But Berenberg, with €39 billion in assets under management and 1,600 employees across Europe and the Americas, isn't interested in caution for caution's sake.

The bank's approach is ruthlessly pragmatic: every AI project must either increase revenue or decrease costs. Period.

"At the end of the day, there has to be a measurable economic impact for the business," Baum explains. "You don't get projects through management if the only outcome is that three employees can go home an hour earlier."

That philosophy drove Berenberg to partner with Google Cloud in 2022, launching a multi-year AI strategy that's already transforming how the bank's equity research analysts, portfolio managers, and investment banking teams work.

"We were able to give portfolio managers a Gen AI tool that made the process of data aggregation and evaluation significantly more efficient," says Baum. "But we also enriched it heavily with context. Our investment IP went into these tools, so they didn't just read reports; they brought our investment know-how into the analysis."

The results: faster research consumption, better coverage, and higher-quality investment decisions. Another tool built similarly helped generate the bank's daily "Morning Mail," a market briefing sent to clients each morning. With AI assistance, the content could be produced 85-90% faster, with human review ensuring quality and compliance.

Then came the evolution. When Google launched NotebookLM, Berenberg adopted it for their "Everyday AI" layer, providing employees with a standard tool for document summarization, translation, and comparison without needing custom development.

Now, with Gemini Enterprise, Berenberg is building what Baum calls a "pyramid strategy": tailored AI at the top for high-impact business problems through contextualized, proprietary data, everyday AI in the middle for productivity, and standard tools at the bottom for individual workflows – all centrally managed, compliant, and connected to Berenberg's data landscape.

"We believe the typical shift is that human roles move to the edges of the models," Baum explains. "Humans generate proprietary data at the beginning, the AI runs the production process, and then at the end, you have a human interface again, because selling still requires empathy and interpersonal skills where AI isn't competitive yet."

That philosophy extends to Gemini Enterprise (formerly AgentSpace), which Berenberg is rolling out across the bank starting in 2026. Employees will access a marketplace of centrally managed agents: some generic (summarization, translation), others role-specific (Research, financial analysis, Due Diligences support, and more). The rollout is intentionally controlled with training and monitoring to ensure measurable impact.

Berenberg chose Google Cloud not just for the model, but for the platform. "I believe the model won't make the biggest difference in the end; it'll be the overall platform," says Baum. "And Google is by far the leader there, especially when it comes to cloud capabilities, storage, flexible performance scaling, and the entire regulatory and governance side." As AI-generated content floods the internet, Berenberg's competitive advantage becomes clearer: proprietary data.

"These AI systems will ultimately become a commodity," Baum predicts. "The companies that vertically integrate these AI systems best on their own data will extract the greatest value." For a 434-year-old bank, that's not a threat: it's an invitation.