A systematic CX improvement programme that transformed customer satisfaction at a major UK fashion retailer — using AI, automation, and smarter queue management instead of headcount.
59% CSAT. That was the number staring at us when we started the customer experience workstream at a major UK fashion e-commerce retailer. To put that in context, anything below 70% is considered poor. Below 60% is a crisis. Customers were unhappy, complaints were escalating, and the CX team was drowning in volume they couldn't manage with their existing headcount.
The obvious solution — the one every consulting deck would recommend — is to hire more agents. Throw bodies at the problem. But as we explain in our piece on deploying AI agents before cutting headcount, headcount is expensive, slow to onboard, and doesn't fix the underlying issues. If customers are contacting you because your systems are broken, more agents just means more people apologising for the same problems. We took a different approach.
Before building anything, we spent two weeks analysing every customer contact from the previous quarter. We categorised 14,000+ tickets by contact reason, resolution type, handle time, and outcome. The findings were stark.
42% of all contacts were what we call "information requests" — queries that required no human judgment, discretion, or empathy. Password resets. "Where is my order?" Return status checks. Delivery time estimates. These were customers waiting 4-6 hours in a queue to get information that existed in a database and could be retrieved in milliseconds.
Another 18% were proactive failure contacts — customers reaching out about problems the business already knew about but hadn't communicated. A shipment delayed by the carrier. A product out of stock after the order was placed. A return received but not yet processed. In every case, the business had the information internally. The customer contacted support because nobody told them first.
Only 40% of contacts genuinely required a human agent: complex complaints, subjective quality disputes, goodwill decisions, multi-order issues. The CX team was spending 60% of their time on work that shouldn't have reached them at all.
We deployed an AI-powered chatbot to handle that 42% of information-request contacts. Not a decision-tree chatbot with pre-scripted responses — an LLM-based agent with real-time access to the order management system, carrier tracking APIs, and returns processing database.
A customer asks "where is my order?" The AI agent looks up the order, checks the carrier tracking status, and provides a specific answer: "Your order was dispatched yesterday via Royal Mail and is expected to arrive by Thursday. Here's your tracking link." No queue. No wait. No human required.
Password resets, address changes, return label requests, delivery date queries, stock availability checks — all handled instantly. The AI agent resolves these in an average of 47 seconds. The human team's average handle time for the same queries was 8 minutes and 20 seconds, not because the agents were slow, but because 7 minutes of that was queue wait time.
Crucially, the AI agent knows its limits. Anything involving a complaint, a refund decision above a threshold, or emotional language gets escalated to a human agent immediately with full context attached. The handoff includes the conversation history so the customer doesn't repeat themselves. We tuned the escalation sensitivity high intentionally — it is better to over-escalate than to have an AI agent fumble a genuine complaint.
Curious what your margin opportunity looks like?
Free Tool
How much margin are you leaving on the table?
Answer 6 questions. Get a personalised margin estimate in under 2 minutes.
Take the Free Margin AuditThe 18% of contacts driven by known-but-uncommunicated problems were the most fixable. These were customers contacting support to ask about issues the business already had data on. The fix was obvious: tell them before they ask.
We built a proactive notification system that monitors three data feeds in real time. Carrier tracking data: if a shipment is delayed beyond its estimated delivery window, the customer receives an automated email and SMS with the updated timeline before they notice the delay themselves. Warehouse processing data: if an order is stuck in picking or packing beyond the expected SLA, the customer gets a notification that their order is taking longer than usual. Returns processing data: when a return is received, scanned, and enters the refund queue, the customer gets a confirmation with an estimated refund date.
The impact was immediate. Within the first month, inbound contacts related to delivery delays dropped by 61%. Returns-related queries dropped by 43%. These customers didn't need to contact support because the business contacted them first. That is not just a CSAT improvement — it is a cost reduction, because every deflected contact is agent time freed up for complex work.
For the 40% of contacts that genuinely required human agents, we overhauled the queue system. Previously, tickets were processed in strict chronological order. A customer with a missing £500 order waited behind a customer asking about a size exchange. First in, first out, regardless of urgency or value.
We implemented priority scoring based on four factors: order value, customer lifetime value, issue severity (detected via NLP analysis of the initial message), and time sensitivity (e.g., a delivery needed for a specific date). High-priority tickets jump the queue. VIP customers get routed to senior agents. Time-sensitive issues get flagged with countdown timers.
Average first-response time for priority tickets dropped from 4.2 hours to 38 minutes. Overall first-response time dropped from 3.8 hours to 1.4 hours — not because we added agents, but because the AI chatbot had absorbed 42% of the volume, freeing the human team to focus on the remaining tickets with dramatically lower queue pressure.
You cannot improve what you don't measure. The retailer had been running quarterly CSAT surveys with a 6% response rate. That is not measurement. That is guessing.
We implemented automated post-interaction CSAT surveys triggered immediately after ticket resolution — both for AI-handled and human-handled interactions. Response rates jumped to 23% because the survey arrives while the experience is still fresh. We also deployed sentiment analysis on the full conversation transcript, giving us a CSAT proxy for every single interaction, not just the ones where customers bother to fill out a survey.
This data feeds a weekly quality dashboard that the CX team lead reviews every Monday. Agents with consistently low scores get additional coaching. Common complaint themes get escalated to the relevant operational team. The AI agent's responses are continuously evaluated, and any pattern of negative sentiment triggers a review of that response category.
The measurement system itself improved CSAT by making problems visible faster. A bad customer experience used to surface weeks later in a quarterly report. Now it surfaces within hours and gets addressed the same week.
Eight months after the programme launched, CSAT reached 80%. From 59%. Without a single additional hire.
The AI chatbot handles 44% of all inbound contacts autonomously, with a 91% resolution rate on those interactions. Proactive notifications reduced inbound contact volume by 22% overall. Average first-response time for human-handled tickets is 1.4 hours, down from 3.8. The human CX team now spends their time on work that actually requires human judgment, empathy, and decision-making — which is exactly what they should be doing.
The cost saving from avoided hires is significant, but it's not the point. The point is that CSAT drives retention, and retention drives lifetime value. A customer who has a good support experience spends 23% more over the following 12 months than one who doesn't. At this retailer's scale, the CSAT improvement maps to a meaningful uplift in repeat purchase rate — which is margin that compounds.
This was one workstream within a broader £6.4M transformation programme. The CX improvements don't show up as a single line item on the P&L, but they underpin everything else. Better customer experience means fewer refunds, fewer complaints, fewer chargebacks, and higher lifetime value. It is the foundation that every other margin initiative — from personalisation to discount suppression — builds on.
Book a free margin audit. We'll assess your CX operation and show you where AI can absorb volume without sacrificing satisfaction.
We go into businesses and make them permanently more profitable. Every initiative is EBITDA-tracked.