More AI Isn’t Always Better: Putting AI Where It Creates Real Business Value

AI has quickly become part of how businesses operate. In Singapore’s fintech sector, it is increasingly difficult to find a firm that has not experimented with AI in some form, whether to support customer service, streamline operations or help employees work more efficiently. Yet widespread adoption has not necessarily translated into business value. According to a report by Quinlan and Associates, 95% of traditional financial institutions are actively exploring or have adopted AI, while 71% struggle to demonstrate clear returns on their investments. The question for leaders is no longer whether to adopt AI, but how to deploy it where it makes a meaningful difference.

One reason firms struggle to realise AI’s potential is that they approach it primarily as a productivity tool. Most people’s experience of AI starts with large language models (LLMs): ask a question, receive an answer, and use it to complete a task faster. Naturally, businesses have followed the same pattern, layering AI onto existing workflows to draft documents, summarise information or assist with research. These applications can deliver genuine benefits, but they often improve individual tasks without fundamentally changing how work gets done. If AI only makes an inefficient process marginally faster, firms may be capturing only a fraction of its potential value.

This matters because the biggest opportunities may lie beyond individual tasks. Consider a customer onboarding process that involves collecting documents, conducting checks, reviewing exceptions and updating internal systems. A firm might introduce AI to extract information from documents or help employees review applications more quickly. However, what if AI could coordinate several of these steps, trigger the necessary checks, update connected systems and escalate only unusual cases for human review? Rather than improving isolated activities, the firm could redesign the entire process around AI. The opportunity is not simply to help employees do the same work faster, but to reconsider which activities need to be done by people at all.

Realising this opportunity requires leaders to rethink their operating models before deciding where AI belongs. One useful exercise is to imagine building the business from scratch today, with access to the most advanced AI tools available. Would the firm structure its teams, allocate its resources and design its processes in the same way? Perhaps not. Some activities could be handled by AI agents working across multiple steps, while employees focus on judgement, complex decisions and customer relationships. Other activities may still require human involvement because of regulatory obligations, risk or the need for accountability. Starting with the desired business outcome, rather than a particular AI tool, helps firms identify where redesign could create the greatest value.

This does not mean every process should be automated end to end. AI introduces its own costs and limitations, and more sophisticated applications are not automatically better. Token consumption, for example, can rise when models repeatedly process large amounts of context, use complex reasoning or make multiple calls to complete a task. Deploying an expensive model for a straightforward, repetitive activity may therefore cost more than the value it creates. Conversely, a more capable model may justify its cost when handling complex work that would otherwise require substantial human effort. The goal is not to maximise AI usage or minimise token costs in isolation, but to find the right balance between capability, cost, risk and business value.

There is, however, a second challenge: firms cannot redesign around AI’s potential if they do not understand its capabilities and limitations. This is where experimentation becomes important. Leaders who only encounter AI through occasional demonstrations or narrowly defined pilots may underestimate what it can do, or overlook where it fails. Regular, hands-on use helps employees develop a more practical understanding of which tasks AI handles well, where human judgement remains essential and how different tools perform in real working conditions. Experimentation should therefore be more than a series of isolated proofs of concept; it should build the organisational understanding needed to make better deployment decisions.

Ultimately, these two challenges are closely connected. Firms need to understand AI to identify the right opportunities, but they also need to rethink their operating models to turn that understanding into meaningful outcomes. We believe the starting point is by experimenting more deliberately and broadly, allowing teams to explore AI across their day-to-day work while measuring the results. Then, use those lessons to identify high-value processes that could be redesigned from end to end, rather than simply adding another tool to the existing workflow. The strongest results will come from matching the right AI capabilities to the right activities, with appropriate safeguards and clear measures of success. The real advantage will not belong to firms that use the most AI, but to those that understand where it works best and build their businesses accordingly.