answer is no, buy help per search and revisit it. Nobody regrets that decision as much as they regret the other one.

Where This Method Stops

Three limits, stated plainly.

  1. Thresholds here are rules of thumb from our own searches, not benchmarks. Eight to ten repeatable openings is where the arithmetic has tipped for the Singapore companies we work with. Your cost of capital, attrition rate and internal mobility all move that line, and a company with heavy internal transfers needs fewer external hires than its growth implies.
  2. This is a method for a first recruiter, and the second one is a different problem. Once you have two, you are designing a function: specialisation, coverage, who owns which hiring managers, and how data is reported. Almost none of the reasoning above transfers to that decision.
  3. No recruiter fixes a loop that cannot decide. If your interview process produces six opinions and no verdict, a recruiter will surface the problem faster and will not solve it. Fix the decision mechanism first or you will have hired a very capable person to watch candidates withdraw.

Frequently Asked Questions

At what headcount does a first in-house technical recruiter pay for itself in Singapore?
Headcount is the wrong variable and it is the one everybody uses. What matters is the number of engineering roles you will open in the next twelve months and how similar they are to each other. A company of forty people opening fourteen broadly similar backend and frontend roles has a clear case. A company of a hundred and eighty opening five highly specialised roles does not, because each of those searches is a different market and a generalist recruiter will be learning from scratch each time while an agency specialist already has the network. As a rough line, somewhere around eight to ten repeatable openings a year is where the arithmetic starts to favour an employee, and the certainty improves above twelve. Below that, buying help per search is cheaper and you keep the option to stop.
Should I hire from an agency background or from an in-house background?
For a first hire, agency background is usually the better bet, with one thing to screen hard for. Agency recruiters are trained on sourcing volume and candidate persuasion, which is exactly what an empty pipeline needs, and they are used to working without the brand pull a well known employer enjoys. What they often have not done is build a process: intake meetings, structured debriefs, a scorecard that survives contact with four interviewers, data a founder can read. So screen for whether they are interested in that half of the job or merely tolerant of it. In-house candidates arrive with the process instinct and sometimes with a pipeline habit formed inside a company whose name opened every door, which does not transfer to a forty person firm nobody has heard of.
What does this role realistically cost compared with paying agency fees?
The comparison people make is salary against fees, and it leaves out the two things that actually decide it. First, an in-house recruiter is a fixed cost that continues when hiring pauses, which it will, so the saving is real only if your volume is durable rather than a single funded push. Second, a good in-house recruiter produces an asset an agency does not: a warm network of people who know your company, a pipeline that survives a cancelled requisition, and a hiring process your managers can run without them. That asset is worth more over two years than the fee differential on any individual search, and it is worth nothing at all if the person leaves in month seven because there was not enough work to be good at. Model it over eight quarters, not over the first three hires.
How do I interview a recruiter when I have never hired one and no engineer on my panel has either?
Use step four, because the work sample is self marking in the same way a debugging exercise is. Hand them your hardest live role, give them two hours, and ask for ten named prospects with reasoning. You do not need recruiting expertise to read the result. You need to check three things: did they disqualify anyone and say why, did they find people who are not the obvious search result for your job title, and did they ask you a question before starting that you could not answer. That last one is the strongest signal in the whole process. A recruiter who reads your job description and starts sourcing has told you they will fill the role you wrote. A recruiter who comes back and asks which of the five listed requirements is actually a gate has told you they will fill the role you have.

Buy the searches while you build the function

We screen developers for Singapore employers per search, which is the right answer until your volume says otherwise. We will also tell you when it does.

Get started