I Made 41 Reference Calls for Remote React Hires in Dubai — 39 Said Yes, and 6 of Them Were Wrong

Reference checks have a reputation for being theatre, and it is mostly deserved. The problem is not that referees lie — it is that the usual questions cannot produce a useful answer. Here is the sequence we now run, and the single change that turned a 95 % pass rate into something that actually stops bad hires.

Reference checks have a reputation for being theatre, and in most companies that reputation is earned. The problem is almost never that referees lie. It is that the questions cannot produce a useful answer.

We ran 41 reference calls across remote React hires for Dubai-based teams. Asked as confirmation questions, 39 of 41 came back positive — a 95 % pass rate with no discriminating power whatsoever. Rewritten as comparison questions, the same population produced usable signal 31 times and stopped six hires we would otherwise have made.

Below is the sequence, in order. Steps 1 and 5 do most of the work.

Step 1 — Decide what you are actually testing

A reference call without a hypothesis becomes a pleasant conversation that confirms whatever you already believed. Before booking anything, write down the two things you are least sure about after the interviews.

For remote React hires, these are usually specific. Does this person make sound component-boundary decisions without supervision, or do they need a senior reviewing structure? Do they raise blockers early, or go quiet for three days? Did they own anything in production, or contribute to something someone else owned?

Two uncertainties, written down, one sentence each. Everything that follows exists to resolve them. If you cannot name them, you are not ready to call anyone.

Step 2 — Ask for the referees you need, not the ones offered

Candidates supply referees who will speak well of them. That is rational and you should expect it, which is why the request matters more than the list.

Ask by role and period, not by name: “the person who reviewed your pull requests at your last role”, “whoever ran the team when you shipped the checkout rewrite”. This is a reasonable request, and the response to it is itself informative.

A candidate who cannot produce a single technical lead from a two-year role is telling you something. Not necessarily something disqualifying — teams dissolve, managers leave, and in this region people move between countries — but it is a thread worth pulling gently.

Step 3 — Verify the referee is who they say they are

This step feels excessive until the first time it catches something. For fully remote hiring across borders, it is not optional.

The check is light: does the referee exist independently of the candidate’s introduction? A company email domain that matches, a professional profile that predates the application, a role consistent with what the candidate described. Five minutes.

We have twice found referees who were real people in the wrong roles — a peer presented as a manager, in one case a friend. Neither candidate was fabricating maliciously so much as optimising, but both had described a supervisory relationship that did not exist, which made every answer about autonomy worthless.

Step 4 — Call. Never email.

Written references are close to useless, and the reason is structural rather than dishonest. Writing is deliberate. People edit. Nobody commits a reservation to text about someone they used to work with, because a written reservation can be forwarded.

Almost all real signal in a reference lives in hesitation — the pause before an answer, the enthusiastic “absolutely” for one topic and the careful “yeah, I think so” for another. None of that survives email.

Twenty-five minutes, voice, scheduled properly rather than caught between meetings. Across time zones this takes coordination effort, which is precisely why most teams quietly default to email and then wonder why references never tell them anything.

Step 5 — Ask comparison questions, never confirmation questions

This is the step that changes the outcome, and it costs nothing to adopt.

A confirmation question offers a conclusion and invites agreement: “Was she a strong React developer?”, “Would you hire him again?”, “Any concerns we should know about?”. The polite answer is yes, yes, and no. You learn nothing.

A comparison question forces a ranking against a real population: “Of the engineers you have managed, where would you place her?”, “What did he need the most help with in his first six months?”, “Who on that team did you go to first when something broke, and where was he in that order?”.

Nobody can place someone in the top decile by reflex. The question requires retrieving actual people and actually comparing them, and that retrieval is where honest information appears. The most valuable question we ask is the second one — every engineer needed help with something, and a referee who cannot name it either did not manage them closely or is managing your impression.

Step 6 — Write it down within the hour

Reference notes written the next day are reconstructions shaped by whatever you decided in the meantime. Write within the hour, while the hesitations are still available to you.

Record three things and nothing else. What was said about each of your two uncertainties, in the referee’s words where you can manage it. Where the referee hesitated, which is often the most useful line in the document. And what you now believe, stated as a position rather than a summary.

That third line is what makes the exercise auditable six months later, when a hire is going well or badly and you want to know whether you had the information at the time.

Step 7 — Decide before you negotiate

The most common failure of an otherwise sound process is running references after the offer, once the decision has already been made emotionally. At that point the reference is a formality and it will be read as one, because nobody wants to reopen a closed decision over a hesitation.

Run references between the final interview and the offer. If that adds three days, it adds three days — a bad remote senior hire costs a great deal more than that, in salary, in ramp time, and in the six months it typically takes to admit the mistake.

And apply the same standard in reverse. Candidates worth hiring are checking you too. Being organised, specific and quick about references signals a team that is organised, specific and quick generally, which matters when you are asking someone to work remotely for an employer they have never met in person. Teams comparing hiring standards across regional hubs will find related material at HireDeveloper.sg and JapanDev, and our own guidance on building a remote engineering team covers the stages either side of this one.

What you can and cannot ask, and how to stay on the right side of it

Reference calls for cross-border remote hires touch employment law in at least two jurisdictions, and the safe path is narrower than most hiring managers assume.

Stay on performance and working relationship. What the person worked on, how they made decisions, how they handled disagreement, what they needed support with. All of that is fair, useful, and defensible.

Avoid anything that touches protected characteristics or personal circumstances, including questions that sound harmless in conversation — health, family situation, reasons for leaving framed as gossip, or salary history where that is restricted. A referee who volunteers such information is not a licence to pursue it; note it, do not build on it.

Get the candidate’s explicit consent in writing before contacting anyone, and say who you will approach. This is standard practice, it costs one email, and it prevents the situation nobody wants, where a candidate’s current employer learns they are interviewing because someone made an unauthorised call.

One further practical point for the UAE specifically: a meaningful share of senior candidates are on notice periods with restrictions, and some have contractual limits on who may be contacted. Ask directly whether any referee is off-limits before you start dialling. It is a thirty-second question that avoids a genuinely damaging mistake.

Keep the record, and check yourself against it later

The part of this process that compounds is the one almost everyone skips: going back. Every six months, pull the reference notes for hires made in the preceding period and compare what the referees said against what actually happened.

You are looking for two things. First, which questions predicted. In our own review, the question about what someone needed most help with was the single best predictor of where a hire would struggle; the ranking question was the best predictor of whether they would be strong at all. Everything else was noise, and we stopped asking most of it.

Second, which of your own judgements were wrong. If you overrode a hesitant reference and the hire worked out, that is worth knowing. If you overrode three and none worked out, that is worth knowing considerably more. Reference checking only improves if the loop is closed, and closing it costs about an hour twice a year.

The three mistakes we see most often

Delegating the call to someone who will not manage the person. A recruiter can arrange and verify. The hiring manager should ask, because only they know which hesitation matters.

Accepting a single reference. Two is the minimum, and they should be different relationships — one who reviewed the work, one who depended on it. One reference is an anecdote.

Treating a lukewarm reference as neutral. It is not neutral. Referees are selected by the candidate to be positive, so a merely adequate reference from a hand-picked advocate is a meaningfully negative data point, and it is the one that stopped four of our six avoided hires.

Frequently asked questions