By Dorian Wattel
Today we are here with Thymo ter Doest, RSM alumni and co-founder of AccountingBox, to discuss the impact of artificial intelligence and machine learning on the future of business. AccountingBox is an online accounting service that uses machine learning to automate bookkeeping. He started the company six years ago with his father, an experienced bookkeeper, after many long nights discussing the repetitiveness of traditional accounting.
The impact of artificial intelligence and machine learning
Artificial intelligence (AI) is a machine’s capability to mimic "cognitive" functions associated with the human mind. It entails processes proper to human beings such as problem solving and learning, except that it is done through the integration of data as a learning mechanism. Thymo explains that machine learning, a subset of AI, is basically the automation of data processing where the machine learns from data to reproduce human senses. In turn, this results in a sort of mechanisation of actions currently done by human beings. The emergence of this new technology and its normalisation can have radical effects on the business world. The entrepreneur captures this impact by contrasting the steep variation in prices that AI and machine learning could provoke. He goes on: “let’s say a bookkeeper bills a company 20 hours a month for recording business transactions; the same bookkeeper will only need to bill 2 hours on these tasks in the future”. In this case, AI could induce a drastic reshaping of the service sector, automating repetitive tasks usually performed by a human workforce, and thus cutting important costs. The Accounting Box co-founder concludes: “of course, forecasting that AI will divide all prices by ten might be an overstatement, and yet it is not that far off from reality. Expensive human labour with all kinds of physical and psychological limitations can, in some cases even entirely, be replaced by low-priced machines that operate efficiently 24/7”.
“a bookkeeper bills a company 20 hours a month for recording business transactions; the same bookkeeper will only need to bill 2 hours on these tasks in the future”
It is obvious that there are numerous efficiency and cost advantages in automating processes. Nevertheless, are there concerns coupled to these benefits?
Thymo ter Doest: “Applying AI to process automation is the keystone that accelerates progress in all kinds of fields. New software applications and autonomous robotics are increasingly assisting and even replacing humans. According to a McKinsey study, 45% of job activities that workers perform today could already be replaced by machines with today’s proven technology. Low-skilled workers could end up unemployed while high-skilled workers would profit lavishly. This is a valid concern, but it can be overcome by adjusting governmental policies and with appropriately regulating the implementation of machine learning. Additionally, if our education system prepares future workers to be flexible and adapt easily to new technologies, there should be no apprehension of massive unemployment.”
In other words, it is crucial that the educational and legislative systems keep up with the current speed of technological change. This is necessary in order to avoid negative social and economic outcomes while grasping the true benefits that come with AI and machine learning.
Finally, an overarching concern that remains the subject of intense debate is the ethical dimension of applying this new aspect of technology in some delicate settings. Thymo states that “although extremely reliable, machine learning will never be 100% infallible. In the case of self-driving cars this poses many questions.” Indeed, if an accident occurs and leads to serious injuries or even casualties, the problem of responsibility is posed. There is a point where AI is more efficient than human cognition in anticipating unexpected movements, hence some already existing early breaking technologies. Yet, if the cars rely entirely on AI and a tragedy happens to occur, it is debatable as to how we should deal with such a situation.
With this in mind, is more always better when it comes to AI?
The enthusiastic entrepreneur suggests that AI is not necessarily essential in every industry or company. However, this holds not because of the latter concerns but more because in some instances, human heuristics and predictable programmable rules are amply sufficient. In the case of AccountingBox, most of the time progress is not a tech-related problem but a business acumen related issue. “An illustration of this is the fact that the digital infrastructure of many companies relies on Amazon S3 which is an interface whose technology dates back to over 10 years ago”, describes Thymo. Another reason why more is not always better in the case of AI is its complexity. He explains that “when something goes wrong, if there is a malfunction for example, it is difficult to understand where the issue lies. This is different from programming where bottlenecks can be reasoned about logically and are therefore easier to identify”.
So, what advantages are brought to the table?
Thymo ter Doest: “Computer applications digitised and automated manual business processes in a broad range of industries. These applications were only able to automate structured and rule-based tasks, and to a high degree, humans still needed to operate them for control and interpretations. Artificial intelligence solves these limitations, as the algorithms are increasingly able to automate cognitive-intensive jobs that previously were unthinkable to be done by machines. In a diverse range of tasks, these algorithms are often able to outperform humans in terms of speed, scale and accuracy. Therefore, machines using such algorithms can be more cost-efficient than human workers. Some tasks could be automated using simple heuristics, while others require more sophisticated algorithms. For example, in the case of the start-up, the tasks of converting invoice documents to booking proposals and matching bank transactions to corresponding invoices were automated using machine learning principles. It is satisfying to see how the applications of Information Extraction techniques are so influential in reducing the total workload of our bookkeepers: it has helped process over 100,000 transactions at AccountingBox.”
"programmers automate the work of humans while machine learning automates the work of programmers”
AI and machine learning go the extra mile when compared to programming. In a way, programming is a patchwork where you reassemble and encapsulate existing pieces, meaning that you are limited to the extent of these existing pieces. This induces numerous valuable applications but also implies a certain level of complexity constraints. As Thymo captured it perfectly: “programming automates humans while machine learning automates programmers”. With an increase in complexity, there is a need for more cognitive power which is what artificial intelligence brings to the table. When in need to process a near to infinite number of data points, the complexity of the task is no longer an insuperable challenge. Of course, the results may not be the most accurate at first, but the more AI is put to work, the more accurate the results will be.
In conclusion, what should managers do to successfully integrate the use of these technologies?
Thymo ter Doest: “For the past five years, I have witnessed how the applications of AI positively affect the role of bookkeepers. At my startup, AccountingBox, we developed numerous automation systems that significantly improved the satisfaction and productivity of bookkeepers. Instead of spending the majority of their time on manual data entry, they are now able to focus on the more valuable tasks of advising entrepreneurs. The job of a bookkeeper has not been replaced but has evolved to better leverage the unique capabilities of humans.”
The entrepreneur highlights the importance of facilitating change by ‘hiding’ the complexity of technology for both the team and the consumer, meaning you have to limit what is immediately visible. Intuitiveness is key and AI should be camouflaged within the end product that the consumer interacts with or the tools that employees use. In the eyes of the end user, the benefits ripped from the implementation of AI must be outlined.
The most critical point is balancing automation and control. AI should assist managers, not take over their role. Always keep in mind it is a synergetic collaboration between humans and technology.