Navigating AI Integration: Understanding Organisational Readiness Beyond Technology

As financial institutions delve deeper into artificial intelligence (AI), the challenge at hand shifts from understanding technology capabilities to reshaping organizational structures. Maximilian Groth, co-founder and CEO of Decentriq, emphasizes that successful AI utilization hinges not on the quantity of AI applications but rather on how institutions align their operational models to integrate AI effectively.

Strategies for AI Integration Challenges

The discourse around AI in financial services often centers on technical abilities, such as model performance, speed, and task automation. However, a critical yet less examined concern is whether these organizations are structurally equipped to leverage AI optimally. With the impending EU AI Act effective from August 2, 2026, organizations face pressure to enhance transparency and accountability regarding AI-driven decisions. Similarly, UK regulators are exploring the potential reach of AI in retail finance, which also highlights this accountability issue.

Groth's perspective is particularly insightful, given Decentriq's role in confidential computing—a niche that permits data sharing between organizations without exposing raw datasets. His assessment of AI readiness is more focused on an organization’s willingness to share data than on specific AI models themselves. This raises a vital question: Are institutions prepared to recognize and resolve their internal apprehensions surrounding data utilization?

The Approval Chain: A Bottleneck for AI Adoption

When asked about operational weaknesses exposed by AI, Groth points to the review chain within financial services—a system designed around human oversight. He states, “The first thing to break is the approval chain”—a process that fails to accommodate the rapid generation of AI outputs. AI’s speed can lead to overwhelming backlogs as human reviewers struggle to keep pace, rendering traditional review methods impractical or even merely performative.

Many firms address this bottleneck with increased governance, typically through committees. However, Groth argues that this model is misguided. "Governance-as-committee fails because it is built for periodic decision-making, not continual oversight,” he asserts. Instead, he advocates for predetermined decision rights categorized by output type, ensuring efficient workflow where certain outputs can be automated, while others may need review or escalation.

Accountability in AI-Driven Decisions

Amid concerns about accountability when AI systems err, Groth maintains that the responsibility does not diminish with machine involvement. “If an institution defines which decisions an AI can make, it is fully accountable for the outcomes,” he states, drawing parallels to the expectations placed on junior analysts operating within defined parameters. He critiques the misconception that AI involvement in decision-making weakens responsibility; on the contrary, clearly documented boundaries can actually strengthen accountability.

Assessing Organizational Willingness for Data Sharing

Through Decentriq’s work, Groth has observed that an organization's approach to data use often reveals its readiness to embrace AI. A case in point involves a collaboration between a global wealth manager and a prominent publisher targeting high-net-worth individuals. Both entities faced legal constraints that prevented them from sharing raw customer data. The real obstacle was not the models but rather the firms’ inability to find safe, compliant means to merge signals without exposing sensitive information.

This dynamic illustrates a crucial insight: if firms hesitate over data integration, no advanced AI model will suffice to bridge that gap. Hence, a solid indication of an organization’s AI readiness lies in how such projects are stymied, particularly whether the cause is technological in nature or related to the infrastructure for decision-making.

Indicators of Successful AI Deployment

Rather than focusing solely on the technology in use, Groth suggests observing points of friction in AI projects. “A better signal is friction location: watch where an AI initiative stalls," he explains. "If it halts due to technical needs, that's a manageable issue. If it stalls because of ambiguous accountability, that's where a real problem lies.”

For instance, Decentriq collaborated with a Swiss bank that had already clarified its policies regarding the use of first-party data. Following this groundwork, the bank achieved a remarkable 129% increase in click-through rates and a 44% decrease in cost per page view. Groth notes that these enhancements arose not from a superior model but from a well-defined operational framework established ahead of time.

The Pitfalls of Misconceiving AI Readiness

Groth identifies a prevalent error among firms: misinterpreting AI readiness as mere procurement. “The most common mistake is treating AI readiness as a purchase decision: acquiring the technology and assuming the operational model will adjust accordingly.” This often leads to sluggish progress and sunk costs, with systems potentially underperforming relative to original business cases while expenses run unchecked.

This issue often remains undetected until significant time has elapsed. Unlike canceled projects that undergo evaluations, stalled initiatives continue to incur costs without accountability, as they appear successful on paper. This gap between perceived and actual effectiveness can strain resources and impede strategic goals.

Future Considerations and Operational Readiness

As the industry braces for stricter regulatory requirements, Groth emphasizes the need for firms to clarify who owns outcomes in AI-assisted decisions. The distinction between investing in capabilities versus resolving internal operational hurdles will likely shape the success of AI initiatives in the years to come. The roadmap ahead isn't about adding more technology; it’s about fundamentally rethinking how organizational frameworks can better accommodate AI-driven processes.