There’s been ample excitement about the impact that machine learning and artificial intelligence can have on the accounting and audit industries. There’s also been some fear, especially after two Oxford academics concluded that accountants were highly susceptible to computerisation. People were worried that the robots were coming for their jobs, but thankfully, the truth is never that simple. The great advances that have been made in other complex fields should give accountants hope that it can reinvigorate the industry.
Skills, not robots
Personally, I prefer the term data science, rather than artificial intelligence, because it emphasises the fact that these techniques involve more than picking a R2-D2 off the shelf and letting it loose on your tax returns. Implementing data science at an accounting practice involves the right mix of skills from your staff, easy access to the right tools, and a commitment to overcoming the various data access and shaping hurdles that will be presented. As developing a data science capability is becoming part of many practices’ strategies for 2019, people are figuring out the right ways to overcome these hurdles.
The good news is that tools and techniques that are being honed in the broader data science world are being brought into the accounting and audit space. This is making it cheaper for tools specific for the industry, such as MindBridge.Ai and Brevis to appear on the scene. It’s also making it significantly easier for anyone willing to learn machine learning theory and coding to get involved in the data science revolution.
Built for everyone
In particular, the open source movement is still going strong. This movement is creating high quality building blocks for a huge variety of software applications, free for anyone to use and maintained by the public. Microsoft’s acquisition of Github and their willingness to build their own Linux OS have shown that even a company that once called open source a cancer has come to embrace the movement.
Access to a coding environment is also getting significantly easier. Jupyter notebooks allow people to write Python code directly in their browser, a combination which is becoming popular. The large cloud providers have noticed, and all provide some form of notebooks as a service. These notebooks as a service give users a website to do their data work in, backed by powerful servers with comparatively little set up costs for firms or users. Once a user has performed some data analysis, their web nature means these notebooks are easy to host and share online.
Things are no different in the world of data science and machine learning. Deep learning, a form of machine learning that is vaguely inspired by patterns in biological nervous systems, has become significantly easier to implement due to open source packages such as TensorFlow and PyTorch. Coming originally from Google and Facebook respectively these tools reliably perform much of the heavy lifting for data scientists.
However, to use something like TensorFlow directly the industry must be building the right kinds of skills. It’s often said that the proper data scientist must have a mix of statistical knowledge, computing skills, and domain expertise. People who can comfortably claim to have these skills are rare, which is what must change if we are to benefit from the data science movement. There are only a few places where an accountant can currently learn the foundations of these data skills. I was lucky that the management of Kingston Smith saw the value, and provided the environment for those in the data analytics team to pursue new skills in this space.
It’s becoming easier to pick up these skills, with coding schools geared towards those popping up in cities around the world. If these are too steep, online courses such as those provided by DataQuest guides students through projects, step by step, all in their browser. Alternatively, there are the video course providers like Udemy that are going from strength to strength.
Auditors in particular share many characteristics with data scientists. Audits are at their core a statistical exercise, with auditors drawing conclusions about a population from a sample. The ability to deal with large datasets, particularly in the form of the general ledger, is already becoming the norm at many accounting practises. Other tech-savvy firms realise the importance of data pipelines by connecting web applications together to ensure data is always kept up to date. One look at the app eco-system that is popping up around the big cloud vendors proves the value that this is providing.
Even before building a machine learning model, the techniques that data scientists use to explore data are useful to accountants, advisors, and auditors. Visualisations of data sets such as a density plot can convey an understanding of the data that the pivot table (an accountants’ best friend) cannot. These are building blocks of a data scientist’s skillset, and are useful exercises in their own right.
The Medici Effect
Whilst it’s getting easier to learn and implement data science techniques, many of the processes in accounting and audit are both complex and difficult to model, with professionals in the industry spending years building niche domain knowledge. Likewise in data science, the skills required to create novel methodologies or to handle huge data sets take years to hone.
Because of this, I’m guessing that to pull off this kind of work will require both accountants to learn some data science, and vice versa. The fundamentals are a good start, and there are many use cases where the fundamentals are good enough.
An AI may never be able to fully replicate what the accounting and audit industries do. Data science, however, will be able to improve the quality and consistency of the work being done. In a time where people are questioning the value of accountants and audits, this opportunity couldn’t have come earlier.