Founder and CX lead says AI can turn customer service from a cost center into a profit center

Jaakko Timonen has a background leading CX organizations, and he's currently the founder and CEO of GitHits.

We sat down with Jaakko to learn how he's using AI to better understand his users. Here's what he said.

Different roles, same CX job

My name is Jaakko (Jack) Timonen. I'm currently the cofounder and CEO of GitHits, where I oversee CX.

We are a developer tool company in beta. Modern AI-coding tools allow agents to navigate local codebases, but they can't access frameworks, libraries, SDKs, or other open-source dependencies. So GitHits provides a version-aware index of open-source code on demand, usually in 10-20 seconds for an average repository.

Currently, the vast majority of our users are individual software developers or small startups, but we also have some users from bigger companies. We engage with them primarily through email, Slack, WhatsApp, Intercom, and X.

Before starting GitHits, I was the Chief Customer Officer at an AI consulting company called Softlandia. And before that, I was responsible for the product and customer experience at NoNoNo.com, which helped American consumers escalate and resolve complaints with big businesses.

How AI allows CX leaders to understand usage before responding to users

Our product is fairly new, and the problem we're solving is complex, so we get loads of support requests — particularly around how to set up different coding agents and harnesses. And the high support load has also been exacerbated by recent software supply-chain attacks. In response, many developers, including our customers, quarantine new package versions — including the fixes they requested — for one week before installing them.

As a result, we needed to make some changes to how we handle support. We connected Claude Cowork to BigQuery. And we began using a tool called Datost that accesses the relevant data on BigQuery and enables us to chat with that data on Slack.

Now, we can ask questions about any user and their usage, allowing us to dive deeply into what they are trying to do. So, if a user complains about an issue, we come back with more accurate, data-driven answers — faster.

It's also helpful when we reach out proactively, because we come to the table informed about the customer's usage. And it gives us a much better understanding of our users as a whole, which means we can build better, more relevant features and, ultimately, gain more traction.

Why humans should touch every part of CX

I used to think that 100% of users preferred human customer service over AI. The reason was simple: Human customer service is more human, and we're all humans. But I was wrong. It just depends.

I think how receptive customers are to AI depends on what they're doing. For example, if they need to find some kind of technical nitty-gritty and they can find that information in documentation, then they'll prefer self-service — and AI's help will likely be appreciated. But if they have complaints regarding how the product works or anything related to the product experience, they'll almost always prefer a human in the loop.

That's the main reason we don't use AI to respond to customer feedback. Another is that we are an early-stage company, pre-product-market fit, and we really need to understand our users. That means we have to do things that don't scale.

In the future, we may use AI agents to respond to simple customer requests. But complex requests will likely still require a human touch.

For now, none of our CX decisions rely 100% on AI. We use AI to get better answers faster, but humans still review everything.

Why customer service chatbots don't meet expectations

We recently started allowing our customers to chat with our docs, using Mintlify. We actively guide them there when they reach out via email.

We expected them to really love this. We thought they'd do deep dives into the product to understand its capabilities, but that hasn't happened. Maybe it's because it's such a recent addition. Or maybe our customers prefer human customer service. Time will tell.

When setting up a chatbot like this, I'd warn CX leaders about one thing: Most chatbots are complete rubbish. They don't know how to answer questions, and they end conversations before users even get to ask a follow-up question.

Here are a few tips:

• Most importantly, provide context. Without it, the chatbot doesn't know what to do.
• Train it to be empathetic and match the tone of the customer. For example, if the customer is asking something analytical, that person might be a number cruncher, so you want to provide more analytical answers. But if the message is emotionally driven, then it's more important to deal with the issue empathetically.
• Train it to use non-violent communication.
• Make sure that its goal is not to be technically superior, but to understand what the drivers are behind the customer's messages.

What an AI tool stack looks like in CX

We use AI to build all our workflows. Here's our AI tool stack:

• Claude Code and Codex for AI-enabled coding
• Intercom for support requests and triage

• PostHog for tracking conversions, clicks, and so on. We also use it for screen recordings.
• BigQuery for housing relevant data
• Datost for chatting with data
• Mintlify for allowing our users to chat with our docs

We're planning on moving away from Intercom. Initially, we thought we might go to Pylon, but it's too B2B-focused, so we're exploring options like UseClad or Parahelp instead.

Why successful AI implementation in CX depends on your data and documentation

Every CX leader needs to know one thing: Your success comes down to your data and your documentation.

Regarding data, I always come back to what Cat Wu from Anthropic said on Lenny's Podcast — that episode is an excellent resource, by the way, if you haven't listened to it already. I'll paraphrase: If you want to succeed with Claude Cowork, the first thing you need to do is to integrate it with all the possible data sources so that you have absolutely everything available for AI before you start creating reports and action items.

The input always affects the output, and the input for AI is data. So, make sure your data is structured and clean.

And regarding documentation, you need to document everything. And your customers need to be able to chat with the docs so that it's very easy to find the needles in the haystack. Luckily, the process can be automated to keep the documentation up to date.

Jaakko Shares

The input always affects the output, and the input for AI is data. So, make sure your data is structured and clean.

Why AI-native products still need dashboards

If you're building a consumer-facing AI product, keep in mind that your ticket rate will increase if your tool's interface is purely AI chat.

I noticed this issue when I built my first website with Lovable. Building it with prompts was great, but it wasn't able to handle SEO, and there was no static dashboard where I could easily access things like meta descriptions and page titles. Being forced to ask AI for absolutely everything — one by one, page by page — was not a good experience.

Make sure your product team is still building the necessary dashboards that your customers need.

How AI can change CX from a cost center to a profit center

The majority of CEOs treat customer service and customer success as cost centers, and that leads to bad CX. But AI allows us to treat them as profit centers.

For example, AI can measure what is happening with the customers whose complaints have been resolved. It can find out if they're a churn risk, or if they are even more engaged than before.

If the latter, that will benefit revenue — not only via retention but via word of mouth too. Statistically, a customer whose complaint has been resolved tells six to seven people within a week about that positive experience. And they are likely to spend 30% more in the next month.

In this regard, I recommend reading Delivering Happiness: The Zappos Story.

Why non-violent communication is even more necessary with AI

My go-to tip for CX leaders is always to study non-violent communication. Every support message from a customer includes:

• The message (what are they trying to say?)
• The emotion or feeling behind the message
• The need behind that emotion or feeling
• The request arising from that need”

CX leaders need to understand that. Before you do anything else, your reply should handle all four components. And now, your AI needs to understand that too.

Overall, remember: You need to convey empathy when you respond to customers — not just technical know-how.

Follow along

You can follow along with Jaakko Timonen's work on LinkedIn and X. And check out GitHits.

More expert interviews to come on The CX Lead!