How can the profession use AI?

There’s a lot of talk (and hype) around Artificial Intelligence (AI) and machine learning at the moment, but there are clearly some very interesting opportunities for the profession. So, we held a workshop with Portsmouth University this week which brought together machine learning experts and accountants, especially those in audit and forensics, to explore the issues. My colleague Rick Payne from the Finance and Management Faculty and I were joined by Dr Edward Smart, a machine learning expert from Portsmouth’s Institute of Industrial Research, and Prof Lisa Jack, from Accounting and Financial Management, to lead discussions.

The initial focus was on novelty detection although it turned into a wide-ranging discussion. Here are some of my key learnings:

  • Machine learning is powerful and could potentially have many applications across the profession. For example, its predictive capabilities provide opportunities for accountants to be far more proactive, rather than just responding to problems that arise. Using machine learning to build models of ‘normal’ processes could provide powerful new insights around abnormalities. The ability to extract meaningful information from unstructured data, such as contracts and emails, could massively improve quality, speed and efficiency of processes.
  • We need to understand its limits and not just apply AI blindly. The development and effective use of machine learning requires a lot of care – ensuring the right questions are asked, picking the best technical methods, understanding the data, ensuring the outputs are interpreted correctly and identifying when there are errors or significant changes in circumstances that reduce the accuracy of algorithms. These all require expert input.
  • Data is always a big challenge – quantity and quality. Machine learning is a data technology – it finds patterns in data, and clusters, categorises and predicts based on data models. This all depends on having lots of data (although how much will obviously depend on the case). It also needs good quality data - ‘healthy samples’ that the model can learn from. And as with all discussions about data, the realities of most organisations – with lots of legacy and unintegrated systems - means that a lot of work needs to be done on the data before it is usable.
  • There are also issues to consider around the economics and business case for investing in AI. Developing bespoke models is still a fairly expensive business, and defining the business case may not be easy. For example, while using predictive models to become more proactive seems sensible, it can be hard to quantify the benefits and secure investment. Similarly, when looking at models for new insights or patterns, you are unlikely to know what you will find (and therefore the specific benefits) in advance.

There were also many other questions raised about the practical application and longer-term implications. To what extent do models really learn and how do they cope with changing human behaviour? How will we gain assurance around the use of ‘black box’ algorithms in businesses? What about the future skills and roles of accountants? And is there space for collaboration across the profession here, or will it be a purely competitive world?

We are developing a short paper summarising the opportunities and challenges for the profession from AI, so if you’re interested in contributing, please get in touch!