Time since we first saw it
Day 72
Up longer than most. Similar jobs usually come down between day 8 and day 60.
From 12,096 similar postings that came down. Coming down is not always a hire.
Almaty, Kazakhstan
Live, first seen by us 72 days agoLocation type: On-siteFrom the employer’s posting
Employment type: Full-timeFrom the employer’s posting
Department: Engineering & ProductFrom the employer’s posting
Build churn and downgrade prediction for subscribers, and repeat-purchase models for credit-pack buyers
Model uplift to find users whose decision an offer actually changes
Design and keep running randomized holdouts for discounts, offers and campaigns
Own customer lifetime value on contribution margin, split by cohort, plan and acquisition discount
Design and read experiments on pricing, packaging, offers and lifecycle with causal methods
Build churn and downgrade prediction for subscribers, and repeat-purchase propensity for one-time credit-pack buyers
Model uplift, not propensity.
Design the randomized holdouts that make uplift trainable and measurable in the first place, and keep them running permanently
Own LTV: cohort-based, censored honestly, stated on contribution margin rather than gross revenue, and split by plan, segment and acquisition discount.
Design and read the experiments behind pricing, packaging, offers and lifecycle
From the employer’s posting.
These dates show when we found the posting. Check the employer’s website to confirm it is still accepting applications.
Day 72
Up longer than most. Similar jobs usually come down between day 8 and day 60.
From 12,096 similar postings that came down. Coming down is not always a hire.
10 open jobs
Open jobs by the week we first saw them, last 12 weeks.
Add your resume and we’ll mark which of these tools you already use, and point out anything in the posting that could rule you out.
- Genuine causal literacy: randomized holdouts, incrementality, Qini and uplift curves, selection effects, and why a lift measured before-and-after is usually not a lift. - Strong SQL and Python: cohorts, funnels and retention on raw event data without help; gradient boosting; and the boring parts — feature pipelines, retraining cadence, a scoring job that runs on a schedule and doesn't rot. - Understanding of subscription plus one-time purchase mechanics: recurring vs one-off revenue, refunds, plan changes, the deferred value of unspent credits, and why revenue and margin disagree about the same customer.
Complete your application on jobs.ashbyhq.com. The employer’s form will show what is required.
Already applied? Track this application
Higgsfield AI is the fastest-scaling generative AI company in history, hitting $1B in annual revenue run rate, 25M+ users worldwide, 6M+ generations per day, and powering 390 of Fortune 500 brands. We're building at the absolute frontier of AI-powered video creation and next-generation creative tools. Joining Higgsfield means becoming part of a high-impact team shaping the future of AI-native experiences, at a company that isn't just moving fast, but rewriting what fast looks like.
Right now we treat 25M+ people as one user. The same paywall, the same discount, the same lifecycle email, the same retention effort. And we mostly conclude that a campaign worked by looking at what the people we sent it to did next — which is not the same thing.
This role makes the company act per user, and know what that action was worth.
You own both halves of that: the models that decide who we aim at, and the causal reads that say whether it landed. Who is about to leave. Who will come back on their own without us spending anything. Whose decision a discount actually changes. And what a payer is worth over their life, once the discount, the refunds and the compute they consume are all inside the number.
You make sure:
Discounts and credit vouchers go to the people whose behaviour they change, and nowhere else
We know who is about to churn early enough to do something about it — and we know whether the something worked
Lifetime value is a number we can defend by cohort, by plan and by the discount a cohort was acquired on, not one blended average that hides all three
Every model and every campaign has a holdout behind it, so we can always state what it is worth in money
Churn, retention & lifecycle propensity
Build churn and downgrade prediction for subscribers, and repeat-purchase propensity for one-time credit-pack buyers — two different machines, not one model with a flag on it.
Know the difference between a model that predicts churn and a model that reduces it. AUC earned by detecting users who already stopped using the product is worth nothing.
Hunt leakage relentlessly. Cancellation-adjacent features will hand you a beautiful offline number and a useless system.
Pair every score with an intervention and a holdout. The deliverable is a retained user, not a ranked list.
Uplift modelling & offer targeting
Own incrementality on everything we aim at a user: discounts, credit vouchers, trials, upgrade prompts, win-back campaigns, retention saves.
Model uplift, not propensity.Targeting the users most likely to convert spends margin on people who were going to convert anyway; the entire value of this work is finding the users whose decision the offer changes.
Design the randomized holdouts that make uplift trainable and measurable in the first place, and keep them running permanently — including the ones marketing will ask you to switch off.
Respect the guardrails. An offer model optimizing conversion alone will happily find the accounts that convert at negative gross margin, and it will find the abuse rings first. Margin and abuse signals are constraints on the objective, not a later cleanup.
Work with Legal on what may be personalized. Personalized pricing and discounting touches consumer-protection and consent rules in several of our markets — you should want that conversation, not route around it.
Lifetime value & payer economics
Own LTV: cohort-based, censored honestly, stated on contribution margin rather than gross revenue, and split by plan, segment and acquisition discount.
Model the parts that actually move it — refunds and chargebacks, credit breakage and expiry, plan migration and downgrades, and the compute an unlimited plan consumes.
Make it a decision input, not a slide: what we can pay for a user, which discount depth pays back, which plan we should stop selling.
Keep it current and versioned. A stale LTV curve is worse than no curve, because people act on it.
Experimentation & causal measurement
Design and read the experiments behind pricing, packaging, offers and lifecycle — with the product analytics team, not in parallel to it.
Reach for the right tool when a clean experiment is impossible: difference-in-differences, matching, switchback and geo tests, synthetic control — and name the assumption each one is buying.
Separate "users who do X retain better" from "making users do X improves retention." Every time, out loud.
Kill your own results. A finding that doesn't survive a second look should die at your hands, not in a board deck.
How we ship
Own the work end to end: framing the decision, features, training, evaluation, the scored audience landing in the tool that sends the offer, monitoring, retraining.
Models ship as batch scores and policy rules into lifecycle tooling, the paywall and pricing — plus the readout that says what they earned. There is no real-time serving stack to babysit.
Monitor for drift. New model launches, price changes, discount campaigns and seasonality all move the ground under a deployed score.
Write down what each model is worth, in money, and keep that number current.
Experience shipping models that changed what the business did to real users— a targeted campaign, a pricing rule, a retention programme. Notebooks and offline benchmarks are not this.
Depth in at least two of: churn and propensity modelling, uplift & causal ML, LTV and subscription economics, experimentation at scale.
Genuine causal literacy: randomized holdouts, incrementality, Qini and uplift curves, selection effects, and why a lift measured before-and-after is usually not a lift.
Strong SQL and Python: cohorts, funnels and retention on raw event data without help; gradient boosting; and the boring parts — feature pipelines, retraining cadence, a scoring job that runs on a schedule and doesn't rot.
Understanding of subscription plus one-time purchase mechanics: recurring vs one-off revenue, refunds, plan changes, the deferred value of unspent credits, and why revenue and margin disagree about the same customer.
Commercial judgment: you name the decision and the metric before you pick the model, and you know when the answer is a rule rather than a model.
Pragmatism. A segment rule shipped this month that lifts retention beats a two-quarter uplift platform. Then you replace the rule.
Clear written and spoken English, B2+.
Backgrounds that often do well:
Growth, retention or monetization data scientists from subscription, gaming, fintech or e-commerce — anywhere offers and churn are modelled with real money attached
CRM and lifecycle data scientists who owned campaign targetingandits incrementality
Causal inference and uplift specialists who ship policies rather than papers
Strong product analysts who moved into modelling and kept the decision instinct
This role is not a fit if you:
Want to train or fine-tune generative video models — that's what our R&D ML Engineer roles do, and they're open
Want to own a recommender or a serving stack — ranking is not in this role's scope today
Would target an offer at the users most likely to buy and call it personalization
Report a lift from a before-and-after comparison, or from comparing the users you targeted to everyone you didn't
Optimize AUC and hand the score to someone else to work out what to do with it
Need clean labelled data, a feature store and an experimentation platform to exist before you can start
Want predictable 9–5 workdays
We move fast:
Screening call (30 min)
Technical interview & case study (1 hour)
Practical home task (7 days)
Team interview (60 min)
Paid on-site trial (1 month)
Competitive base salary in USD, based on your experience, skills, and the scope of the role.
Equity participation through the company’s stock option program, giving you the opportunity to share in Higgsfield’s long-term growth.
Relocation support to Almaty for candidates moving from another city or country.
A highly collaborative, fast-paced environment where you can work directly with experienced leaders and have a meaningful impact on the product and company.
Opportunities for professional growth, ownership, and career development as the company scales.
Company-provided equipment, meals, transportation, or other office benefits.
New Data Scientist postings at the employers we follow are up 4% over the last 6 weeks compared with the 6 before.
Each bar is one week of new postings. The darkest bars are the last 6 weeks.
Counted from the hiring pages we’ve followed for at least five weeks, so employers we add don’t look like growth. Updated weekly.
We file this job under Data Scientist, by the work the posting describes.
Data scientists use statistics and models to answer harder questions, such as whether a change worked, what demand will look like next quarter, or which customers are likely to leave.
Postings usually ask for Python or R, SQL, statistics, experiment design and common machine learning methods.
Unlike an ML engineer, a data scientist's main output is often an analysis or a first model, not software running inside a product.
Source: US Bureau of Labor Statistics, Occupational Employment and Wage Statistics (May 2025) and Employment Projections (2025 to 2035). Occupation code 15-2051.
Selected passages from the saved posting. Check the full description for conditions and exceptions.
No pay amount identified in the saved description.
Almaty, Kazakhstan
- Team interview (60 min) - Paid on-site trial (1 month) What We Offer
No clear work-authorization passage found. Eligibility is unconfirmed.