Key Takeaways
- Prioritize cleaning and organizing your data before developing AI models.
- Empower your existing experts with AI tools, rather than aiming to replace them.
- Focus on AI systems that are transparent, explainable, and operate under clear rules.
- Concentrate on practical AI applications that solve real internal problems, not just flashy demos.
- Consider specialized, smaller AI models tailored to specific tasks for better results and easier management.
- Ensure your teams understand data and AI basics to use new tools effectively.
Artificial intelligence in finance is moving beyond just talk, but making it work often hits roadblocks like poor data quality and a lack of clear rules. Insights from industry leaders suggest a practical path forward for banks, insurers, and fintechs wanting to use AI safely and effectively.
The biggest hurdle for successful AI isn’t complex technology or regulations; it’s messy data. Experts agree that information often lives in too many disconnected places and unsuitable formats. Christian Buckner from Altair, which owns RapidMiner, highlighted that real progress begins by integrating and contextualizing data, perhaps using knowledge graphs to connect systems. This creates a solid base for reliable automation.
Devavrat Shah, an AI professor at MIT and co-founder of Ikigai Labs, agreed. He emphasized the need for “specialized, purpose-built models that live within the enterprise.” Instead of chasing flashy chatbot demos, the focus should be on tools that tackle the essential, unglamorous work of planning, reconciling, and forecasting.
When it comes to trusting AI, Shah explained that AI isn’t flawless but provides valuable direction. It should be treated as an input for decision-making, not a final answer. Michael Berthold, CEO and co-founder of KNIME, added that the level of trust depends on the AI’s task. Spotting trends is different from forecasting revenues, where accuracy is critical.
Berthold stressed that AI systems in finance must be transparent, allowing users to understand how a result was reached. Buckner pointed out that governance should be built into the data layer itself. This involves defining who sees what, what models can do with data, and how outcomes are evaluated, ensuring every action is traceable and builds confidence.
There’s a common misunderstanding that no-code AI platforms are too simple for serious work. However, Berthold explained that tools like KNIME help teams build sophisticated, automated workflows more safely and transparently than older methods. He emphasized that data literacy is crucial: users need to understand what the AI does, even if they don’t write the code.
Buckner expanded on this, noting how platforms like RapidMiner can empower non-technical teams to conduct their own analyses within secure guidelines. This approach frees up expert users for more complex challenges and helps business teams move faster, fostering collaboration.
The experts also challenged the idea that bigger AI models are always better. Shah was clear that the future likely involves “small, contextual models that live within the enterprise.” These specialized agents are often more efficient, cheaper to run, and far less risky than relying on massive, external general-purpose AI systems.
A key theme was that AI will enhance human capabilities, not simply replace jobs. Berthold predicted a shift away from the “chatbot everything” trend towards AI that “quietly observes and offers meaningful suggestions, like a co-pilot.” Shah sees this as machines learning our language, leading to more intuitive teamwork.
However, Shah cautioned that “explainability is now just as important as accuracy.” If people don’t understand an AI’s output, they won’t trust or use it. Buckner framed AI’s benefit in terms of providing teams with leverage to work smarter and faster. According to an article on Forbes, the real winners in AI will be those quietly building robust, understandable systems where humans and machines effectively collaborate.
