Tech hiring Data & AI
Hire data scientists who answer the right question.
A data scientist uses statistics, experiments and models to answer questions a report cannot: whether a change caused the result, what is likely to happen next, which customers differ and why. We shortlist people assessed on reasoning and on explaining uncertainty, not only on fitting models.
01 The role
What data scientists do all week, and what comes out of it.
A data scientist turns a loosely worded business question into something that can be tested, then tests it. They design experiments, build forecasts and predictive models, estimate the effect of a decision, and report what the evidence supports along with how confident anyone should be. The output is usually a recommendation, sometimes a prototype model.
Fitting a model is rarely the hard part. The hard part is asking a question the data can answer, and not fooling yourself with the result: confusing correlation with cause, stopping an A/B test the moment it looks good, validating a forecast on data it has already seen. A data scientist who cannot say “this does not show what you hoped” is of little use.
The role sits between two neighbours. A data analyst describes what happened, with SQL and dashboards. A machine learning engineer runs models as production software. The data scientist works on why it happened and what will happen, and normally hands a proven approach to engineers. In small teams the lines blur, so be clear which end you need.
What they ship
- A/B test designs, power calculations and readouts with a clear decision
- Forecasts for demand, revenue or capacity, with honest intervals
- Churn, propensity and lifetime-value models used by marketing and sales
- Customer segmentation grounded in behaviour, not demographics alone
- Causal estimates of what a price, feature or campaign change did
- Pricing and promotion analyses that weigh revenue against volume
- Prototype models and written findings a non-specialist can act on
02 Skills and stack
Skills to look for, in three columns.
Data science covers several distinct jobs: product experimentation, forecasting, risk modelling, research. A strong forecaster is not automatically a strong experimenter. Tell us which kind of question you most need answered.
Core
The role cannot be done without these.
- Statistics: inference, regression, uncertainty
- Experiment design and A/B test analysis
- Python or R for analysis and modelling
- SQL, to get their own data
- Predictive modelling and validation
- Problem framing: from business question to testable one
- Communicating findings and their limits
Tooling
Varies from team to team.
- pandas or Polars, NumPy, SciPy
- scikit-learn, XGBoost or LightGBM
- statsmodels; R with the tidyverse
- Forecasting: Prophet, statsforecast, ARIMA and ETS models
- Bayesian modelling: PyMC or Stan
- Jupyter or Quarto notebooks
- Matplotlib, Plotly or ggplot2
- Snowflake, BigQuery or Databricks
- An experimentation platform: Statsig, Eppo, GrowthBook
- Git and reproducible environments
Adjacent
Useful, not required.
- Causal inference: difference-in-differences, uplift modelling
- Deep learning with PyTorch
- Natural language processing and LLM-assisted analysis
- Model deployment and MLOps
- dbt and data pipeline basics
- BI tools: Looker, Power BI, Tableau
- Domain knowledge: finance, marketing, operations, health
03 When to hire one
When a data scientist is the right hire, and when not.
We would sooner point you to a different role than fill the wrong one. Both lists are worth a minute.
Hire a data scientist when
-
You run experiments and argue about what they showed
Tests are launched without a sample-size plan and stopped when someone likes the number. You need someone to design them properly and call the result.
-
Planning depends on a forecast built in a spreadsheet
Stock, staffing or cash decisions rest on last year plus a percentage. A tested forecast with error bounds changes how much buffer you carry.
-
You need to know whether something worked
Revenue rose after the price change, the campaign and the redesign all landed in one month. Separating their effects is causal analysis, not reporting.
-
Dashboards say what happened but not what to do
The analysts are good and the reporting is solid, yet questions about why customers leave or which lever matters most go unanswered.
-
You want to know if a predictive model is worth building
Before funding an ML team, someone should check whether the data contains enough signal and what a model would be worth if it worked.
Look elsewhere when
-
The need is regular reporting and dashboards
Weekly numbers, KPI tracking and ad hoc SQL are analyst work. A data scientist doing it will be costly, bored and soon gone.
Hire data analysts -
Your data is scattered across exports and tools
Without a warehouse a data scientist becomes a part-time, reluctant data engineer. Get the data into one reliable place first.
Hire data engineers -
The model must run in the product, at scale, every day
Serving, monitoring and retraining are engineering. Most data scientists can prototype; fewer should own a production service.
Hire machine learning engineers -
You want a chatbot or a document assistant
Building on language models is software engineering with its own methods. Statistical training helps with evaluation but is not the main skill.
Hire AI engineers -
There is one question to answer, once
A pricing study or a single forecast is a defined project. Commission it, and hire permanently if the questions keep coming.
Post a freelance project
04 How we assess
Four stages, before a data scientist reaches your shortlist.
A CV says what someone claims. We test what they can do, then send you the evidence with the shortlist.
-
Stage 1
Profile review
We look for analyses and models that changed a decision: the question asked, the method chosen, the result and what the business did next. A portfolio of tidy competition notebooks tells us less than one piece of work on untidy company data.
-
Stage 2
Technical test
A timed test covering probability and statistics, experiment analysis, modelling concepts such as validation and overfitting, SQL and Python or R, with practical exercises that involve interpreting results as well as producing them.
-
Stage 3
Conversation with our team
Our team asks the candidate to walk through a project from question to recommendation: why that method, what could have invalidated it, and how they explained it to people without a statistics background. We listen for candour about uncertainty.
-
Stage 4
Review and levelling
Reviewers score accuracy, speed, problem-solving, communication and teamwork, then assign a level using our AI framework. For data scientists communication carries real weight, because a correct analysis nobody understood changes nothing.
-
Then
Your shortlist
Assessed candidates, with notes and scores.
The engine
Evalia
Assessments run on Evalia, the assessment platform we built ourselves. Across the ProDevs network, more than 100,000 assessments have been completed.
The levels we assign are public. Read the AI framework, or see how a hire runs from brief to offer.
05 Seniority
Same title, different scope. What each level can own.
Levels differ in scope more than in speed. Match the level to how much direction you can give.
Junior
Owns a defined analysis
- Carries out an analysis or fits a model when the question and method are set by others
- Writes clean SQL and Python or R, and checks their own data before modelling
- Knows the standard tests and models, and when their assumptions fail
- Needs a senior colleague to frame problems and challenge conclusions
Mid-level
Owns a question, start to finish
- Turns a vague request into a testable question and picks a suitable method
- Designs and reads experiments properly, including when the answer is “no effect”
- Builds and validates forecasts and predictive models that others rely on
- Presents findings to non-specialists and states plainly what remains uncertain
Senior
Owns how the company decides
- Sets experimentation standards, metric definitions and review practice
- Chooses which questions deserve modelling and which a simple cut will settle
- Handles hard causal problems where no experiment is possible
- Advises leadership directly, and mentors analysts and junior scientists
06 Interview questions
Questions for data scientists, and what to listen for.
None of these has one right answer. The way someone answers tells you how they work.
-
An A/B test shows a 3% lift with p = 0.04 after four days. Do we ship it?
A good answer shows: Questions before an answer: was the duration and sample size fixed in advance, were results checked repeatedly, does four days cover a weekly cycle, how many metrics were examined. A candidate who says “it is significant, ship it” will mislead you regularly.
-
Tell me about an analysis where the result was not what the stakeholder wanted.
A good answer shows: A real case, how they checked their own work first, and how they delivered the message. You are looking for someone who held their position with evidence and still kept the relationship, not someone who softened the finding until it meant nothing.
-
Sales rose after we launched a loyalty scheme. How would you estimate its real effect?
A good answer shows: Recognition that members chose to join, so a simple comparison is biased. They propose a holdout group if one can be created, or difference-in-differences, matching or a synthetic control, and state the assumptions each needs.
-
How do you validate a forecasting model?
A good answer shows: Backtesting on rolling time windows, never a random split. Comparison with a naive baseline such as last year’s figure, an error measure that suits the business, and prediction intervals. If the model cannot beat the baseline they say so.
-
Your churn model is 95% accurate. Is that good?
A good answer shows: It depends on the base rate: if 5% of customers churn, predicting “nobody” scores the same. They move to precision, recall and lift in the group you would act on, and ask what the retention team will do with the list.
-
How do you decide between a simple regression and a more complex model?
A good answer shows: They begin with the purpose. If the aim is to explain or to estimate an effect, an interpretable model usually wins. If it is pure prediction, complexity must earn its place against a simple baseline on held-out data.
-
What do you do before trusting a dataset you have just been given?
A good answer shows: Practical checks: how the data was generated, what a row represents, missing values and why they are missing, duplicates, sudden shifts in volume, definitions that changed over time. People who skip this step produce confident, wrong results.
07 Ways to engage
Not only permanent hires. Four ways to work with a data scientist.
A permanent hire is the usual route, not the only one. Choose by how long the work lasts and who should employ the person.
- Placement fee, of first-year salary. One-time.
- 10–15%
- Replacement guarantee on direct hires.
- 2 weeks
- To a first shortlist, for most roles.
- 48h
-
Direct hire
A permanent employee on your contract and your payroll. One-time placement fee of 10–15% of first-year salary.
Tech hiring -
Dedicated team
A data scientist inside a team we assemble and run for you: product, design and engineering on your roadmap, for a monthly retainer.
Dedicated teams -
Outsourcing & payroll
A full-time person in a country where you have no legal entity. ProDevs employs and pays them: 20% of salary if we find the person for you, or 5% of salary, capped at $500 a month, for payroll and compliance on someone you have already chosen.
Outsourcing -
Freelance project
A defined piece of work with a clear end. You pay the agreed price per milestone through escrow, plus a 5% client fee. Free to post.
Freelancers
08 Around the role
A data scientist is one seat. These are the others.
A data scientist is most useful close to the people making decisions: product, growth, finance or operations. Placed in a separate research corner they produce interesting work that nobody acts on. They also rely heavily on whoever maintains the data they analyse.
- Product manager
- Agrees the decision an experiment or model is meant to inform, before any work starts.
- Data analyst
- Shares metric definitions and takes over the deeper questions that reporting raises.
- Data engineer
- Asks for reliable, documented tables and flags data problems found during analysis.
- Machine learning engineer
- Hands over a validated modelling approach to be rebuilt for production.
- Finance and commercial leads
- Supplies forecasts and scenario estimates, with the assumptions written down.
09 Client testimonials
Hear it first-hand, from companies that hired.
Founders, operators and executives describe working with ProDevs, in their own words.
Play video: Client testimonial: Olufemi Seyi, COO, Casafina development company
Play video: Client testimonial: Aproko Doctor, Building AwaDoc
Play video: Client testimonial: JT Liddell, Promenade CEO
10 Questions
Hiring data scientists, answered.
Something else? Book a call (opens in a new tab) and ask us directly.
-
What is the difference between a data scientist and a data analyst?
A data analyst reports and explains what has happened, mainly with SQL and a BI tool, and keeps the regular numbers flowing. A data scientist uses statistics and modelling to predict, to run experiments and to estimate cause and effect. If your open questions begin with “how many” or “what was”, hire an analyst. If they begin with “why” or “what if”, consider a scientist.
-
What is the difference between a data scientist and a machine learning engineer?
A data scientist establishes whether a question can be answered and which method is valid, and often builds a prototype model. A machine learning engineer turns a proven approach into production software with pipelines, serving and monitoring. If your model needs to make decisions inside the product every day, you will eventually need the engineer as well.
-
Do we have enough data to justify a data scientist?
Volume matters less than people assume. What matters is whether you record the outcomes you care about, whether the data is in one queryable place, and whether there are recurring decisions that better evidence would change. A company with modest data and many pricing, growth or planning decisions can get more from a data scientist than one with a large, unused lake.
-
Has generative AI replaced the need for data scientists?
No. Language models speed up coding and summarising, and a good data scientist uses them. They do not design a valid experiment, notice that a comparison is biased, or take responsibility for a forecast. Demand has shifted towards people who can frame problems and judge evidence, and away from those whose main skill was writing model code.
-
Should a data scientist be our first data hire?
Rarely. If the data is not yet collected and modelled in a warehouse, a data engineer or a strong analyst will add more in the first year. A data scientist hired too early spends most of the time on data preparation. Hire one first only if a specific modelling or experimentation problem is central to the business.
-
How does ProDevs assess data scientists, and how fast can we hire one?
Every candidate passes four stages: a review of past work, a timed technical test, a conversation with our team and a final review that assigns a level. For most roles you see a first shortlist within 48 hours of agreeing the brief, and most companies find their hire within two to three weeks.
Talk to ProDevs
Tell us about the role.
Tell us the level and the stack. We will tell you how quickly we can put assessed data scientists in front of you.
Project work
Need it for a project instead?
Not every job needs a full-time hire. Describe the work, get proposals from vetted freelance data scientists, and pay per milestone through escrow. Free to post.