описание
TyrAds is a leading tech-driven loyalty and rewards platform that partners with businesses to create meaningful and rewarding experiences for their customers. With millions of users across our platforms, we specialize in innovative AdTech solutions powered by big data, machine learning, and deep learning technologies. Our systems process over 10,000 events per second, delivering real-time insights at scale. We build seamless user experiences across web, Android, and iOS, while leveraging ML/AI to power personalization and advertising effectiveness.
задачи
- Model payout against funnel progression, event design against completion, and user quality against advertiser return
- Produce elasticity estimates that can be used to configure a campaign, not only to describe one
- Size the opportunity in revenue or margin before the work starts
- Design controlled experiments: randomisation unit, power analysis, primary and guardrail metrics
- Set run time from the funnel's own event timing, and agree stopping rules before launch
- Write a decision record for every experiment, including the ones that fail
- Run isolation reads before any causal claim, separating the effect of a funnel change from the effect of a change in publisher mix
- Difference in differences, matching, holdouts, incrementality testing
- Say clearly when the available data cannot answer the question
- Hand models that go live to the ML engineering team with an evaluation specification they can serve against
- Write findings that AdOps and Sales can act on the same day, without a translation layer
требования
- 5+ years as a data scientist or quantitative analyst in a product, marketplace or performance marketing business
- Applied statistics: experiment design, power analysis, confidence intervals, and the discipline to report a null result as clearly as a positive one
- Causal inference beyond A/B testing: difference in differences, matching, holdouts, incrementality
- Advanced SQL on large event tables (CTEs, window functions)
- Python for modelling and analysis (pandas, numpy, scikit-learn)
- Comfortable with the defects of real event data: duplicates, late arrivals, rejected installs, inconsistent partner data
- Can explain a model to a commercial stakeholder without simplifying it into something that is no longer true
- Works well alongside a Head of Product and a COO who are both hands-on in the same problems
- Будет плюсом: Mobile advertising, user acquisition, offerwall or rewarded advertising; MMP data (AppsFlyer, Adjust) and the limits of multi-touch attribution; Pricing, incentive or elasticity modelling in a commercial setting; Databricks or Spark; Fraud and anomaly detection
условия
- Условий в вакансии нет