AI for e-commerce companies of 20–200 people — from customer service to demand forecasting

A 20–200 person e-commerce business usually runs multi-channel: own store (most often Shopify, Shoper, PrestaShop or BigCommerce), Amazon, eBay, local marketplaces, plus brick-and-mortar showrooms. The central system is often an order management platform or a custom integration with the ERP (Comarch, SAP, Microsoft Dynamics). The warehouse is typically 5–50k SKUs, with order volume from dozens to a few thousand a day.

In this setup the most overloaded part of the organisation is customer service / returns and logistics. AI in e-commerce here isn't about buzzwords — it solves three concrete problems: growing query volume, pressure on reaction time, rising cost of logistics mistakes.

The first AI area in e-commerce that pays back fastest is customer service. A chatbot on site and a voicebot on the helpline handle 50–70% of repeat queries: order status, product availability, return policy, delivery questions, basic product advice. For a store handling 500–5000 queries per week that means recovering several service FTEs.

What matters is that the chatbot has access to real data — order status in the OMS, availability in the ERP, customer history. Without that it stays at FAQ level. A well-built chatbot today is a 6–12 week project, EUR 7–18k, with ROI in the first 6 months.

Returns and complaints are the second big time-sink in e-commerce. A typical store has 2–8% of orders as returns or complaints, each handled manually taking several minutes. An AI agent classifies tickets, asks customers for missing photos, extracts data from receipts, generates protocols and opens procedures in the OMS.

The best projects combine ticket handling (customer conversation through a chatbot) with back-office workflow (classification, decision, ERP correction, refund). The effect: handling time shrinks from days to hours and post-sale NPS rises visibly.

Holding excess stock is a classic mid-sized e-commerce problem — end-of-season discounts, markdowns, disposal cost. AI sales forecasting analyses historical data, seasonality, marketplace trends, promotions and ad campaigns to indicate what to buy, when and how to balance availability across channels.

For 5–50k SKU stores a realistic forecasting project is 3–6 months focused on top categories (Pareto). Typical effects: 10–25% inventory reduction without stock-out growth, 3–8% margin lift from lower markdowns and better best-seller availability.

Dynamic pricing in e-commerce is where AI is still picking up speed — mainly due to marketplace box pressure on price. Even so, more stores introduce dynamic models on their own store and marketplaces, adjusting prices to demand, availability, competition and category margin.

Personalisation (product recommendations, dynamic content, campaign segmentation) only makes sense once the store has a clean customer database (buyers and prospects) and integration of marketing (Klaviyo, Mailchimp, Edrone) with the ERP. Without that, personalisation wastes time and money.

Marketplaces generate 30–70% of revenue in many stores and require huge volumes of small decisions: listing, position monitoring, answering buyer questions, complaint handling, competitor monitoring. AI combined with an OMS and Power Automate automates most of this — from listing descriptions to buyer Q&A.

The second area is scaling to more marketplaces (Amazon, eBay, Allegro, local platforms). Each has its own technical and operational requirements. An AI agent can prepare description variants, map categories, manage campaigns and check regulatory compliance.

Most mid-sized e-commerce boards see reports with a 1–3 day delay, in a form that doesn't allow rapid reaction. Power BI combined with AI gives near real-time views: sales per channel, margin per category, cost of service, returns per category, top SKU availability.

The most valuable add-on is an AI assistant in Power BI or Microsoft 365 Copilot that reads the dashboard, drafts management commentary and answers natural-language questions ("why did margin drop in March?", "which categories sold best on the weekend?"). It shifts the board's culture from reactive to proactive.

A practical 6–9 month path: months 1–2 — chatbot in customer service (FAQ + order status + availability). Months 3–4 — returns and complaints automation. Months 5–6 — sales forecasting on top categories. Months 7–9 — dynamic pricing, personalisation, marketplace automation and management reporting.

This model delivers visible results each quarter and avoids overloading the organisation. Total programme cost for mid-sized e-commerce is typically EUR 45–95k spread over a year, with ROI in the first 12–18 months.