News
A leading neobroker shuts down its chatbot: What companies must learn about AI in customer service
A leading European neobroker is investing tens of millions in human service – and replacing its chatbot. Why this isn't an AI problem, but a concept problem, and how hybrid AI strategies actually work.
As publicly reported recently, a leading European neobroker (Trade Republic) is investing a double-digit million sum into its customer service. More than 1,000 service agents, 24/7, in eight languages, across locations all over Europe. Until recently, the only contact channel was an automated in-app chat with FAQ logic. No human. No phone number. Within a year, complaints to consumer protection agencies had more than doubled. The company’s response: back to humans.
This is not an AI problem, but a concept problem – and far from an isolated case. It is a misleading signal currently being sent to the entire industry. We have observed the underlying pattern for years. Yet AI in customer service can work very well and serve as the ideal complement in customer interactions, especially when deployed in a hybrid model.
The headlines sound like a defeat for AI in customer service. But look behind the façade and you’ll see that the problem isn’t the AI itself, but how companies deploy it. Today, buzzwords like “AI customer service”, “AI chatbot”, and “automation” are thrown around freely, and many companies fall for marketing promises. As a result, they implement artificial intelligence in their service operations but fail to ensure that it actually works as a bridge between the customer and the company.
The solution isn’t to switch off the AI, but to adopt a hybrid approach built on a unified knowledge base — AI for the recurring requests, humans for the complex cases. This article shows what that looks like in practice.
- Why the shift in sentiment is happening now
- The six typical mistakes when deploying AI in customer service
- Solving the quality problem technically, not rhetorically
- How to do it right: the hybrid approach
- There is no one-size-fits-all, but a fitting solution for every company size
- The real lesson from the Trade Republic case
- Free strategy check
- FAQ
Why the shift in sentiment is happening now
For more than 15 years, we at OMQ have been supporting companies across the DACH region in automating their customer service. Insurers and banks, industrial companies and e-commerce brands, but also public institutions such as chambers of commerce and IHKs. During this time, we have seen hundreds of chatbot, email, and agent-assist projects.
Customer frustration doesn’t arise because AI is involved. It arises because the chatbot blocks instead of helping. Because it spits out canned answers when there’s a concrete problem. And because there is no path to a human when the machine reaches its limits.
Media outlets and consumer advocates increasingly report critically about automated service. The “return to humans” is being marketed in press releases as a quality differentiator. That sounds appealing, but it doesn’t solve the structural problem. Removing AI doesn’t address the root cause. It only avoids the symptoms. The demand for fast, available, personal service remains. With every step of growth, those demands rise – and so do the personnel costs.
The six typical mistakes when deploying AI in customer service
Anyone who works with AI projects up close sees the same mistakes repeatedly. Six of them appear especially often, and they explain almost every failed automation. They are not technical, but strategic mistakes.
The chatbot is supposed to do everything, serve every channel, take work off every employee. The result: customers are forced into a one-way street. As soon as their issue becomes complex, helplessness emerges instead of help, and cost-per-contact rises through repeat contacts and escalations in the end anyway.
Many chatbots are at their core just FAQ systems in a polished UI. They answer the ten most frequent questions and fail at the eleventh. Anyone who wants to deploy AI seriously needs a deep, well-maintained, and structured knowledge base as a central strategic asset – not as an IT side project.
There is no escalation path to a human. No “Let me connect you with a service agent.” Instead, an endless loop of “I didn’t understand that.” No surprise that customers get angry and the Net Promoter Score plummets.
AI is not a piece of furniture. It needs to be maintained. Products, prices, processes, and legal requirements change constantly. An AI trained a year ago works with outdated knowledge today, and under the EU AI Act, that quickly becomes a compliance risk.
A chatbot on the web, a bot in the email inbox, a voice assistant somewhere in between – but no shared logic. The result is an inconsistent customer experience, a confused service team, and three parallel maintenance streams that consume the service budget.
The chatbot says A. The email bot says B. The human agent says C. Three truths within one company, because every channel runs with its own logic and its own knowledge base. That destroys trust faster than any bad answer, and silently triples the total cost of ownership of automation.
Solving the quality problem technically, not rhetorically
Behind almost all of these mistakes lies a deeper doubt: Can I trust an AI that can also hallucinate? It’s a fair question. It deserves a technical answer, not just a rhetorical one.
That’s why at OMQ we developed a multi-stage AI agent pipeline. Every answer our system generates is checked by a second, independent AI model before it is sent. Automatically, in real time. We call it the four-eyes principle, only between two models. Everything is based exclusively on the verified service knowledge of the respective company, hosted in Germany, ISO 27001 certified, with no training data sharing, and fully compliant with GDPR and the EU AI Act.
This setup directly addresses the central loss of trust in AI in customer service. It tackles the first three mistakes before they can even occur, and makes the solution audit-ready and board-ready.
Dual Control for AI: Two Models, one answer.
How to do it right: the hybrid approach
AI and humans are not an either-or. Set up correctly, they complement each other. Only in this combination does the quality leap that many companies are looking for emerge – and only there does the ROI materialize that lasts beyond the pilot phase.
Up to 80% of all customer requests are recurring. Resetting passwords. Shipping status. Changing the billing address. Checking contract terms. This is where AI is clearly superior: available 24/7, answering in seconds, infinitely scalable, with no marginal personnel cost.
The remaining 20% are the complex, emotional cases. The worried customer. The angry caller. The person on the verge of cancelling. This is where humans belong, with time, attention, and real decision-making authority – because exactly these 20% determine customer lifetime value and churn rate.
And in between: AI as a real-time assistant for service agents. It suggests answers, provides context, surfaces the relevant passages from the knowledge base. Humans remain the decision-makers. AI accelerates them. The customer on the other side notices nothing of this – except a clear, fast, consistent answer. And the onboarding of new agents shrinks from months to weeks.
The foundation for all this is a unified knowledge base – a single source of truth that simultaneously feeds the AI chatbot, the email bot for email automation, and the agent assistant. That is precisely the OMQ approach. We solve the sixth mistake – contradictory answers per channel – technically, because all channels read from the same source. And the total cost of ownership is cut in half because only one knowledge base needs to be maintained.
There is no one-size-fits-all, but a fitting solution for every company size
What we’ve learned from hundreds of projects: the right use of AI depends heavily on company size, request volume, and budget. The question “do we need a bot?” almost always falls short. The better question is: which combination of AI and humans fits our volume, our team, our compliance requirements, and our customer base?
| Company size / budget | Recommended approach | Benefit |
|---|---|---|
| Small company, limited budget | AI as the primary channel | 24/7 availability without FTE growth |
| Mid-sized company | Hybrid approach: AI plus service team | AI relieves the team, humans solve complex cases, measurable cost-per-contact reduction |
| Large enterprise, high budget | AI as a turbo for the service team | Maximum quality plus scalability without linear FTE growth |
In practice, this means: a mid-sized retailer like myphotobook achieves an automation rate of around 80% in email support with OMQ Reply, without any drop in service quality. On the contrary: service agents focus on the cases where they are truly needed, and the team scales with growth instead of with tickets.
The real lesson from the Trade Republic case
Trade Republic didn’t fail because AI was used. The company failed because AI was used as a façade. As a blocker between itself and its own customers. Not as a bridge to them.
The lesson is not “abolish AI.” The lesson is: deploy AI correctly. With a deep, well-maintained knowledge base. With automatic quality checks. With a clean handover path to humans. With architecture that meets GDPR and the EU AI Act. And with the insight that good technology doesn’t replace humans – it makes them most valuable exactly when they are truly needed.
Anything else is not an AI problem. It’s a concept problem.
Good technology doesn’t replace humans – it makes them most valuable exactly when they are truly needed.Matthias Meisdrock, CEO OMQ GmbH
Free strategy check: Where does AI really help in your service?
If you’re currently asking yourself where your customer service stands today and where AI actually makes sense, we offer a free 30-minute strategy check. We look at your current setup, place your cost-per-contact within the DACH benchmark, and show concretely where automation has the biggest ROI lever, where humans are indispensable, and where a clean combination secures your scaling strategy.
Book an appointment or write to me directly on LinkedIn. I’m curious to hear about your experience: where has AI worked in your service so far, and where hasn’t it?