From Zero to £2M ARR: An E-Commerce AI Transformation Story

The Starting Point: £800K Revenue, Flat for Two Years

StyleFwd (name changed) is an independent UK women's fashion retailer with a strong wholesale presence and a growing but underperforming e-commerce channel. They had invested in a Shopify store and were running Meta and Google ads, but revenue had plateaued at approximately £800K for two years. Conversion rate was 1.2% — below industry average. Cart abandonment was 78%. Email revenue was inconsistent.

The founder reached out to TechGeneses with a specific brief: "Use AI to help us compete with brands 10× our size. We cannot outspend them — we need to outthink them."

Phase 1: Data Foundation (Months 1–2)

Before any AI work, we needed to understand the data. StyleFwd had 24 months of order history, 45,000 registered customer accounts, and fragmented analytics across Google Analytics, Klaviyo, and their Shopify backend. We built a unified customer data model that combined purchase history, browse behaviour, email engagement, returns data, and customer service interactions into a single view per customer.

The initial analysis revealed immediately actionable insights: 22% of customers who had purchased once never received a second-purchase email. 67% of high-value customers (£500+ lifetime value) had never been enrolled in a VIP programme. The top 15% of customers generated 68% of revenue — and were receiving the same communications as first-time buyers.

Phase 2: Personalisation Engine (Months 2–5)

Product Recommendations

We built a collaborative filtering recommendation model trained on purchase and browse history. The model powers three surfaces: homepage recommendations (personalised for each logged-in user), product page recommendations ("Customers who bought this also bought"), and email product blocks (personalised per recipient based on their style profile).

A/B test results after 60 days: personalised homepage drove 34% higher add-to-cart rate. Personalised email blocks drove 28% higher click-through rate and 19% higher revenue per email send.

Dynamic Pricing and Promotion

We implemented smart discount logic: instead of blanket promotional codes, the system offered personalised incentives based on customer behaviour. Price-sensitive customers (identified by browse-without-purchase patterns) received targeted discounts. High-LTV customers received early access to new collections. This reduced promotion cost while improving conversion for segments that respond to incentives.

Phase 3: Demand Forecasting and Inventory (Months 4–7)

StyleFwd's biggest operational problem: consistently stocking out of bestsellers while sitting on slow-moving inventory. We built a demand forecasting model using 24 months of sales history, seasonality factors, trend data from their Instagram engagement, and weather patterns (outdoor occasions drive demand for specific categories).

The model produces weekly forecasts by SKU with confidence intervals. The buying team now plans orders using model output rather than intuition. Stockout rate dropped 42% in the first quarter. Slow-moving inventory (defined as 90+ days without a sale) decreased 31%.

Phase 4: Automated Marketing (Months 5–10)

Behavioural Email Triggers

We built 14 automated email sequences triggered by customer behaviour: cart abandonment (personalised to the items abandoned, with related recommendations), browse abandonment for high-intent sessions, post-purchase sequences timed to review windows, win-back campaigns for customers inactive for 90 days, and birthday offers. These flows run autonomously and generate 38% of email revenue.

AI-Generated Email Content

GPT-4 generates first-draft subject lines and email body copy from product and campaign briefs. The marketing team reviews and refines. Email production time dropped 60%. Output increased from 4 to 10 campaigns per month. Revenue from email increased 3.2× year-on-year.

The Results: 14 Months Later

The investment in AI and our engagement paid back within 6 months. The systems we built are now assets that compound — they improve as more data flows through them. StyleFwd now competes effectively with brands 5× their size on customer experience, despite being a fraction of the budget.