Python package and Streamlit dashboard, deterministic demo data, real public data, documented contracts, and a tested modelling pipeline.
Historical roadmap
Directions recorded when this repository was archived.
This roadmap is closed and retained as historical context. No further development is planned in this repository. Active analytics and measurement work continues in Responsible Neobank Growth.
No pricing, packaging, or go-to-market, and no authentication or governed storage for confidential company data.
New work belongs in the successor repository; the directions below are historical.
Methodology directions recorded at archive
Deeper Bayesian MMM
Move from the conjugate posterior over the fixed design matrix toward a sampler that also treats adstock and saturation parameters as random, with posterior predictive checks.
Stronger Calibration
Richer geo-lift and incrementality designs, and a clearer reconciliation between experiment evidence and modeled contribution.
Validation Depth
Time-series cross-validation, backtesting across multiple holdout windows, and sensitivity analysis on priors and transformation parameters.
Identifiability
Diagnostics for collinearity between channels and controls, and for how well spend variation identifies each response curve.
Engineering Quality
Type checking, expanded test coverage on modelling paths, and reproducibility via pinned environments, run manifests, and deterministic seeds.
Former contribution lanes
Measurement Methods
MMM, uncertainty, calibration, diagnostics, causal evidence, and validation.
Data Platform
Connectors, schemas, storage, data quality, and reproducible model inputs.
CRM Experimentation
Audience logic, contact policy, readouts, experiment registry, and learning library.
Engineering Quality
Type checking, test coverage on modelling paths, reproducibility, and run manifests.
Documentation
Examples, governance notes, product boundaries, deployment notes, and contribution guides.
Non-goals
A proprietary app, pricing, or go-to-market
Raw PII in the public demo
Production approval claims
Vendor replacement claims
Pretending demo data is real brand performance
Principle
Useful measurement tools should be transparent about assumptions, data limits, security boundaries, and where human judgment remains necessary.