Customer Growth & Pricing Intelligence Platform

Live Streamlit Dashboard

Executive Summary

Customer Growth & Pricing Intelligence Platform is a synthetic fintech data product that connects customer analytics, experimentation, model decisioning, pricing intelligence, and cloud deployment into one end-to-end workflow.

The project answers the kind of questions a product, growth, or customer strategy team would ask in a neobank:

• Which customers activate, retain, and become higher-value users?
• Which onboarding treatment should be shipped, monitored, or iterated?
• Are referral incentives incrementally acquiring customers, or just subsidising signups that would have happened anyway?
• Which pricing or incentive offers are commercially attractive after guardrails?
• Can the scoring, monitoring, and API surfaces be operated as a small data product rather than a one-off notebook?

All data is synthetic. No customer data, internal bank data, or proprietary business metrics are used.

Delivered Product

• Public product dashboard for activation, engagement, retention, feature adoption, pricing, experiments, and monitoring.
• dbt metrics layer for product health, experiment readouts, pricing outcomes, and regional referral incrementality.
• Activation decisioning model with calibration, explainability, segment checks, and model card.
• Pricing intelligence workflow with offer economics, recommendation guardrails, scenario runs, and sensitivity analysis.
• Experimentation workflow covering CUPED A/B testing, SRM checks, guardrails, heterogeneous effects, DiD, synthetic control, and placebo checks.
• FastAPI service boundary for activation, churn, upsell, offer recommendation, and pricing scenario contracts.
• GCP path using Cloud Storage, BigQuery, Cloud Run, Cloud Scheduler, and Cloud Monitoring.
• Operations documentation for release checks, monitoring, score drift, realised-label calibration, rollback, and cost control.

Business Outcomes

Area	Output	Decision Value
Activation	D7 activation model and onboarding A/B readout	Identify users needing onboarding help and decide whether to ship the treatment.
Retention	Post-activation retention and engagement views	Monitor whether activation quality translates into continued usage.
Pricing	Offer economics and scenario analysis	Compare margin, expected conversion, support risk, and fairness guardrails before rollout.
Referrals	Geo incrementality analysis	Estimate true lift and cost per incremental referred customer.
Monitoring	Release gates, drift checks, calibration, and alerts	Keep model and data outputs reviewable before operational use.

Production-Style Evidence

The portfolio version has been exercised beyond local development:

• Public Streamlit dashboard deployed on Streamlit Community Cloud.
• Cloud Storage raw landing path loaded into BigQuery.
• BigQuery raw tables verified against the export manifest.
• dbt graph run against BigQuery with 107 passing checks.
• Activation score extract loaded to neobank_ml.customer_scores_daily.
• BigQuery monitoring result written to neobank_monitoring.score_monitoring_daily.
• Private Cloud Run API deployed and smoke-tested via authenticated /health.
• Cloud Run Jobs deployed for activation scoring and score monitoring.
• Cloud Scheduler configured for daily scoring and monitoring runs.
• Cloud Monitoring alert policies created for Cloud Run job failures and API service errors.
• Budget alert and Cloud Storage lifecycle policy configured for cost control.

Architecture

Synthetic event generator
        |
        v
DuckDB / Parquet local development
        |
        +--> dbt marts --> Streamlit dashboard
        |
        +--> modelling and experiments --> model card, memos, monitoring
        |
        v
Cloud Storage raw landing
        |
        v
BigQuery raw tables and marts
        |
        +--> Cloud Run batch jobs --> scored users and monitoring table
        |
        +--> private Cloud Run API --> prediction and pricing contracts
        |
        v
Cloud Scheduler + Cloud Monitoring alerts

Technology Stack

Layer	Tools
Data generation	Python, Faker-style synthetic event generation, Parquet
Analytics engineering	dbt, DuckDB, BigQuery
Experimentation	CUPED, SRM checks, guardrails, DiD, synthetic control
Modelling	scikit-learn, calibration, explainability, model card
Application	Streamlit dashboard, FastAPI prediction service
Cloud	Google Cloud Storage, BigQuery, Cloud Run, Cloud Scheduler, Cloud Monitoring
Quality	pytest, ruff, GitHub Actions, monitoring snapshot workflow

Review Guide

For a five-minute review:

1. Open the live dashboard.
2. Read the case study in docs/CASE_STUDY.md.
3. Review product health, pricing intelligence, experiments, and monitoring.
4. Skim the architecture in docs/ARCHITECTURE.md.
5. Read the API contract in docs/API.md.
6. Review the model card in docs/model_cards/MODEL_ACTIVATION_DECISIONING.md.
7. Review the runbook in docs/OPERATIONS_RUNBOOK.md.

Key Documentation

Document	Purpose
docs/CASE_STUDY.md	Business-facing case study, decisions supported, evidence, and limitations.
docs/ARCHITECTURE.md	Local and cloud architecture overview.
docs/API.md	Prediction and pricing scenario API contract.
docs/CLOUD_RUN_DEPLOYMENT.md	Private Cloud Run API and job deployment evidence.
docs/GCP_WAREHOUSE.md	Cloud Storage and BigQuery warehouse path.
docs/MONITORING.md	Data, model, score, and GCP monitoring checks.
docs/OPERATIONS_RUNBOOK.md	Rollback triggers, triage, and GCP operations.
docs/PRICING_SCENARIOS.md	Pricing recommendation and scenario framework.
docs/PUBLIC_LAUNCH.md	Portfolio, LinkedIn, and CV wording.

Local Reproduction

The project is reproducible without GCP. The local path builds a synthetic dataset, dbt marts, tests, and the dashboard.

uv sync --group dev
uv run python -m data_generator.generate --users 5000 --months 6 --output-dir raw/ci
uv run dbt build --project-dir dbt_neobank --profiles-dir dbt_neobank
uv run pytest
uv run streamlit run app/streamlit_app.py

For the full portfolio-size data run, use:

uv run python -m data_generator.generate --users 50000 --months 12 --output-dir raw/portfolio_full
uv run dbt build --project-dir dbt_neobank --profiles-dir dbt_neobank --vars "{raw_path: raw/portfolio_full}"

Public Release Notes

This is a synthetic portfolio case study inspired by public fintech product analytics themes. It is designed to show how product analysis, causal inference, pricing analytics, model decisioning, and cloud operations can be combined into a small but credible data product.

The project should not be interpreted as representing any real financial institution. It uses no real customer data.

What Would Be Hardened For Production

A real regulated-financial-services deployment would add:

• Stronger API protection through API Gateway, IAP, JWT, or service-to-service IAM, plus rate limits and structured request logging.
• Formal data governance with row-level and column-level controls, retention policy, ownership, lineage, query labels, and cost controls.
• Keyless CI/CD deployment through GitHub OIDC, container vulnerability scanning, SBOMs, and infrastructure as code.
• A model registry or feature store for reproducible training, shadow deployments, and online/offline feature parity.
• Formal privacy, Consumer Duty, model-risk, audit, and approval controls before any live customer decisioning.