Case Study
A UK online fashion retailer with 85,000 customers automated 61% of its support volume with an AI chatbot, cutting first-response time from four hours to 28 seconds and recovering the project cost within five months.
UK online fashion retailer
A UK online fashion retailer with 85,000 customers automated 61% of its support volume with an AI chatbot, cutting first-response time from four hours to 28 seconds and recovering the project cost within five months.
A UK online fashion retailer with around 85,000 active customers was spending a significant amount each month running a nine-person support team that handled the same questions day after day. Order tracking queries, return requests, size guides, delivery timescales, and stock availability made up roughly 78% of all inbound contacts. The team fielded an average of 1,400 tickets per week, with peak periods during Black Friday and January sales pushing volumes beyond 2,800 in a single week.
The problem was not that the team was slow or unmotivated. The problem was structural. Each query required an agent to log in to the order management system, check the order status, copy a tracking reference, and paste it into a reply. For stock queries, agents checked the product database manually. These tasks could in principle be answered instantly by any system with access to the right data, yet they consumed the majority of the team's hours every day.
Abandonment rates on live chat had reached 34%, largely because wait times during peak hours stretched beyond four minutes. Customer satisfaction scores were declining, and a growing number of Trustpilot reviews specifically mentioned slow response times as a source of frustration. The business recognised it needed to either significantly expand its support team or find a smarter way to handle the volume.
There was a second layer to the problem. The team's knowledge was held informally, spread across individual agents, shared documents, and an outdated FAQ page that had not been updated in over two years. New agents took three to four weeks to reach a standard of competency that kept satisfaction scores stable. The business could not rely on hiring alone to solve a capacity problem that was growing season by season.
The managing director approached Softomate with a clear brief: reduce ticket volume, speed up response times, and protect the quality of the customer experience without adding headcount. The business had previously trialled two off-the-shelf chatbot products, both of which were abandoned within six months. The first could only respond to exact keyword matches. The second offered some natural language capability but could not connect to the order management system, meaning it could not give personalised responses and customers found it more frustrating than helpful.
The engagement began with a three-week discovery phase. Softomate's team analysed six months of support ticket data, categorising every query by intent, frequency, and the level of human judgment required to resolve it. This produced a clear priority matrix: the top 12 query types, which accounted for 81% of total volume, were entirely answerable by a system with real-time access to order data, product information, and returns policy documentation.
The solution was a retrieval-augmented AI chatbot built on GPT-4o and integrated directly with the retailer's Shopify back end and warehouse management system via REST APIs. The chatbot could pull live order status, tracking references, estimated delivery windows, and current stock levels in real time, returning accurate, personalised responses without any agent involvement at any stage of the interaction.
Softomate built a custom knowledge base from the retailer's returns policy, shipping terms, size guide data, and promotional FAQs. The knowledge base was structured as a vector store and queried using semantic search, so the chatbot retrieved relevant information even when customers phrased questions in unusual or unexpected ways. A customer asking where their parcel was and a customer asking whether their order had shipped both reached the correct answer without needing identical phrasing.
The chatbot was embedded as a floating widget on the retailer's website and integrated with their existing Zendesk instance, so complex or escalated queries were handed to a human agent with full conversation context already visible in the ticket. Agents no longer had to re-read a long message history; a structured handover summary was generated automatically from the conversation.
A critical part of the build was tone calibration. Softomate trained the chatbot using a corpus of the retailer's own highest-rated support conversations so that responses matched the brand's warm, informal style rather than sounding like a generic automated system. The retailer reviewed 200 sample outputs before launch and confirmed the tone was consistent with their voice.
The deployment was phased. The chatbot went live for order tracking and FAQ queries first, with a two-week monitoring period during which Softomate reviewed every escalation to identify gaps in the knowledge base. Product stock queries and returns initiation were enabled in week three once accuracy was confirmed. Full autonomous coverage of all 12 query types was live within six weeks of integration starting. Throughout, weekly review sessions with the retailer's operations manager ensured the knowledge base was refined iteratively, with 11 targeted updates made before it reached a stable, production-ready state.
Within 90 days of full deployment, the results were clear and measurable. Ticket volume handled by human agents fell by 61%. The chatbot resolved 1,105 of the average weekly 1,400 contacts without escalation. Average first-response time dropped from 4.1 hours to 28 seconds. The abandonment rate on live chat fell from 34% to 7%.
Customer satisfaction scores measured via post-interaction surveys improved from 3.6 to 4.4 out of 5 for chatbot-handled interactions. The human agents who remained reported that the quality of their work had improved significantly. They spent their time on returns disputes, delivery exceptions, and complex sizing queries, the kind of nuanced work they were actually suited to, rather than pasting tracking numbers into reply templates.
The retailer's Trustpilot rating improved from 3.8 to 4.3 stars over the following six months. Several reviews specifically mentioned the speed of the chat support as a direct reason for the higher rating.
The managing director confirmed that the business had redirected the equivalent of four full-time support positions into growth activities, including a proactive customer outreach team and an expanded social media presence. The cost of the Softomate engagement was recovered within five months of go-live.
One outcome the retailer had not anticipated was the operational intelligence the chatbot surfaced. Because every query was logged and categorised automatically, the business could see clearly which product lines generated the most support contacts. A consistent pattern of sizing queries around one category led to a decision to add a measuring guide video directly to those product listings, which reduced that specific query type by a further 40% over the following quarter.
The chatbot has since been extended to handle post-purchase review prompts and loyalty scheme FAQs, adding further automation coverage without material additional development cost beyond a two-day configuration engagement.
Names withheld to preserve confidentiality.
Work with us
Every project we take on has a measurable outcome. Talk to our London team and we will show you exactly how we would approach your challenge.
Softomate Solutions Limited
Deen Dayal Yadav
Online