Demand ForecastingClient DeliveryConsultingFinancial ServicesMedia

Demand and audience forecasting for financial and media clients (Artefact)

Artefact, Senior Data Scientist (2025) · Lead Data Scientist

End-to-end pipeline diagram: Data triage (quality audit · BigQuery) → XGBoost · Prophet (fit per problem) → Client validation (sign-off vs intuition) → Documented handoff (model card · retraining)
Pipeline overview

Two consulting projects ran in parallel. A media group needed reach and CPM forecasts to plan advertising inventory. A financial services client needed transaction forecasts for liquidity and capital planning.

Each project had two weeks from kickoff to an executive demo, leaving little time for the data investigation that forecasting needs. Both datasets had gaps: sparse histories for some ad placements and thin records for niche financial products. When forecasts challenged client expectations, I had to explain the result immediately while leading the technical work and the stakeholder discussion.

In each two-week sprint, I handled data gaps, selected models, added revenue and reporting constraints, and wrote handoff documents in parallel. I translated forecasts into planning terms that senior leaders could use.

I delivered a model card, feature dictionary, retraining protocol, and dashboard guide. Before sign-off, client teams compared the outputs with their domain knowledge and resolved any differences with me.

The media client could commit inventory using forward reach estimates. The finance client could set liquidity buffers from probabilistic transaction ranges. Both systems were delivered on schedule with full handoff to client teams.

GCP (Vertex AIBigQuery)PythonXGBoostProphetdbtSQL