Coinflow is seeking a Fraud Data Scientist in Chicago, IL (onsite) to analyze and improve fraud models and rules. This role monitors model performance over time, detects shifting fraud trends, develops and validates new signals, and evaluates experiments and label quality.
What You Will Do
- Track fraud model and rule performance over time and by merchant segment, including approval rates, fraud loss, false positives, and drift, while documenting what changed and why.
- Identify emerging fraud patterns early, including fraud rings and account takeover, card testing, friendly fraud, and bonus abuse, assess the impact, and help prevent escalation into chargebacks.
- Analyze transaction, device, behavioral, identity, and network data to discover new signals, prototype and validate features, and prepare an evidence-backed recommendation for model updates.
- Own ground truth by interpreting delayed and noisy labels from chargebacks, disputes, and manual review, and flag when poor labels mislead model learning.
- Design and analyze backtests, shadow tests, and champion or challenger experiments, and evaluate third-party fraud vendors against Coinflow models using Coinflow data.
- Partner with risk operations to tune thresholds, rules, and step-up authentication, recommend decision strategies by merchant segment, and quantify tradeoffs between loss and approvals.
- Translate analysis into clear reporting for leadership, merchants, banks, and auditors, covering drivers of model decisions, performance trends, and ongoing actions.
Requirements
- 4+ years of experience as a data scientist or analyst working on fraud, risk, or abuse, ideally at a payments company, fintech, or marketplace.
- Proven experience using analysis to improve production fraud models or rules, including measurable outcomes such as new features, improved thresholds, reduced loss, or higher approvals.
- Strong SQL and Python skills.
- Familiarity with the statistics and ML concepts behind fraud models, including gradient-boosted trees, calibration, and imbalanced data (deployment experience not required).
- Solid understanding of the fraud landscape, including card-not-present fraud, account takeover, card testing, friendly fraud, chargeback reason codes, and how fraudsters adapt.
- Comfort working with messy, imbalanced, delayed-label data, with the judgment to evaluate model performance honestly.
- Experience designing and analyzing experiments where the costs of false positives and false negatives differ materially.
- Strong feature intuition to translate observed fraud patterns into signals that would detect them.
- Ability to explain model behavior and tradeoffs to merchants, risk operations, banks, and auditors in plain language.
- Graph or network analysis experience for detecting linked accounts and fraud rings is a plus.
- Comfort using AI-native tooling as a daily driver, or genuine excitement to use it.
- A bias toward action, favoring timely recommendations over waiting for perfect analysis.
Technologies
- SQL
- Python
- Gradient-boosted trees
- Calibration
Compensation
The base salary range for this role is $120,000 - $180,000 USD + equity.
About the Role
In this role, you will serve as the analytical core behind Coinflow’s fraud modeling. You will investigate how fraud models and rules are performing, identify where fraud is changing, and communicate which feature and strategy adjustments are most likely to move performance metrics.
The work spans card, bank, and alternative payment rails, supporting merchants in gaming, marketplaces, remittance, and ecommerce. You will collaborate with engineers who build and deploy models, along with risk operations and compliance, to convert data insights into decisions that can be executed.
You will also work hands-on with AI-native tooling to accelerate analysis, feature exploration, and experimentation.
