ECOM IGNITE is a growth platform for D2C brands (ecomignite.ai) for ad performance tracking, built on Supabase (Edge Functions + Postgres). Beyond basic trend forecasting, we're building a causal measurement layer — separating true incremental demand from captured demand, and identifying spend efficiency ceilings per channel/account. This is a deeper, methodology-driven build than a standard forecasting feature, closer to what tools like MMM vendors and analytics platforms offer their enterprise clients.

What You'll Build

- A causal / Marketing Mix Model (MMM) that separates demand generation, demand capture, and baseline (organic) revenue — controlling for seasonality and confounders
- A spend-efficiency-ceiling model: saturation/response curves per channel or account showing where marginal ROAS drops below a usable threshold
- A defensible accuracy framework — in-sample and out-of-sample error reporting with confidence intervals, not a single headline accuracy number
- Documentation explaining model assumptions and structure in plain language, so the methodology can be defended to clients

Our Data Reality (read before applying)
- Accounts vary wildly in volume — some run high monthly spend with rich daily data, others are much smaller and sparser. The model needs to handle both without breaking or producing unstable estimates on thin data.
- Data arrives via Meta/Google Ads APIs and can have gaps, expired references, and inconsistent formatting.
- We're not trying to build a research-paper-grade model — we need something rigorous enough to trust and defend, but pragmatic enough to ship and maintain.

Requirements
- Strong background in causal inference or econometrics (difference-in-differences,



instrumental variables, or Bayesian causal graphs) applied to marketing/business data
- Hands-on MMM experience — ideally with Robyn, Google Meridian, LightweightMMM, or a custom Bayesian implementation (PyMC / Stan)
- Robust Python (pandas, statsmodels, PyMC or equivalent probabilistic programming)
- Able to explain causal structure and modeling choices in plain language to a non-technical stakeholder
- Prior experience with marketing/ad performance data strongly preferred — understanding adstock, saturation curves, and seasonality confounders
- Comfortable working with Supabase (Postgres + Edge Functions) or willing to hand off a model that a data engineer can integrate

Engagement Structure
- Format: Fixed-price for defined scope (model design → validation → documentation → integration handoff), paid trial task first
- Trial task: Given a sample dataset, outline the causal structure you'd use (a simple DAG) and explain how you'd separate demand generation from demand capture, plus how you'd validate the model isn't overfitting on sparse accounts
- Estimated scope: 60–100 hours for v1, depending on model complexity
- Note: This is a separate, later-stage build from our existing forecasting module — no need to integrate directly with the trend-forecasting feature unless proposed

To Apply, Tell Us
1. A past project involving causal inference or MMM on real marketing/business data (link or case study if possible)
2. How you'd approach separating demand generation from demand capture, and how you'd validate that split
3. How you'd handle accounts with thin/sparse historical data without producing unreliable estimates
4. Your rate and availability

📌 Freelance Data Science (Causal Inference / Marketing Mix Modeling) (Tamil Nadu)
🏢 eCom Ignite
📍 Tamil Nadu