Klarna’s innovative approach to customer service using an AI chatbot, powered by OpenAI, has stirred significant discussion in the tech and business sectors. The company boldly aims to replace a substantial portion of its customer support roles with this technology. But is this AI-driven solution truly revolutionary, and is it poised to reshape the future of customer service jobs as dramatically as Klarna suggests? This article delves into the reality behind the hype.
Klarna’s commitment to Artificial Intelligence, particularly OpenAI’s technology, is undeniable. Sources within the company confirm a company-wide initiative to integrate AI across operations, aiming for enhanced efficiency and product innovation. This push is spearheaded by Klarna’s CEO and co-founder, Sebastian Siemiatkowski, a known advocate for OpenAI. Fortune reported on Siemiatkowski’s proactive engagement with OpenAI CEO Sam Altman, highlighting Klarna’s ambition to be at the forefront of AI application.
On February 27th, Klarna released impressive initial results, just a month after deploying their AI assistant:
“Klarna today announced its AI assistant powered by OpenAI. Now live globally for 1 month, the numbers speak for themselves:
- The AI assistant has had 2.3 million conversations, two-thirds of Klarna’s customer service chats
- It is doing the equivalent work of 700 full-time agents
- It is on par with human agents in regard to customer satisfaction score
- It is more accurate in errand resolution, leading to a 25% drop in repeat inquiries
- Customers now resolve errands in less than 2 mins compared to 11 mins previously
- It’s available in 23 markets, 24/7 and communicates in more than 35 languages
- It’s estimated to drive a $40 million USD in profit improvement to Klarna in 2024”
Siemiatkowski further emphasized these statistics on X, sparking debate about the implications for customer support professionals:
“In our case, customer service has been handled by on average 3000 full time agents employed by our customer service / outsourcing partners. Those partners employ 200, 000 people, so in the short term this will only mean that those agents will work for other customers of those partners.
But in the longer term, as more companies adopt these technologies, we believe society needs to consider the impact. While it may be a positive impact for society as a whole, we need to consider the implications for the individuals affected.We decided to share these statistics to raise awareness and encourage a proactive approach to the topic of AI. For decision makers worldwide to recognise this is not just “in the future”, this is happening right now.”
To understand the real-world impact and capabilities of this Klarna Customer Service chatbot, a hands-on experience is essential. Is it truly a threat to customer support jobs, or simply an evolution of the field? After testing the chatbot, it’s evident that while sophisticated, its disruptive potential might be overstated.
Experiencing the Klarna Chatbot for Customer Service
Interacting with the Klarna customer service chatbot is straightforward. Users can initiate a conversation with a general inquiry or focus on a specific transaction. The response time is notably quick, typically under 20 seconds.
Klarna chatbot interface displaying how it works
The chatbot functions by retrieving relevant information from Klarna’s documentation and presenting it, sometimes verbatim and sometimes in summary. Its intelligence shines when contextualizing responses with order-specific details, such as purchase price, merchant, and date.
Klarna chatbot providing context about a specific purchase including price and date
The chatbot can also engage in broader conversations, like providing company history, but it appears deliberately limited to pre-approved topics. While these restrictions can be circumvented, it requires intentional effort.
Common Customer Service Queries Handled by the Chatbot
Interestingly, when prompted about frequently asked questions, the Klarna customer service bot provided a categorized list:
- Payment Inquiries: Guidance on making payments, payment confirmations, and due date extensions.
- Order Issues: Questions or problems related to orders, such as non-delivery or damaged goods.
- Personal Data Updates: Assistance with updating personal information.
- Unauthorized Transactions: Reporting and addressing unauthorized purchases.
- Klarna Policy Questions: General inquiries about Klarna’s terms and policies.
Less frequent, more complex scenarios also addressed include:
- Technical glitches causing purchases to not appear in accounts.
- Clarification of unique payment arrangements.
- Troubleshooting uncommon error messages.
Safety Measures and Limitations of the AI Chatbot
A primary concern with AI chatbots is their potential to “hallucinate” or fabricate information. Klarna’s chatbot, utilizing ChatGPT APIs, is inherently susceptible to this. While ChatGPT’s functionality relies on predicting the most probable next word based on input, this can lead to inaccuracies if not carefully managed.
Air Canada’s chatbot incident, where it provided incorrect policy details resulting in legal repercussions, highlights the risks of unchecked AI. Klarna has implemented safety protocols, and initial testing suggests the bot avoids generating incorrect information. A likely verification process compares the chatbot’s output against a pre-approved list of topics, ensuring policy information is delivered precisely.
When faced with topics outside its defined scope, the Klarna customer service chatbot seamlessly transitions the user to a human agent.
Klarna chatbot handing off to a human agent when asked about topics outside its scope
Examining the System Prompt
System prompts are crucial for guiding chatbot behavior, defining conversation boundaries, topic limitations, and high-level instructions. While the complete system prompt for Klarna customer service chatbot remains undisclosed, probing revealed insights into topics triggering human agent handover.
Examples of topics the chatbot is programmed to avoid discussing, leading to human agent transfer
These restrictions, likely embedded within the system prompt, explain the chatbot’s limitations and its tendency to defer to human agents when conversations deviate from pre-defined parameters.
Prompt Injection Vulnerabilities
Prompt injection is another significant security challenge for chatbots. As cybersecurity expert Simon Willison noted, even with extensive testing, unforeseen grammatical structures can potentially bypass defenses and manipulate a language model.
Despite Klarna’s careful development, users have successfully employed prompt injection techniques to elicit unintended outputs, demonstrating the inherent vulnerability of even sophisticated chatbots.
Example of prompt injection where a user tricked the Klarna chatbot into generating code
Multilingual Customer Support
A standout feature of Klarna customer service chatbot is its multilingual capability, supporting conversations in approximately 30 languages. Remarkably, it extends to less common languages, like Hungarian, which even major platforms like Amazon do not fully support. The chatbot accurately translated information and maintained grammatical correctness comparable to a human speaker, showcasing its potential for broad international customer support.
The True Scope of AI Chatbot Replacement in Customer Service
While Klarna’s CEO attributes the chatbot to replacing 700 customer service roles, it’s crucial to understand the nature of this replacement. The chatbot primarily handles basic, initial-level support (L1), acting as a filter for simpler queries before escalating complex issues to human agents.
This AI chatbot excels at automating Level 1 support, addressing frequently asked questions with standard answers. It serves as an efficient initial point of contact, reducing the need for numerous, less specialized L1 support staff. This aligns with the CEO’s message regarding job displacement within this specific tier of customer service.
Klarna’s future strategy likely involves expanding the chatbot’s capabilities to manage more intricate queries autonomously. However, the inherent limitations of LLMs in handling sensitive data and ensuring deterministic responses necessitate a cautious approach, particularly for a Fintech company where data security is paramount. Sensitive financial information will likely remain within the domain of human agents.
Is AI-Driven L1 Customer Service Truly Groundbreaking?
Klarna’s AI chatbot for Klarna customer service is undoubtedly a sophisticated implementation, potentially leading the industry in scale and comprehensiveness at present. This is reinforced by Klarna’s close partnership with OpenAI, suggesting a collaborative environment for AI innovation.
The claimed $40 million annual savings are plausible, considering:
- The chatbot handles the workload of 700 agents concurrently.
- 24/7 support necessitates roughly 3 agents per role in shifts.
- This equates to a potential displacement of around 2,100 L1 support positions.
- This figure aligns with Klarna’s reported 2/3 reduction in customer support load.
- The implied average annual cost per agent was approximately $19,000.
However, automating L1 support itself is not a novel concept. Automated phone systems, using “press 1 for X, 2 for Y” menus, have long served as a rudimentary form of L1 support automation.
A software engineer’s experience at Citibank around 2008 illustrates this point. Citibank automated 95% of phone support requests over two years, closing a call center employing 7,000 people in Mexico. Interestingly, they also implemented a system to prioritize high-value customers, routing them directly to human agents to avoid frustrating them with automated systems.
The key differentiator between Klarna and earlier examples like Citibank is Klarna’s public announcement and emphasis on the AI aspect.
In essence, the revolutionary aspect lies within AI’s potential, not necessarily in the business concept of L1 support automation itself. Klarna’s adoption of AI for Klarna customer service highlights a shift from potentially overspending on traditional customer support in a previous economic climate (ZIRP era) to embracing efficient, AI-driven solutions. Many established companies had already implemented non-AI automation for basic customer queries long before LLMs became prevalent. Companies like Uber, for instance, combine readily available information and automated processes for common issues (like UberEats order discrepancies) with human agent support for complex cases, often without relying on AI agents for basic tasks.
The accessibility of LLM-based L1 support automation is expanding rapidly. Klarna’s publicized success in saving millions using an AI chatbot built relatively quickly will undoubtedly spur other companies to explore similar AI-driven Klarna customer service solutions.
Is AI Chatbot Development Addressing the Core Issue?
Consider this perspective:
Image depicting that creating an AI chatbot is easier than fixing complex user interfaces and product design
While offering chat support enhances customer convenience, over-reliance on it as a primary feedback channel might indicate a lack of focus on user interface (UI) and product simplification. If customers frequently resort to chat support, it could signal underlying usability issues within the product itself. Therefore, while AI chatbots offer valuable customer service solutions, they should not overshadow the importance of intuitive design and user-friendly products.
Subscribe to my weekly newsletter to get articles like this in your inbox. It’s a pretty good read – and the #1 tech newsletter on Substack.
