119 EBITDA-tracked initiatives across 7 workstreams. A UK fashion e-commerce retailer with 16.4M customers and 72,000+ products — transformed by a full-stack Claude-powered programme that replaced external vendors with in-house AI and automated operations department by department.
A major UK fashion and lifestyle e-commerce retailer running flash sales and a curated marketplace. The business operates a Magento 2 platform on AWS, processing approximately 2 million requests per day with a customer database of 16.4 million profiles. The marketplace has 196 active sellers and over 72,000 products.
The technology stack includes three Aurora MySQL databases (465GB), a Snowflake data warehouse with CDC replication, dbt Cloud for transformations, Varnish caching, Cloudflare WAF/CDN, Algolia search, and 20+ integrated SaaS platforms.
The client's margin was under pressure from every direction. Marketing spend was growing faster than revenue. The technology stack was sprawling — 20+ SaaS vendors, many overlapping, several no longer fit for purpose. Cloud hosting was at premium AWS rates. Pricing was managed by an external vendor charging £75K/year with limited transparency into their algorithms.
There was no single view of where margin was being lost. The board needed visibility into every initiative, every department, every pound of improvement — tracked against EBITDA.
The business needed a partner who could build the AI systems, deploy them into production, and operate them — not deliver a slide deck.
Every AI system deployed at this client runs on Anthropic's Claude. We chose Claude for three reasons: the safety guarantees required for production business systems, the agentic capabilities needed for autonomous operations, and the technical depth of Claude Code, MCP, and the Agent SDK for real infrastructure work.
We didn't use Claude as a copilot. We used it as the engine — building, deploying, and running production systems that replaced external vendors, automated manual processes, and transformed how every department operates.
We replaced The client's external pricing vendor with a Claude-powered pricing engine built entirely in-house. The system analyses 3.6 million historical transactions across 37,800 products to build elasticity models and generate markdown recommendations.
The warehouse engine runs weekly across 15,000+ products using Claude's reasoning to generate pricing recommendations based on demand patterns, stock velocity, and margin targets. The consignment engine runs daily across ~1,600 products with two-way pricing logic — price cuts for zero-sellers, price increases for fast movers.
Real-time monitoring: A Claude-powered pricing monitor runs every 15 minutes with a 7-check system: demand shifts, stock velocity changes, discount code conflicts, zero-seller waves, margin alerts, and competitive positioning.
Beyond the engine: Personalised voucher codes tied to Claude-generated customer segments. Dynamic social proof badging on product pages. Bundle pricing that spreads discounts across items to protect per-unit margin. Discount code optimization that prevents stacking and leakage.
By week five: +80% units sold, +77% revenue, +72% gross profit versus the prior period.
We replaced a £45K/year third-party CDP with an in-house platform running on Snowflake, powered by Claude for all segmentation logic, scoring, and campaign intelligence.
Discount suppression: Claude analyses purchase behaviour to identify 90,000 customers who would buy at full price. Suppressing discount offers to these customers identified £8.3M in potential margin protection. Four discount tiers — suppress, light, standard, deep — are dynamically assigned based on Claude-generated customer scores.
CRM lifecycle overhaul: Claude redesigned the entire CRM programme — 190 automated marketing flows including welcome sequences, win-back campaigns, intelligent SMS timing, and personalised voucher distribution. Campaign logic, copy variants, and send optimization are all Claude-generated and Claude-maintained.
Acquisition: Claude-powered analysis identified high-ROI acquisition channels across Meta and TikTok. Renegotiated marketplace seller marketing contributions. Built a post-purchase advertising programme on order confirmation pages as a new revenue stream.
The client's entire infrastructure ran on AWS at £38K/month. Using Claude Code as the primary engineering tool, we migrated non-critical workloads to dedicated servers on Hetzner, keeping AWS only for services that genuinely benefit from the managed ecosystem.
Claude Code drove every step: Server provisioning, Nginx configuration, CI/CD pipeline setup, CDC pipeline management for data consistency, and zero-downtime deployment orchestration. Within three months, hosting dropped to £15K/month — a 60% reduction saving £276K/year.
Database optimization: Claude analysed 465GB across five production databases, identified 257GB of recoverable space. Six index fixes on the primary database — identified and implemented through Claude Code sessions — delivered query speedups ranging from 1,200x to 59,000x, saving 12.2 CPU days of processing.
Vendor replacement: Replaced the monitoring stack, cancelled underused SaaS tools, reduced data warehouse costs, swapped the PIM system, and built a Claude-powered visual search replacement for an external vendor. All built and maintained through Claude Code.
A Claude-powered CX agent handles first-line customer queries, marketplace seller support, and ticket triage. The agent continuously analyses support patterns, identifies friction points, and feeds recommendations into product and operations teams.
Full CX review: Claude conducted a comprehensive analysis of every customer touchpoint — from browse to purchase to post-delivery — identifying friction points, support bottlenecks, and NPS drivers. The insights fed directly into the headcount optimization and platform improvement workstreams.
Marketplace seller automation: The CX agent handles first-line marketplace seller queries across 196 active sellers. Previously manual — now automated with Claude, with human escalation only for complex issues.
The largest workstream by initiative count. The principle: automate the work with Claude before removing the role. No capability is lost.
Department-by-department: Customer service (P1 and P2), finance, merchandising, buying (three phases), data, technology (two phases), operations (two phases), studio, and samples. Each phase has a go-live month and a tracked EBITDA value. Monthly execution from March through October.
Claude-powered automation: AI-generated product descriptions replace manual copywriting. A Claude chatbot handles marketplace seller queries. Image automation replaces manual studio work. Claude agents absorb routine analyst, campaign manager, and operations tasks before each restructuring phase.
Structural changes: Closed the photography studio (moved to supplier imagery and Claude-generated content). Consolidated the samples team. Closed a margin-negative international unit.
A full personalisation overhaul across the customer experience, with Claude powering the intelligence layer behind every recommendation and ranking decision.
Search & discovery: Migrated to AI-powered search with neural capabilities. Claude generates product attributes for better filtering and implements advanced personalisation that adapts results based on browsing and purchase history. Custom ranking attributes surface high-margin products.
On-site experience: Claude-generated social proof badges (purchase counts, trending indicators) across PLPs, PDPs, and shopping bag. Redesigned homepage with dynamic, segment-driven content. Every personalisation change is A/B tested against revenue and margin impact before full rollout.
Warehouse optimisation: Shifted from 7-day to 6-day operations after Claude analysed order volume patterns. Renegotiated 3PL contracts. Automated book-in processes. Cross-docking for fast-moving lines. £1M+ in annual warehouse savings.
Three Claude agents run autonomously across The client's marketplace, SEO, and analytics functions.
Marketplace Agent: Zero-seller identification and automated offboarding across 36,000+ products. Seller performance monitoring, product quality checks, and automated communications. Previously required a full-time analyst — now runs on Claude 24/7.
SEO Agent: Technical SEO auditing, content analysis, meta optimization, and LLM ranking visibility monitoring. The agent runs continuous audits and implements fixes autonomously through Claude Code.
Data Analysis Agent: Cross-platform analytics, anomaly detection, automated reporting, and board-ready dashboards. The agent identifies trends, flags issues, and generates insights that previously required a data team to produce manually.
DevOps Agent: Site performance monitoring, database health checks, error tracking, and deployment validation. Runs 24/7, escalating only when human intervention is required.
Every agent runs on Anthropic's Claude. Built with the Claude API, orchestrated via the Agent SDK, connected to live systems through MCP, and maintained through Claude Code.
Warehouse and consignment pricing with real-time 15-minute monitoring across 15,000+ products. Replaced £75K/year vendor.
CDP management, 190 automated marketing flows, discount suppression, campaign analysis, and deep customer segmentation across 16.4M profiles.
Customer experience analysis, CSAT tracking, chatbot optimization, support ticket triage, and marketplace seller query handling.
Zero-seller identification, automated offboarding, seller performance monitoring across 196 sellers and 36K+ products.
Site performance, database health, error tracking, deployment validation, and infrastructure monitoring. Built and maintained with Claude Code.
Cross-platform analytics, anomaly detection, automated reporting, board-ready dashboards, and trend identification.
Technical SEO auditing, content optimization, meta analysis, and LLM ranking visibility monitoring — running autonomously.
This isn't a surface-level AI integration. Claude is embedded at every layer of the technology stack we built and operate at this client.
Every agent runs on the Claude API. Pricing decisions, customer segmentation, content generation, anomaly detection, and operational reasoning — all powered by Claude's production API with enterprise-grade reliability.
Our primary engineering tool. The entire cloud migration (AWS to hybrid), database optimisations, CI/CD pipelines, server provisioning, and ongoing infrastructure maintenance are built and managed through Claude Code sessions.
Multi-agent orchestration across all seven production agents. The pricing agent feeds the CRM agent. The CX agent informs the marketplace agent. The DevOps agent supports them all. Coordinated agent teams sharing context across workstreams.
Model Context Protocol connects agents to live databases, APIs, and business systems. The pricing agent reads from Snowflake. The marketplace agent writes to Magento. The DevOps agent monitors server metrics. All through MCP.
The 1M token context window and native agent team orchestration enabled the coordinated multi-agent architecture. Previous models couldn't hold the full context of The client's operations. Opus 4.6 changed that.
Human-in-the-loop oversight for pricing decisions above threshold. Full audit trails on all agent actions. Guardrails on customer-facing communications. Anthropic's safety standards applied throughout.
Every initiative is tracked, EBITDA-tagged, and board-reported. The transformation programme spans 7 workstreams with Claude at the centre of each one. Headcount optimization runs monthly, enabled by Claude agents that absorb work before roles are removed. The programme continues to grow — new initiatives are added as opportunities surface, and the tracker is the single source of truth for the board.
“The transformation programme gave us board-level visibility into every initiative. We went from gut-feel decisions to EBITDA-tracked execution across every department.”
— VP of Operations, UK Fashion Retailer
Claude-powered AI agents deployed across pricing, marketing, CX, and operations. Every initiative EBITDA-tracked. Every system built on Anthropic's Claude.