AI and Business Today

There has been a wave of AI hype in recent years, with many companies looking to start a project involving AI before even understanding what AI is about. This is not the case anymore, as companies have a good understanding of what AI can and cannot do. Most companies now see what is reality versus what is science fiction, and start to grasp AI’s strengths and weaknesses.

As with any solution or tool, before they jump to using AI, bank executives must understand what it can deliver, its requirements, the time frame involved, and the risks and costs of AI compared to other solutions. That is the starting point.

Research from Lucidworks with 2500 senior executives and entrepreneurs in June 2024, showed that the growth rate of investments and expenses in Generative AI was starting to slow, mainly due to executives realising the cost to benefit ratios. Research from Gartner in May 2024 showed that there was an important gap in estimating and demonstrating business value as the main barrier to the adoption of generative AI.

There are clear signs that the market is starting to understand that AI is another tool/technology. It’s certainly a powerful tool, just like others in the history of humanity, like the steam engine, the production line, robots, and computers. The key here is to follow how AI evolves, understand it, and use it tto maximise the efficiency and return to your business.

Another factor is that companies started to experience the growing cost of having large, updated and curated databases to obtain more precise outputs. However, it is important to highlight that Machine Learning can work with any types of databases if the data is organised in way that it can be read and processed, and many banks have excellent stores of data.

You can for example use an Excel series of spreadsheets to feed into your model using SQL or CSV (comma separated values), and then using algorithms such as Decision Trees, Random Forest, XGBoost and Logistic Regression. You can use these spreadsheets to store and process data needed to analyse credit, finance, health, insurance information and decision, HR etc.

Data and Algorithms

Any data source must enable usage by an algorithm, but again we stress the need for enough high-quality data to deliver the proper outcomes. We begin with the premise that we can process data in an algorithm, and we can have unstructured data elements that use text such as phrases, articles, messages, reviews, etc.

These elements of text are of course everywhere in email processing, customer service chats, and social media with relevant use cases like satisfaction analysis, text classification, and unrelated intents from spam detection. This is known as Natural Language Processing, which is enabled by models such as GPT and BERT (Bidirectional Encoder Representations from Transformers).

The same can be said about images, which are stored in formats like JPEG, PNG, or BMP, and used in scans of documents, photographs, and general images. Applications for these formats include facial recognition, medical diagnostics, and document parsing using methods such as convolutional neural networks (CNN).

Likewise, audio files are also stored in formats like WAV and MP3, and NLP can extract meaning from voice commands, music, or recordings of meetings or lectures. We can use algorithms that can interpret and transcode voice signals, such as recurrent neural networks (RNN) and audio transformers.

We also work with chronologically ordered and labeled data such as stock market records or weather records. (Such chronologically ordered and labeled formats are important in forecasting, anomaly detection, and controlling processes.) In these specific situations, we often use algorithms such as ARIMA, LSTM, and Prophet.

Also, as a company generates more and more complex outputs from larger prompts, the cost of curating, checking, and supervising the output also grows. Many suppliers will charge based on the size of prompts, complexity and level of supervision of outputs, and the cost can grow exponentially.

In many initial projects, these growing costs were not considered and had a negative impact on the estimated ROI of such projects. It is important to have a clear understanding of what is the problem where AI may be a contender for a solution, and a clear understanding of the alternatives. When deciding on an AI solution, have a clear picture of what is expected, the time needed, and its full cost.

Also, as a company generates more and more complex outputs from larger prompts, the cost of curating, checking, and supervising the output also grows. Many suppliers will charge based on the size of prompts, complexity and level of supervision of outputs, and the cost can grow exponentially.

In many initial projects, these growing costs were not considered and had a negative impact on the estimated ROI of such projects. It is important to have a clear understanding of what is the problem where AI may be a contender for a solution, and a clear understanding of the alternatives. When deciding on an AI solution, have a clear picture of what is expected, the time needed, and its full cost.

AI and Business Strategy and Vision

As former HSBC CTO Rumi Contractor said recently, “What 90% of users forget is that neither AI nor any other technical innovation (except a specific stand-alone functionality) is easily implemented and integrated within a business process and a corporate environment. The challenge for companies is NOT what AI strategy they should adopt and use but HOW they can adopt and use that within their corporate ecosystem. This has been a challenge for the past and will continue being a challenge for the future.”

Despite all this AI can be a very powerful tool and can achieve great results. It can create competitive advantages, improve processes, and reduce costs, once the bank or company implementing it truly understands AI and how AI integrates into and impacts the company or bank.

These results can only be achieved if the bank or company has a strategic vision and the leadership necessary engagement. Today most AI initiatives are led by the CTO, or within specific areas such as risk and fraud, customer services, and marketing. However it is the leaders of the company who must understands the role that AI plays in the strategy of the company. AI must have the engagement and support of the Senior Executives and boards. Tactical implementations are fine, but they do not maximise AI’s potential impact.