BackCase Study

Club21 — AI-Powered E-Commerce Analytics

Client: Club21 (COMO Group) | Industry: Luxury Fashion E-Commerce | Platform: Shopify Plus | Region: Singapore, Malaysia, and 20+ shipping destinations

The Scale

A complex, multi-channel operation

Club21's e-commerce operation is not small. The online store carries 40,000-50,000 active SKUs across 250+ luxury brands, ships to 20+ countries, and runs paid advertising through 400+ campaigns across Google Ads and Meta — managed through five regional ad accounts spanning Singapore, Malaysia, and several other markets. Klaviyo sends 100,000-150,000 emails per month through 40+ campaigns and 70+ automated flows. Combined paid media generates 15-20 million ad impressions monthly.

For a lean e-commerce team, keeping analytical eyes on all of this — across every channel, every market, every product category — is the core challenge.

The Challenge

Data in silos, insights out of reach

Like many mid-size e-commerce teams, Club21's analysts faced a recurring problem: data lived in silos. Orders data sat in Shopify. Email and SMS performance lived in Klaviyo. Paid media metrics were split across Google Ads and Meta. Website behaviour was in its own analytics stack. Market-level trends required manual cross-referencing.

Every reporting cycle meant hours of pulling exports, switching between dashboards, and manually connecting the dots — often under time pressure, and often missing the cross-channel patterns that matter most.

The Approach

An 11-agent AI system across 7 channels

We designed and built a multi-agent AI system powered by Anthropic's Claude that analyzes e-commerce data across seven channels — orders, Klaviyo, SKU performance, website analytics, market intelligence, Google Ads, and Meta — then synthesizes cross-channel insights automatically.

How it works — two phases

Phase 1: Channel Analysis

Seven independent channel agents each analyze a single data domain. Each agent ingests standardized data exports, applies its domain-specific analytical framework, and writes 20 structured insights. Individual agents complete in 25-45 minutes each.

Phase 2: Synthesis

A synthesis agent reads all 140+ channel-level insights and identifies cross-channel patterns. A full cycle completes in under 3 hours.

Three support agents complete the system: data standardization, feedback processing, and instruction improvement. Eleven agents total.

Key design decisions

No custom infrastructure

Runs on Claude and Notion. No servers, no databases.

Simple data pipeline

Standardized CSV exports processed locally with Python.

Self-improving

Analyst feedback flows into a persistent knowledge base that sharpens each agent.

Built-in validation

Automated audits scan all 13 instruction pages after every change, catching 5-10 issues per scan.

By the Numbers

The system at a glance

40-50K
Active SKUs across 250+ brands
100-150K
Emails/month (40+ campaigns, 70+ flows)
400+
Campaigns, 15-20M impressions/month
5
Regional ad accounts
11
Agents (7 channel + 1 synthesis + 3 support)
140+
Insights per cycle
<3 hrs
Per full analysis cycle
12+
Instruction iterations in first 3 months
$0
Infrastructure cost
Impact

What the system delivered

Unlocking Hidden Revenue in Paid Media

The system’s first analysis surfaced that approximately 10-12% of combined paid media budget was flowing to campaigns that had not yet converted. At the same time, nine highest-performing Google Ads campaigns were running at 12x+ ROAS but hitting budget caps. By reallocating, the system identified SGD 1-2 million per year in additional attributed revenue potential — with no increase in spend.

Operational Efficiency

Full cross-channel analysis in under 3 hours. More than 2x increase in analytical depth — from 60 insights per run to 140+.

Data Quality Safeguards

The system caught a data processing issue affecting approximately 50% of transaction line items — a Shopify multi-item order export nuance. It identified the root cause, corrected its processing logic, and updated instructions to prevent recurrence — within the same cycle.

Cross-Channel Intelligence

In its first full-cycle run, the synthesis agent identified seven cross-channel connections spanning ad spend, SKU sell-through, email engagement, and operational disruptions.

Compounding Value

The system’s instruction set has been refined over 12 iterations in its first three months, with automated audits scanning all 13 instruction pages after every change.

Technical Summary

Stack overview

AI PlatformAnthropic Claude
Knowledge LayerNotion
Data ProcessingPython (local CSV)
E-CommerceShopify Plus
Email / SMSKlaviyo
Paid MediaGoogle Ads, Meta Ads

Have a similar operation?

If your e-commerce team is managing multiple channels, markets, or brands — we should talk.