About the Role
GrowTal builds artificial intelligence and machine learning products for marketing measurement. Our platform connects a client’s marketing and revenue data, measures performance, models what drives outcomes, and reports the results.
We run per-customer marketing mix models in Google Meridian today, hand-specified per client. This role converts that into a self-serve marketing mix modeling feature in VibeMA: a system that specifies, fits, validates, and grades a model for any qualifying account without an analyst in the loop. This role reports to the Chief Technology Officer.
Responsibilities
- Generalize the existing per-customer Meridian implementation into a multi-tenant system that runs unattended.
- Automate control selection: candidate screening, ranking on detrended correlation, and a full-sampling convergence check before any control is added or dropped.
- Automate prior setting, including the media and baseline split, which currently requires a per-client judgment call that out-of-sample metrics cannot adjudicate.
- Automate modeling window selection, including detection of structural breaks and of the point where an upstream metric becomes available.
- Automate per-channel identifiability diagnostics: detect channels whose contribution is prior-driven or unstable across holdout seeds, and surface that state in the product rather than reporting a point estimate.
- Own holdout design, including leakage from carry-over and from a baseline fit on both sides of a held-out point.
- Own the model acceptance gate: out-of-sample error, convergence and divergence thresholds, degenerate baseline detection, credible interval width, and tier assignment.
- Automate data sufficiency gating that determines whether an account can be modeled at all.
- Own data quality assertions on curated inputs and the tests that prove each assertion fires.
- Define the normalized output schema, and how contributions, uncertainty, and assumptions are presented to end users.
- Establish quality monitoring and regression detection across accounts and successive refits.
- Work with engineering on runtime, cost, and refit cadence at volume.
Requirements
- Experience building and shipping marketing mix models against real marketing spend, where the output informed budget decisions.
- Experience systematizing modeling work into a repeatable automated pipeline rather than analyst-run one-off engagements.
- Fluency with Bayesian sampler diagnostics: R-hat, divergences, posterior geometry, and why convergence pathologies do not reliably reproduce at reduced sampling.
- Ability to reason about identification, including collinearity, low-variance regressors, and telling a data-driven contribution from a prior-driven one.
- Holdout and validation design for time series with carry-over effects.
- Ability to translate analytical judgment into automated diagnostics and acceptance criteria that hold without human review.
- Calibration against incrementality or geographic lift experiments where available.
- Understanding of where platform-reported and last-click attribution mislead, and how modeling and experimentation address that gap.
- Strong Python and SQL, and production-quality code that runs unattended on a schedule.
- Ability to present modeled results and uncertainty to non-technical end users without overstating confidence.
- Fluency with artificial intelligence assisted tooling, and full accountability for your methodology.
Preferred
- Direct experience with Google Meridian.
- Hierarchical or pooled modeling across many accounts.
- Incrementality or geographic lift experiment design.
- Bayesian workflow at scale, or probabilistic programming beyond a single modeling framework.
- Familiarity with major advertising and analytics platform data models.
- Experience productizing analytics for non-analyst end users.
Stack
- Python, SQL
- Google Meridian
- BigQuery, Funnel.io, Looker
- Postgres on Cloud SQL
- Google Cloud Platform: Cloud Run
- Anthropic software development kit
- Linear, Notion
Ready to Apply?
Fill out the form below to apply for the Data Scientist