Scalable Capital links €60 billion of client assets to external AI assistants.

      Scalable Capital has expanded its investment platform by allowing clients to use ChatGPT and Claude to analyze their portfolios and execute trades, instead of relying solely on the broker’s own app. The Munich-based firm claims it is the first European bank to implement this, coinciding with the industry's exploration of how AI agents can manage trading capabilities in practice.

      Founded in 2014, Scalable is not a small entity; it boasts over a million clients and manages more than €60 billion, primarily serving Germany and Austria while also having a presence in Italy, Spain, France, and the Netherlands. The strategy of partnering with external AI assistants before integrating such functionalities into its own app is noteworthy. Typically, institutions first develop in-house assistants and may expose them to third parties later, if at all, but Scalable has opted for a different approach.

      Erik Podzuweit, the company's founder and co-CEO, described this launch as an initial step rather than a complete offering. He noted that many people might still be reluctant to allow ChatGPT to review or manage their portfolios, referring to the initiative as "a first step." Podzuweit candidly mentioned that while he believes using AI could yield better average returns, this has yet to be proven—a refreshing honesty compared to the usual claims in this field.

      It’s important to clarify what has changed. A chatbot capable of analyzing a portfolio serves as a convenience, while one that can execute trades represents a different level of functionality. The distinction becomes significant when considering the implications of a model misinterpreting an instruction.

      Prompt injection poses a risk with no clear solution at present. A chatbot that reads a portfolio can interpret any text it encounters, and a model authorized to make trades is inherently a more appealing target than one restricted to summarization. Scalable maintains that security measures are in place to protect customer accounts, although it has not publicly detailed these protocols. Key questions remain unanswered, such as how a trade instruction is verified, what limitations apply, and who is responsible if an assistant acts on an unclear request.

      European banks have been progressively adopting conversational interfaces, with bunq introducing its generative AI assistant and Citi’s Jane Fraser suggesting that the future of banking will hinge on two AI races. Most of this innovation has kept the technology within the institution's control.

      Engaging external platforms alters the regulatory landscape. A German broker is subject to MiFID II suitability and appropriateness standards, and it is unclear how these rules apply when the advising interface for a retail client belongs to OpenAI or Anthropic rather than the firm managing the assets.

      Scalable appears to adopt the view that the assistant is executing commands rather than providing advice, which is a defensible position, albeit a delicate one. When a model summarizes a portfolio and the user responds to that summary, the line between information and recommendation becomes intricate.

      There is also a broader distribution concern at play. Brokers have spent the past decade vying to dominate the app on a customer’s home screen, and if the entry point shifts to a more generalized assistant, that effort may not yield continued growth.

      The commercial rationale is also straightforward. If users increasingly start financial activities within a chatbot rather than an app, the broker linked to that chatbot will capture the engagement, while those waiting to develop their own assistants may miss out.

      Competitors won't take long to react. Trade Republic, Revolut, and traditional banks face the same distribution challenge, and none has a clear motive to surrender this channel to a Munich-based competitor.

      This is the gamble, and Podzuweit has effectively acknowledged it. Whether this will lead to improved returns for anyone remains an unanswered question, one that will require significantly more time than the launch phase to resolve.