6 minute read · Published July 5, 2024

AI won’t kill support - it makes it more important

Latest Update July 11, 2024

Opening an article with “AI changes everything” is the most trite, cringe and boring thing you can do in 2024. So that’s exactly what I’ll do here 😎

AI changes everything, especially support. Klarna famously said their chatbot resolved as many tickets as 700 full-time human agents.

And it’s not just Klarna: More and more companies use AI solutions like CommandBar’s Copilot to automate support and deflect tickets. Will this put an end to the era of call centers filled with rows of headset-wearing support agents?

Support is becoming a different function: Away from answering tickets all day and towards - well, what exactly?

Before AI, support wrote (and writes) responses to questions and sometimes jumps on calls. With AI, support could become a function that contributes directly to the product and owns in-product surfaces that help users help themselves.

This might sound odd in a world where many see AI more as a threat than as an opportunity, especially for support. But AI turns support into more than a necessary expense - it turns it into a proactive contributor to user assistance.

We’re pioneering this at CommandBar, where our support engineer Matt and his team are actively contributing to the product roadmap. I interviewed Matt to inform this article about our take on how support teams will evolve and the role AI will play in that.

How support (really) works

Prospects interested in Copilot constantly tell us that half of their support tickets are resolved with “read this help doc.”

Our support engineer Matt calls these solved problems. They have known answers in our documentation. It’s stuff like “How do I change the color of that button?”. They’re answered in seconds.

For many prospects for Copilot, these kinds of questions are 40-50% of the support tickets they get - and LLMs are great at solving these questions, instantly and more cheaply than a human agent. This is annoying because it ties up support agents from solving issues that require human ingenuity - the unsolved problems.

Here’s an example from our own product: Customers and prospects constantly ask about the ideal dimensions for thumbnails in HelpHub. This doesn’t take long to answer, but happens repeatedly. So Matt added the answer to our documentation and rarely has to answer the question again:

commandbar copilot example

We believe these solved problems will be handled more and more by AI. LLMs are already great at answering questions that can be answered with documentation. As AI chatbots grow into full-fledged support agents, they’ll resolve even more issues with known answers.

Right now, LLM chatbots can solve issues where the information is known. AI agents like Copilot can even solve issues where the process is known.

Unsolved problems

Unsolved problems are things nobody’s asked about before and that need human assistance. An example of this: Testing the limits of an integration or iterating on specific CSS overrides.

You can’t teach an AI nonexistent information or workflows. So when you’re in uncharted territory, your support team can shine and deeply engage with the user to solve the problem.

With AI handling the smaller stuff, they’ll have the time to invest in these issues. There’s another reason this is great: In deeply engaging with customers, support often discovers important sticking points that can become roadmap items.

We believe this is the future of support functions: Conserving human work for the unsolved problems and making a big impact in solving them. Not just for the user, but for the product and the company.

To make sure our support team (mostly) deals with issues worth their time, we use Copilot ourselves - for which Matt owns the source materials.

Why support owns more than just tickets

At the core of good support is knowledge. Knowledge on how to do something, knowledge on data the user can’t access, etc. That knowledge mainly exists in two places:

  • The heads of support workers
  • Documentation

These have a problem: Support workers have limited resources and work hours. Users don’t like to read docs. But using AI, we can surface the knowledge from the docs in more and more interfaces.

AI impact on support

Matt owns our documentation, which effectively means he owns two in-product surfaces:

  • HelpHub (our resource center)
  • Copilot (our AI chat)

Both feed on our documentation, which makes our docs an important part of deflecting the solved problems.

Matt effectively manages a 24/7 support agent that lives inside our product and handles easy questions users could answer themselves. Imagine hearing a decade ago that support owns surfaces inside the product!

Now, if you hired a new human agent, you’d probably start by teaching them how your product works, then told them the answers to the most common questions and finally taught them how to do more complex things using the tooling you have.

Training Copilot is similar:

  1. First, you point it at your documentation and it learns it.
CommandBar Copilot training

Then you give it answers to the most common questions.

CommandBar Copilot answer
  1. Then you can teach it your tools and processes by connecting it to APIs and instructing it with workflows:
CommandBar Copilot workflow

Because capabilities like these now exist, we believe that in the age of AI, support functions will increasingly own an AI system they create content for and optimize.

Matt told me he only spends 20% of this time answering “known questions” and focuses on the tricky stuff. These are things like testing the limits of an integration or building highly specific things to customer use cases.

Why support should contribute to the roadmap

It’s a well-known problem that product teams don’t talk to users enough. But you know who talks to users? Support teams solving real problems together with customers!

That’s why our support function does much more than respond to tickets. As Matt told me: “We do a little bit of user research. We have such a great chance to watch and learn customers actually using our product.”

Matt told me he frequently asks questions like “How would you like this to work?” or gathers feature wishlists that impact the roadmap.

That’s a part of his role he calls Mini PM: “We really are drivers of removing user confusion from the product.” That’s because in directly talking with users, Matt uncovers many issues a top-down product manager may not have found.

As an example, Matt recently realized that customers were throwing off analytics when testing nudges. He realized it’d be better if log events (which power analytics) were suppressed when someone is logged into the editor. Since he’s an engineer, he created a quick PR and submitted it.

That way, support contributes to building the product while also solving their own problems (the analytics question won’t happen again).

We think it’ll be more and more normal for support to contribute directly to the roadmap and shape the product. As the problems they solve become more complex, they can flex their problem-solving muscle more, which makes them even more impactful.

The support team of the future

Support functions will evolve: With AI handling the easy stuff, we’re seeing support teams become more proactive like Matt’s team - doing user research, impacting the roadmap and solving challenging problems.

People might say support work is dead - and support teams might shrink. But support’s closeness to users and its ownership over the AI that handles the easy stuff sets it up perfectly to make a bigger impact in the future.

Copy icon
X logo
LinkedIn logo

Zero fluff.

Our stories 🤝 your inbox

Email Icon

Up Next

Unveiling the top PLG companies: A comprehensive guide for SaaS leaders in product-led growth

Never miss another
fabulous article.

Weekly digest in your inbox.

Email Icon
Continue Reading

Up Next

What product people need to know about LangChain

Read Now

What's hot