Define the Role Before You Post It
The first question to answer is: what does success look like in this role at 6 months and 18 months? Not "what skills do we want" — what actual outcomes do we need? This forces clarity about whether you need:
- An analytics-focused data scientist — someone who can take business questions, work with existing data infrastructure, build models, and communicate insights clearly to non-technical stakeholders. Usually doesn't need deep ML engineering skills.
- An ML engineer / applied scientist — someone who can build, train, evaluate, and deploy models in production. Needs strong software engineering skills alongside the statistical foundation.
- A research scientist — someone working at the frontier of a technical domain, typically with a PhD and publications. Appropriate for organisations doing genuinely novel AI research; not what most business-focused data science roles need.
- A data engineer with some ML exposure — often what teams discover they actually need when they realise their data infrastructure isn't ready for modelling work.
The title "data scientist" gets used for all four of these. The skills, experience, and compensation are materially different. Getting this wrong at the job description stage means attracting the wrong candidate pool and wasting everyone's time.
💡 Ask your hiring manager to describe the last three decisions the business made where a strong data scientist would have changed the outcome. Their answer tells you more about what you actually need than any job description template will.
Where to Find Data Scientists
The best data scientists in APAC are often not visible through conventional sourcing. Here's where to actually look:
Kaggle
Kaggle is the most direct signal of applied data science skill available publicly. A candidate who has placed in the top 10% of a Kaggle competition on a relevant domain problem has demonstrated real practical ability in a way that no CV description can replicate. Filter by competition category (NLP, tabular data, computer vision) relevant to your use case. Kaggle grandmasters and masters are internationally recognised at the top of the applied data science profession — if you can attract one, do.
GitHub
Look for data scientists with active repositories showing real projects — not tutorial completions, but actual applied work. Things to evaluate: code quality, documentation, whether the projects show genuine domain curiosity or are just portfolio-filling exercises. A data scientist who has built and published an interesting model applied to a real dataset tells you something meaningful about how they work.
ORCID and Google Scholar
For research-oriented roles or positions where deep mathematical foundations are important, publication records on ORCID and Google Scholar are invaluable. You can assess both the depth of the candidate's technical work and whether they're engaging with current research questions rather than just applying existing techniques.
Conference communities
The NeurIPS, ICML, and ICLR communities produce the world's best ML researchers. If you're hiring at that level, targeted outreach to conference contributors and workshop participants is worth the investment. DataScience SG and similar local meetup communities are valuable for building awareness of your employer brand in the Singapore market.
FreeFindTalent searches across GitHub, Kaggle, ORCID, and 40+ other platforms simultaneously — useful for building a starting longlist across these channels without the manual effort of searching each one individually.
How to Assess Data Scientists Without Wasting Their Time
The data science assessment process at many companies has become detached from reality. Multi-stage processes involving weeks of take-home work, multiple rounds of coding tests, and presentations to committees — all before anyone has had a genuine technical conversation — are a reliable way to lose the best candidates to competitors who move faster.
What good assessment looks like
- A brief technical screen (30 minutes): Have a senior data scientist on your team talk through the candidate's actual past projects. Not algorithmic puzzles — their work. What were the business questions? How did they formulate the problem? What modelling approach did they take and why? What didn't work? This is the fastest and most informative assessment available.
- A focused practical task (2 hours maximum): A real-ish dataset, a defined business question, and a clear brief. The output should be a short analysis with recommendations — not a polished presentation. You're evaluating reasoning, not presentation skills. Two hours, not a weekend.
- A discussion of the practical work (45 minutes): Walk through their analysis together. Ask about the choices they made, what they'd do differently with more time, and how they'd explain the findings to a non-technical stakeholder. This is often more revealing than the analysis itself.
- A team culture conversation (30 minutes): Have them meet one or two of the people they'd work closely with. You're assessing mutual fit — whether there's intellectual chemistry and shared working style.
What to look for that most assessments miss
- How they handle uncertainty. The best data scientists are precise about what their models can and cannot tell you, and are honest about confidence intervals and data limitations. Overconfidence in model outputs is a serious red flag in production contexts.
- Whether they can explain their work clearly. Ask them to explain a model they've built to a hypothetical non-technical business stakeholder. This skill is genuinely rare and genuinely valuable.
- Domain curiosity. Are they interested in your business problem specifically, or just in the technical challenge? The best business data scientists care about the outcome, not just the model.
🚩 Red flag: a candidate who can't explain when their model would be wrong, or who talks about accuracy metrics without mentioning the business decision context. Knowing a model's limitations is as important as knowing how to build it.
Making the Role Attractive to Strong Candidates
Experienced data scientists in APAC have options. The things they consistently tell me they care about when evaluating roles:
- Data quality and access. There is nothing more demoralising for a strong data scientist than spending 80% of their time on data cleaning because the organisation doesn't have its data infrastructure in order. Be honest about the state of your data. If it's messy, that's okay — but framing a data engineering problem as a data science role will end badly.
- A clear path from model to impact. Can models actually be deployed and affect decisions? Or does the data science work disappear into slide decks that business stakeholders ignore? Organisations where data science demonstrably influences decisions attract and retain better practitioners.
- Technical peers. Strong data scientists want to work with other strong data scientists. If your team is thin on genuine expertise, that's worth acknowledging — and your plan for building it is worth sharing.
- Learning and research time. The field moves quickly. Data scientists who can't stay current become stale quickly. Conference attendance, learning time, and access to compute for experimentation are meaningful signals about how seriously an employer takes the function.
The Infrastructure Question You Must Answer First
Before you hire a data scientist, answer this question honestly: is your data infrastructure ready to support data science work? Do you have clean, accessible, well-documented data? A clear understanding of the business questions you want answered? Engineering support to deploy models when they're built?
If the answer to any of these is no, the highest-value hire you can make is probably a data engineer, not a data scientist. Hiring data scientists into an environment where they can't actually do data science work is expensive, demoralising, and results in high turnover. Get the foundations right first.