5 minute read · Published May 20, 2024

Do we still need humans in support? Klarna’s AI chat is the future

Latest Update May 20, 2024

It’s rare that a software company’s product decisions make the mainstream news. It’s even rare when those product decisions barely affect the user’s core experience. Yet that’s what happened when Klarna announced its AI support chat does the equivalent work of 700 full-time human support agents.

From Bloomberg to CBS, Klarna CEO Sebastian Siemiatkowski gave many interviews to highlight the impact of its AI support chat.

The numbers are staggering—the chatbot had 2.3 million conversations, resolves issues in 2 minutes (previously it was 11) and works in 35 languages, 24/7. The impact: $40 million in profit improvement. We’ve heard similar feedback from customers for our AI support Copilot: It deflects tickets and enables the support team to do more of what they’re best at.

Mainstream media barely congratulated him on his business acumen, but focused more on potential jobs impacted by AI. That aspect is important, and Siemiatkowski makes sure to highlight it: “We wanted to make policymakers aware that this isn't something that's happening in the future — it's happening now. We think it is critical that society start thinking about this major change.”

The societal impact matters, but we want to focus on its impact on building great software companies. And while it’s fashionable to proclaim AI will kill job X or job Y, those are usually overstated. But there’s no question that support will evolve.

So let’s dive in and explore how support evolves in the age of AI. First, let’s start with a fundamental question: Why does support matter?

Support does more than resolve tickets (and will become more strategic)

Many successful founders nearly failed in the early stages. They usually tell stories like this: “At first, we built [PRODUCT THAT SOUNDS REALLY COOL], but nobody wanted it. There was no product-market fit. But people loved [MINOR FEATURE], so we pivoted and built [PRODUCT THAT SOUNDS LESS COOL], and growth took off.”

Those stories are ubiquitous—and they only happen if you talk to users (a lot). This highlights an important truth:

Support isn’t a necessary evil, something you have to pay for so users don’t hate you.

Nobody has more contact with users than support. Especially in early(ish) stage companies, user feedback regularly impacts the product roadmap or inspires complete pivots.

A good support function surfaces user sticking points and highlights areas for product improvements. That’s especially important for B2B use cases, where products can be complicated and require a lot of user input to set up.

That’s why support has always been important—and will continue to be. If you replaced a support function with a perfect chatbot that answers every question perfectly, you’d miss a lot of important insights.

That’s why we’ve invested into building analytics into Copilot:


Whether you currently use Copilot or not, you need to know what users are thinking about, complaining about and wishing for.

That’s why I don’t think humans will be banished from support functions. In fact, their work will make a bigger impact.

How human support work will evolve

One of the biggest frustrations of support agents are, dare I say, stupid questions. In most companies, half (or more) support tickets are resolved with “Read this help article”.

Doing the same work over and over again is annoying. It takes away from the time available for real problem solving with customers. When AI surfaces the important docs or, even better, the relevant content inside those docs, those tickets never get submitted.

That’s probably why Klarna’s AI assistant handles two thirds of their customer support tickets. But the fact that Klarna didn’t reduce its support team by two thirds signals that not all tickets are created equal: The third of remaining tickets are probably far more difficult to handle and require the empathy, knowledge and problem-solving ability of a human.

When support agents only receive those tickets, they’ll make a bigger contribution to the company and enjoy their job more.

But that’s not the only impact of AI:

From agent to supervisor

In our entirely non-biased and totally-not-influenced-by-our-incentives opinion, AI support is great. But it isn’t perfect from the get-go. In addition to handling complex support tickets, humans in support will come to supervise AI and improve it:

  • Plugging holes in documentation: If AI is unable to answer questions about certain topics (or users are consistently dissatisfied with the answers), that signals a need for better or more documentation.
  • Fine-tuning AI responses: With things like system prompts, you can improve AI responses and tailor them to the way you want your support to behave.

Copilot does the latter natively. You can add both custom instructions and personality traits to influence the way it answers:

CommandBar Copilot custom prompts

From support team to product team

Even when you have AI support with robust analytics, someone needs to understand users and surface insights from those chats and make them useful to the product team.

This is exactly why humans are still important: The chats and data don’t do much when they’re languishing in an analytics dashboard.

That’s why human support will come to collaborate more with product and other teams in the company: Directly surfacing user needs and suggesting solutions.

That way, the support function will drive a bigger impact on the organization.

So far, we’ve only explored how AI could change how humans work in collaboration with machines. But we think AI will also change what support means.

The end of text-only support

So far, most support (AI or not) is done with text. The user writes a text input and receives a text response. But if you imagine the perfect human user assistant, they’d do much more than tell the user what to do: They’d also show the user how to do something or get pesky little tasks out of the way.

That’s also our vision for Copilot—to do more than text responses and instead guide the user with product tours or any other assistance they may need (that’s what we call “co-browsing”). Here’s what that looks like:


Klarna used its in-house AI teams and a convenient intro to Sam Altman to build its AI assistant. But you don’t need to have billions in revenue and specialized engineering talent to achieve this level of AI support: Copilot lets you offer your users enterprise-grade AI assistance without building from scratch.

If you want to deflect support tickets, elevate your support team and assist your users with more than just text, schedule a demo today.

Copy icon
X logo
LinkedIn logo

Zero fluff.

Our stories 🤝 your inbox

Email Icon

Never miss another
fabulous article.

Weekly digest in your inbox.

Email Icon
Continue Reading

Up Next

Command Palette: Past, present, and future

Read Now

What's hot