Customer queries resolved in under two minutes with AI

Eleven minutes to under two. That is the resolution time shift an AI-powered customer service assistant delivered when Klarna, the payments business, deployed it across its chat channels in early 2024. In its first month, the assistant completed 2.3 million conversations, handled approximately two-thirds of all customer service chats, and matched the satisfaction scores of human agents. Repeat enquiries fell by 25%. For customer service managers and operations directors at UK SMEs, those numbers are worth examining carefully, because the model is more replicable than it might appear.

The problem that AI addressed

Customer service teams at growing businesses face a familiar bind. Query volumes rise faster than headcount budgets. Response times lengthen. Customers repeat themselves because the first interaction did not resolve their issue. Staff spend most of their day on routine, low-complexity questions: order status, account changes, FAQs. The high-value conversations get less attention than they deserve.

This is not a niche challenge. It is structural. And hiring more agents is not always the right answer, particularly for SMEs operating with tight margins and unpredictable demand.

What the AI assistant actually did

The business deployed an AI assistant integrated directly into its existing chat channels. The assistant was scoped to handle routine query types: common account questions, order status checks, standard account changes, and frequently asked questions. It was not asked to handle complaints, complex disputes, or anything requiring judgement beyond its defined scope.

For those in-scope queries, it worked well. Average resolution time dropped from 11 minutes to under two. The workload handled was equivalent to 700 full-time agents. Customers interacting with the assistant reported satisfaction scores on a par with human agents.

Critically, the deployment included clear escalation paths. When a query fell outside the assistant's scope, it handed off to a human agent with the conversation history intact. The human team was freed to focus on exactly the cases where human judgement added most value.

Why this matters for UK customer service teams

At gecco, our view is that AI adoption is 80% people and culture, and 20% technology. This case demonstrates that clearly. The technology involved, an AI assistant connected to a chat platform, is not extraordinary. What made it work was the decision-making around scope.

The team responsible for this deployment did not try to automate everything. They identified the high-volume, low-complexity queries that consumed most of their agents' time, and they built the assistant to handle those. Everything else stayed with humans.

For a customer service manager or operations director at a UK SME, that distinction matters enormously. The question is not "can AI handle our customer service?" The question is "which specific query types account for most of our volume, and are those queries routine enough to automate reliably?"

For many businesses, FAQs, status queries and standard account changes make up the bulk of contact volume. That is the segment worth targeting first. For a fire maintenance firm, that might be service visit bookings and certificate requests. For a membership body, it might be renewals, CPD records and event bookings.

What a practical implementation looks like

The approach used here follows a pattern that managing directors and operations managers can apply without building bespoke infrastructure. The core steps are straightforward.

First, audit your query types. Pull three months of contact data and categorise every query. Identify the top ten recurring query types by volume. These are your candidates for AI handling.

Second, define scope tightly. Build the assistant to handle only those specific query types. Do not ask it to handle edge cases, complaints, or anything requiring access to sensitive account data it cannot verify.

Third, design the escalation path before launch. Every interaction that falls outside scope must reach a human agent, with full conversation context, within a defined time. This is non-negotiable for customer satisfaction.

Fourth, measure resolution time and repeat contact rate from day one. These two metrics tell you quickly whether the assistant is working or creating additional work downstream.

Fifth, iterate on the scope boundary. As you accumulate data on where the assistant succeeds and where it fails, adjust the handoff criteria. This is an ongoing process, not a one-time setup.

Considerations and limitations

The results described above came from a large-scale deployment with significant resources behind it. UK SMEs should not expect to replicate those numbers immediately, and should be cautious about any vendor promising equivalent outcomes from a basic chatbot integration.

Scope discipline is the single biggest risk factor. If the assistant is asked to handle queries it is not equipped for, satisfaction scores will fall, not rise. A bot that attempts to handle a billing dispute without the right data access or judgement will frustrate customers more than a slow human response would.

Adoption among smaller firms has grown significantly, but confidence remains low: recent data suggests that while AI use among small firms has roughly doubled, fewer than one in five business owners feel confident using it. That confidence gap is real, and it matters for implementation quality.

There are also regulatory requirements UK businesses must meet. UK GDPR transparency duties and ICO guidance mean customers should be told when they are interacting with an AI system. Personal data handled during AI interactions must be processed lawfully and securely. Any customer who wants to escalate a complaint or speak to a human must have a clear and accessible route to do so. These are not optional considerations. They should be built into the assistant's design from the outset, not added afterwards.

Finally, an AI assistant is not a staffing strategy. Klarna itself later said it would make sure customers can always reach a human. It handles volume. It does not replace the relationship-building, empathy, and judgement that human agents provide on complex or sensitive cases. The businesses that see the best results use AI to protect their human team's capacity for that higher-value work, not to eliminate the team.

Making this work for your business

The pattern is clear: scope the assistant tightly to high-volume routine queries, design a robust escalation path, and measure from day one. For customer service managers ready to act on that, the next step is defining exactly which query types are in scope for your business.

gecco builds custom AI assistant workspaces for UK SMEs using its GRAFT methodology, covering goal, role, audience, format and tone, and can scope assistants specifically for customer service use cases such as FAQ handling, order queries and account changes, so your team handles more volume without adding headcount.

Find out where AI fits in your customer service operation

If your team is weighing up which customer service queries AI can handle reliably and where the handoff to humans must happen, that is exactly what the AI Readiness survey is designed to surface.

Take the free AI Readiness survey. You will receive 65+ free resources and a custom AI Readiness report based on your answers. From there you can book a free 30-minute showcase to walk through the results with a gecco adviser.

Take the AI Readiness survey

---

Website · LinkedIn · Case Studies · Newsletter