The AI, ML, and RPA conundrum with a Fintech flavor

AI in Fintech

From the start of 2021 to mid-2022, I led the marketing efforts for a company called Noesys Software. In the time I was there, I had a first-hand view of what Machine Learning (ML) was. While Artificial Intelligence (AI) had been used a whole lot in our marketing discussions, ML was what our management said would disrupt the industry.

As I sit down to write this reflection paper, I realize that I have had the best ringside view of how ML models are built. Moreover, I have also seen first-hand what Machine Learning can do to disrupt the payments industry. While certain jobs would remain, I believe that ML and most importantly, Robotic Process Automation (RPA) would disrupt the way financial institutions operate. In turn, this would also disrupt fintech.

The AI and ML industry today

Despite having roots in algorithms developed several years ago, both AI and ML have become popular today. The rise of platforms such as ChatGPT and Bard has seen an increased awareness of the industry. Several investors ranging from venture capital to private equity and investment bankers are all looking at AI and ML as investment opportunities. Therefore, despite being in its nascent stages, AI and ML are rapidly growing industry sectors.

Before moving forward, it is necessary to understand the way AI and ML work from an industrial understanding standpoint. For me, the concept of AI originates from training bots to mimic human actions. Of course, not all humans are the same, but AI should be able to mimic most actions. For instance, bank tellers at ATMs are humans performing repetitive tasks related to customer support. This can be easily accomplished by an AI bot. The banking industry is highly susceptible to automation and stands to be replaced and disrupted by AI bots.

Machine Learning on the other hand is a subset of AI. While AI is used to train the first bot, ML can be used to repeat the process and scale operations. Let us take the same instance of the ATM tellers at the bank. While AI can automate one teller, ML can replicate this automation across the board. Think of AI as the head of Hydra from Captain America. If one head is cut, a thousand more would take its place — the thousand referring to Machine Learning algorithms.

The rapid growth of the industry can be ascertained by the following statistics:

1. From a global standpoint, the global AI in fintech market is expected to reach USD 4.16 billion by the end of 2030.
2. The CAGR (Compounded Annual Growth Rate) for the above will be 16.5% (source: Grand View research)
3. McKinsey estimates that 70% of executives believe that AI and ML will automate the industry going forward.
4. AI and ML will automate service and menial tasks that are prone to human error and can be easily automated. This has led to the development of the RPA technology within Machine Learning.

AI and ML use in fintech

1. Fraud detection in transactions — Research shows that despite all efforts, cybersecurity is an issue with several fraudulent transactions happening daily. An AI bot will be able to single out fraudulent transactions and hand them over to necessary agencies for further investigation. Given time, the AI will also be able to implement its governance and regulations against fraudulent transactions.
2. Investment research — Bard and ChatGPT are being used for research purposes by several academics and working professionals today. Trained ML models can be used to conduct research on data collected and derive insights.
3. Customer service — This is an important use case for most companies as their biggest wage costs come from customer service. However, today, most companies have managed to automate this by deploying chatbots. This also enhances customer interaction and has proved to enhance retention.
4. Risk assessment — This is connected to the research portion that AI and ML can undertake in terms of financial instruments to invest in. It can also be used to value the risk of portfolios should there be a requirement to do so.

Specific Use Case

Coming back to my story of marketing AI and ML products. While it was not the ideal opportunity I was looking for at the time (and the company also did not seem too impressive as well), I decided to use the opportunity as a learning experience. However, as I began devising methods on how to best market the product, I came across certain ML models built for clients and how RPA (Robotic Process Automation) can be used.

Working for a client in Australia, the following were elements of the problem statement we tried to solve:

1. The client was in the energy industry and was the fastest-growing energy provider in the country. This meant that they were supplying utilities (cooking gas and electricity) to 700,000 households across the country and growing the numbers by almost 50 to 60% each month. Important note — the population of Australia is roughly 26 million.
2. The company wanted to incentivize customers who had paid early. They instituted a program that gave discounts to customers who paid before the due date. Moreover, they also wanted to penalize the people who paid late by adding fees for those who paid after the due date.
3. While this improved business systems, several clients had issues with discounts. At times, due to the volume of data, managing ‘Big Data’ became a challenge. Certain customers who were eligible for the discount were not given the discount due to a lack of remittance processes within the company.
4. The Chief Strategy Officer eventually decided that it was time to introduce Machine Learning tools to ensure customer satisfaction while also bringing onboard new clientele.

From the outset, this is a simple case. Ideally, the client would have been able to solve this issue by hiring more staff for the remittance team. However, using our company’s RPA technology, the following was achieved:

1. An RPA bot was deployed to bring together a database of customers who had paid early.
2. Depending on how early they had paid, the RPA bot determined how much discount they should be receiving. Conversely, the bot also determined how much additional payment needs to be made in case of past-due payments.
3. The RPA was also able to do this at 100% accuracy, something that could never be achieved by humans.
4. Moreover, the RPA can also complete this remittance in a few minutes (or hours) depending on the dataset’s size. This was the biggest value ad for the Australian client, something that allowed them to attract more customers, save costs, reduce human errors, and keep their existing customers satisfied.
5. In effect, the RPA tool was able to complete menial tasks that would have otherwise taken a human a few days to accomplish.

Challenges with the usage of AI and ML

While the world is gung-ho about the use of AI and ML, there are several challenges while attempting to implement this technology in today’s landscape.

1. Data quality — It all comes down to data quality. Should data quality be poor, the results of the AI tool used would also be poor. This is a counterproductive task
2. Big data bias — The data collected will at some point inevitably lead to bias. However, it is up to the company to decide at which point they should stop relying on AI and have human intervention.
3. Culture fit — Not all organizations can benefit from deploying AI. Moreover, if the senior management does not buy into AI or any data strategy for that matter, it will never see the light of day in that company.
4. No one-size-fits-all solution — There will never be one AI bot that can solve everything. Even ChatGPT and Bard have several limitations.

Future use recommendations

After looking at the pros and cons of using AI and other supplemental technologies, I have come to one immutable conclusion. These technologies are far from being perfect and there needs to be human intervention at some point. While that threshold is yet to be determined, companies would have to decide upon the breaking point of when to stop using AI and use humans instead. This is something for companies to ponder over.