Sota ForgeSota Forge
Back to cases
E-commerce · Beauty

From gut-feel CRM to a system that validates every campaign with statistics — in 7 weeks

We unified scattered customer, order, and messaging data onto a single CRM rail and shipped a production system that runs A/B validation on a statistical basis — all in 7 weeks.

Client
Company E (D2C e-commerce brand)
Period
Apr–Jun 2026 (7 weeks)
Role
Sole design, build, and operational handover of the CRM system
Completed
7 weeks
Time to production
19
Real-time event types
4 channels
Sending channels
17
Cohort condition fields
67
Zero-downtime schema migrations

The Problem

  • No framework existed to diagnose the entire customer base quantitatively before running CRM
  • Customer, order, and messaging data were fragmented, making it impossible to compare performance between organically acquired and influencer-driven customers
  • Campaign decisions relied on the operator's intuition

Approach

  • Redesigned the workflow itself into a four-stage "CRM rail": diagnose → target → execute → validate
  • Built the data foundation on 19 real-time commerce-platform webhooks plus 9 overnight aggregation batches
  • A/B sends use deterministic sha256 assignment and a two-proportion z-test to declare the winner
  • Beyond meeting every contractual success criterion, delivered four out-of-scope areas: the diagnostic rail, customer scatter plots, webhook sync, and a GA4 + first-party tracker

Systems Built

ModuleDescription
Data integration & auto-aggregation9 overnight batches, idempotent processing
Cohort management17 AND/OR fields, snapshots
Acquisition-source engineAuto-classifies organic vs. influencer-driven customers
Experiment design & A/B sendingBuilt-in statistical testing
Multi-channel sendingSMS, LMS, AlimTalk, brand message — 4 channels, compliance built in
Tracking & evaluation5-tab reporting, 8-stage message-response funnel, RFM 5×5 heatmap (build a cohort straight from any cell)
Handover package30 API docs, 4 operations runbooks

Screens

Company E CRM project wrap-up stat cards — anonymized, aggregated data
AlimTalk / brand-message composer improvement wireframe

Tech Stack

Next.js 14Node.jsPostgreSQL (RDS)AWS EC2pg-bossSentryCafe24 API/WebhooksGA4

Metrics reflect system scale and delivery scope. Revenue-performance metrics will be compiled once the measurement infrastructure is live.

Have a similar challenge?

Let's shape your system design in a free 30-minute consultation.

Other cases