TAKEAWAYS
According to a 2024 study, 87% of C-suite executives in Singapore rank generative AI (GenAI) as one of their top three business priorities. In 2025, Singapore’s artificial intelligence (AI) market is projected to reach US$1.4 billion.
Despite its growing prominence, AI is often misunderstood. At its core, AI models are merely mathematical constructs. They have been around for decades and are freely available for use.
However, AI models can operate in tens or hundreds of dimensions (that is, high dimensional spaces), far beyond the three spatial dimensions that humans can perceive. AI’s complexity can be compared to human DNA – each individual’s genome is unique, encoding biological instructions that define traits and functions. Just as every person has unique DNA, every organisation has its “organisational DNA”. Each organisation has its unique business objectives, risks, processes, transactions and data.
To unlock the full potential of AI, organisations must encapsulate their unique “organisational DNA” into their AI models, customising them to function accurately within their specific operational environments. Consequently, generic, off-the-shelf AI models or solutions rarely deliver optimal results for organisations.
Before AI, rules-based systems based on predefined “if-then” rules were generally used to flag suspicious transactions. These systems leverage on simplistic algorithms and often generate an overwhelming number of false positives, due to their inability to cope with the complexity we face in the real world.
My first encounter with AI’s transformative power was about 15 years ago, when I led a team of financial forensic professionals. We experimented with predictive AI models to detect procurement fraud in a large institution based in Singapore. We developed an ensemble of AI models that pinpointed 10 unusual transactions out of tens of millions of transactions. Nine of these were confirmed as fraudulent, and the perpetuators were prosecuted in court.
About three years ago, we applied predictive AI models to detect potential money-laundering transactions for a Singapore-based bank. We did not use a single rules-based algorithm. Our highly complex and customised ensemble of predictive AI models flagged less than 10% of the transactions flagged by the bank’s legacy rules-based system. Remarkably, not only did the AI models capture all the true suspicious transactions captured by the bank’s legacy system, they also uncovered additional true suspicious transactions. If the bank had transitioned from their legacy rules-based system to a full AI-enabled system, the bank could have reallocated more than 90% of its resources (assigned to investigate transactions captured by its legacy system) to perform higher-value activities while being significantly more effective in detecting potential money-laundering transactions.
These experiences underscore the immense capabilities of predictive AI models, despite their being free to use. Such predictive AI models demonstrated a level of logical intelligence beyond human capability. When thoughtfully developed and strategically deployed, predictive AI models can help organisations achieve their business objectives, such as uncovering customer preferences, identifying high-risk transactions, and predicting transaction defaults with unparalleled accuracy.
The recent introduction of GenAI modules like OpenAI’s ChatGPT, Google’s Gemini and Microsoft’s Co-Pilot has significantly increased public awareness of AI.
But is GenAI more powerful than predictive AI? The answer depends on the context – it’s both yes and no.
Predictive AI excels in the ability to accurately uncover trends or prediction outcomes that are beyond human capability. For example, it can predict equipment failures in manufacturing or customer churn in retail with remarkable accuracy, helping organisations achieve business objectives. However, it lacks cognitive abilities.
GenAI, as its name implies, is designed to generate content. It can create unique images using image models or summarise reports using large language models (LLMs). While LLMs exhibit cognitive traits, they lack logical intelligence and do not understand the content. Instead, GenAI simply mimics patterns from its training data, much like a parrot repeating phrases without truly understanding their meaning.
Both GenAI and predictive AI have unique strengths and limitations. Predictive AI is unmatched in prediction and uncovering insights, making it ideal for tasks like forecasting and anomaly detection. GenAI, on the other hand, shines in creative and generative tasks but may produce inaccuracies due to its inherent lack of understanding.
By understanding their respective capabilities, organisations can deploy the right technology to maximise the value AI brings to their operations.
LLMs generate content word by word, predicting each new word based on preceding ones. They convert words into numbers (tokens) and calculate probabilities in high dimensional spaces (vectors) to determine the next word. This process, while powerful, is inherently random and can lead to “hallucinations” – outputs that seem credible but are factually inaccurate or even nonsensical. The unpredictable nature of LLMs can be a strength in creative content generation but a liability in contexts requiring factual accuracy and reliability.
For instance, if a company relies on an LLM to check their customers for adverse reputation, errors could have serious consequences. Legitimate customers might be turned away because of fictitious adverse reputations reported by the LLM, leading to lost business opportunities. Conversely, criminals may slip through undetected if the LLM fails to detect adverse reputation accurately, potentially exposing the company to regulatory penalties.
AI agents built on LLMs inherit both their cognitive capabilities and inherent randomness, making their deployment a double-edged sword. For corporate enterprises with strict policies and procedures, careful consideration is necessary to balance the benefits of LLMs against the risks they pose.
Many organisations deploy AI tactically, relying on employees to repeatedly and manually collect, enter, validate, and normalise data, as well as execute AI models, which can be unknowingly altered. These manual processes are not only time-consuming and costly, they are also prone to human error. This is not an efficient use of AI.
To fully harness AI’s potential, organisations should adopt a strategic approach to deployment, as such:
1. Automate data processing and model execution
Data required by AI models should be automatically extracted, processed, validated and normalised, even when sourced from multiple systems. Automation enhances data accuracy and timeliness, enabling AI models to function optimally without delays or errors.
2. Integrate customised AI models into operations
AI models should encapsulate the company’s unique “organisational DNA” to deliver the most accurate results. By customising AI models to align with specific business objectives, risks, processes and data, and integrating them seamlessly into the organisation’s operations, the company can drive better business outcomes, such as to enhance customer experiences and improve efficiency.
3. Ensure effective remediation of AI results
The value of AI lies not only in predicting accurate insights but also in acting on them. Results from AI models should be properly tracked and managed, with appropriate actions taken to address the issues identified. Effective remediation ensures that underlying problems are resolved and the company derives maximum benefit from AI.
The proper deployment of AI and GenAI can give companies a significant competitive edge over their competitors. Leaders play a critical role in helping management understand AI and avoid costly missteps. Here’s how:
1. Advocate for tailored models
2. Understand and manage GenAI risks
3. Use the right tool for the task
AI has the groundbreaking potential to drive significant business growth and operational efficiency. Organisations that effectively harness the benefits of AI can gain competitive advantage over those that are slow to embrace it. By fostering a deeper understanding of AI, business leaders can help their organisations leverage this transformative technology to achieve their business objectives effectively while mitigating the associated risks.
Lem Chin Kok, FCA (Singapore), ISCA FFP, is CEO of AiRTS Pte Ltd.
© 2024 Institute of Singapore Chartered Accountants
