Note: The job is a remote job and is open to candidates in USA. Dice is recruiting for a remote Senior Forecasting Lead Data Scientist role focused on applied forecasting and comparable-unit analysis. The role involves communicating forecasts to non-statisticians, teaching forecasting methods to client teams, collaborating with stakeholders, and delivering strategic recommendations alongside hands-on data science work.
Responsibilities
- Presents to non-statisticians: business outcome first, method second; confidence stated in plain language; explicitly states what the forecast cannot do; never opens with an undefined statistical term
- Can teach the method to a client team, not only execute it
- Participate actively in stand-ups and backlog refinement, engage business stakeholders directly, understand why the business is asking a question, and challenge or refine the request when it is wrong
- Strategic recommendations are expected alongside hands-on delivery
Skills
- The candidate must be able to name the industry and the outcome variable for each such engagement
- * Presents to non-statisticians: business outcome first; method second; confidence stated in plain language; explicitly states what the forecast cannot do; never opens with an undefined statistical term
- * Can teach the method to a client team, not only execute it
- * Participate actively in stand-ups and backlog refinement, engage business stakeholders directly, understand why the business is asking a question, and challenge or refine the request when it is wrong
- * Strategic recommendations are expected alongside hands-on delivery
- * Must be able to work EST hours
- * 8+ years of applied forecasting
- * Two or more comparable forecasting engagements led start to finish
- * Comparable-unit / same-store forecasting experience
- * Executive communication
- * Thought leadership
- * Multivariable regression, plus collinearity analysis and VIF interpretation
- * Forecast model development, tuning, selection and holdout validation
- * Metric fluency: R², WAPE, MAPE, p-values — and why WAPE is used at event grain (many events sell zero, which breaks MAPE)
- * Sparse and zero-inflated data. Many variables populate on under 25% of events, some as low as 10%. Nulls must never be silently treated as zeros
- * Data-leakage discipline and point-in-time correctness: every feature must exist before the event starts
- * Python and SQL; reproducible notebooks
- * Snowflake, including Snowflake ML Model Registry (model versions carry metrics and training-dataset references)
- * Git and pull-request workflow; all code merged to the client repository, no private forks
- * Architecture Decision Records (ADRs) and written process documentation
- * Categorical encoding at scale (~30–35 source variables expand to ~70 columns)
- * Sports, streaming, ticketing or subscription-business domain exposure
- * Hierarchical or mixed-effects models for low-volume segments
Benefits
- Remote work arrangement
- Contract employment
Company Overview