Skip to content
← cd ~/projects

Matched groups, repeatable runs

Experiment pipeline & outlet matching

Automated outlet-matching pipeline for smooth experiment runs; these experiments contributed to modest business growth (details on request, DM me).

1-2%
intra-cluster variance
6 mo
sales history

(Python · SQL · Amazon Redshift · scikit-learn · psycopg2 · pandas · Linux cron)

  1. # Problem

    Offers were tested without a repeatable way to build balanced test and control groups, or to pair lookalike outlets before reading results.

  2. # Approach

    Clustered outlets with K-Means on six months of sales and footfall, split by price tier, landing at 1-2% variance inside each cluster. On Amazon Redshift, a configurable pipeline turns transactions into customer features and experiment-ready groups, with stratified sampling, data-integrity checks, and balance validation.

  3. # Result

    Experiments now start from the same matching and cohort pipeline instead of one-off notebooks. They contributed to modest business growth; details on request, DM me on X.