Role: Senior Forecasting Lead Data Scientist
Location: Remote
Visa: NO GC
Experience: 10+ years
The candidate must be able to name the industry and the outcome variable for each such engagement. Retail same-store analysis is the classic form; the analogue here is comparing similar schools and events rather than following one trend line.
* 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.
Qualifications
Required:
* 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.
Preferred:
* 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.
