Accounting weekly

What Artificial Intelligence Means and How CFOs Think About It 

We asked an expert to break down what artificial intelligence means and to talk us through the challenges it creates for CFOs.  

What do we mean when we speak about artificial intelligence? 

To understand what AI is, we must understand how technology has developed. Five years ago, a finance department at a large firm typically used different software for different functions. However, these applications didn’t talk to each other, and staff still had to do a lot of retyping and repetitive tasks.   

For instance, well-paid staff may spend hours checking if a supplier listed on the internet matches the company’s approved supplier database and retyping the information on the internet into one of the company’s software platforms. 

Then, along came RPA (Robotic Process Automation). These are programs, sometimes called scripts or macros, which will do that typing or verifying for you.  

“What you are doing with these tools is to create bridges between applications that don’t talk to each other. By doing that, you are safeguarding legacy systems that you have,” says Armand Angeli. He heads the RPA, intelligent automation, shared services and outsourcing committee of the DFCG (the French CFO Institute) and also the digital transformation working group of the CFO Alliance, the newly formed federation of EMEA CFO institutes, including SAIBA.  
Another advantage of RPA’s is accuracy. “Unlike an employee, the robot, when it takes the data from one screen to another, doesn’t make any mistakes if it has been well coded.” 

There is a problem. These RPA’s are ‘hard coded’. Every so often, exceptions occur that even the best coders couldn’t predict and build into the coding. “If you try to do a task with the robot, outside the standard activities, then the robot doesn’t work,” says Angeli.  

 This brings us to the current generation of software developments, collectively called AI technology, but which consists of several individual software fields.  

Firstly you have ‘computer vision’, which is software able to take or read data from images. Previously software struggled to understand and obtain meaningful data from images. This is why websites ask you to identify objects in photos to prove you’re human.  

Another technology is referred to as ‘natural language processing’. “It can go into an email and take out the information that you need and transform this unstructured information into structured information that can be dealt with by a robot,” explains Angeli.  

 The most important new technology is ‘document understanding’ says Angeli. “It’s a tool that can look at very complex documents or complex invoices and extract data,” 

  Lastly, you get ‘process mining’. This is software that analyses entire systems and processes and figures out how they can be improved upon.  

  These technologies are collectively referred to as artificial intelligence, and require large amounts of data to train the software to identify patterns through thousands of repetitions. This learning makes it possible for AI technologies to handle exceptions because it has ‘seen’ similar examples and solutions in the data and can guess how to deal with an exception with a high probability of success. This also requires a lot of computing power.  

CFOs are faced with difficult choices and no easy answers  

While artificial intelligence technology has amazing potential, CFOs face several practical problems when trying to include them in a finance function.  

The technology stack at a large company is a bit like an old house. It may have recently added rooms or even a second story that is modern and keeps the cold out, but the foundations often consist of old technologies which do the job but not very efficiently. In a business, this foundation is usually in the form of ERP (Enterprise resource planning) software such as Oracle or SAP.  

 CFOs who are unhappy with their legacy software have a few choices. You could wait for a new version of the legacy software, but that may take years. Another option is to try and patch up the holes with RPAs that automate some tasks. “But that’s not the best route. The best route is to look into to upgrade your legacy systems if you can,” says Angeli.  

Upgrading is not without risk. Software development or changes, like building, frequently takes twice as long and is twice the forecast price.  

 For instance, Angeli points out that some CFOs are currently experiencing disappointment at the reality of AI. “They were told that they can cut 80% of their costs. Either they were overpromised, or they didn’t understand what was really going on.” 

AI adoption is about more than direct costs 

“Right now, when you ask finance people or CFOs, they will tell you we are doing all this automation, not to reduce cost, but to improve the customer experience,” says Angeli. The other driving factor is employee happiness.  

  “Employee quality of life is key because right now you can’t hire. So if you have an old type of company, nobody wants to join. You need to show the new employees or the existing employees that you are using new technology.” 

A problem AI implementation faces is that it is expensive and it doesn’t lead to a direct return on investment, as customer and employee satisfaction metrics are hard to quantify. 

“You cannot go to your boss and say, ‘Mr. CEO, I want to spend two million to improve my customer experience and employee satisfaction,’” says Angeli. “But you are going to see the results of not spending within a few years because your employees are going to leave, your customer is going to leave, and you are going to have to shut down your company.”