As CFOs scramble to meet the evolving demands of their CEO (see my previous post), the introduction of AI in finance is sometimes lauded as the silver bullet that CFOs need to accelerate their transition into the digital world. Despite unprecedented levels of enthusiasm, however, 85% of Big Data, Advanced Analytics and AI projects fail.
One reason for the high failure rate is that AI evangelists tend to present a sugar-coated and hype-laden view of the technology to executives. Promoters of AI reference "self-driving cars" as proof of how advanced the technology is. Surely something that is smart enough to drive an SUV through peak hour traffic or beat the world chess champion can predict company revenue?
Read on to learn:
According to Wolters Kluwer, AI will grant superpowers to a CFO by providing them with a "Crystal Ball" that gazes into the future of their companies financial fortunes. Really?
I get it; sometimes you need a simple analogy to help contextualise an otherwise complicated topic like the use of AI in finance. The problem, however, is that some people then go on to believe the crystal ball story.
Oversimplifying any topic leads to solving the wrong problems and assuming outcomes that were never possible.
Some finance executives believe you can pour raw data into an "AI black box" and have something meaningful come out the other end. Unfortunately, as amazing as that would be (particularly for the company offering said "black box"), we are a long way out from having AI that can do such a thing.
Most CFOs will encounter this oversimplification when talking about AI in finance. When combined with hype, it can be easy to invest in projects that will not only fail but also divert limited resources from your longer-term automation vision.
The oversimplification of AI in finance is partly due to people assuming that AI is smarter than it is.
However, Google's AI beat the Go world champion? A six-year-old couldn't do that! This is where things start to get complicated for anyone who opts for the oversimplification route.
AlphaGo (the project that beat the Go world champion), while impressive, is an example of "Narrow AI". Narrow AI focuses on teaching a machine to do one thing (or a small subset of cognitive abilities) well. Now for some important points:
Important point #1: Virtually every AI product marketed to you in the present day is Narrow AI. That includes everything from Siri to self-driving cars as well as enterprise technology such as predictive analytics, chatbots and "Intelligent Automation" robotics.
Important point #2: An advancement in one Narrow AI solution does NOT automatically mean all AI solutions are suddenly better. No matter what that pushy salesperson tells you, the fact that cars now drive themselves does not mean you are going to be left behind if you don't immediately leverage intelligent automation in your finance department.
Important point #3: Creating a Narrow AI solution is often a tedious and challenging task. In addition to technical and problem-domain expertise, solutions require access to vast amounts of reliable and easy-to-label data and deterministic outcomes that can be modeled and are cheap (computationally) to simulate.
Important point #4: When people talk about robots taking over our jobs or "black boxes" that can ingest data and automatically make sense of it, they are referring to Artificial General Intelligence (AGI). The idea is that AGI can successfully perform any intellectual task that a human being can. AGI is a long way from being a reality.
There is plenty of room for AI in finance departments whether it be in the use of Machine Learning Models, Natural Language Processing or Generation, Predictive Analytics, Computer Vision or other branches of AI.
However, given that AI won't be replacing staff anytime soon, it needs to take its place as a component of the automation tool-belt. That is, a technology that augments finance staff so they can spend more time away from their desk, meeting with and communicating insights to the business:
For those that do consider AI as a component of their solution, be prepared. AI is not a crystal ball, a magic "black box" or sci-fi robot. It takes an enormous amount of collaboration and commitment between data and business experts, vast amounts of clean, credible data and a clear sense of the problem to successfully create a solution.
Most importantly, CFOs should not be leading conversations with "how can we use AI?". Instead, they need to start with "what is the problem we are trying to solve and why?". For many CFOs, the priority this year is to focus more of their resources on analysis and insights (see previous post).
Within most finance departments, staff members spend 25% of their day in Excel, collecting, processing and analysing data. That's probably the right place to start. Large companies (with $1b+ in revenue) have around 70 staff in their finance department - that equates to 17+ people doing nothing but massaging data in Excel all year round. Imagine if you could claw back half of that time and re-allocate effectively a team of 9 people to do nothing but analysis and communication?
There is substance behind the hype, and AI has made tremendous strides in recent years. That said, the scenario of a "robot" that can automate and replace a finance staff member won't exist for some time. The Artificial General Intelligence (AGI) needed for that to happen is - at the time of writing - an incredibly well-funded science project.
Narrow AI, focusing on doing one specific task well, can add value as a component of automation for a finance department. As with automation, CFOs need to view this as an augmentation to their staff allowing them to be more productive and spend more time on analysis and communicating with the business.
Final thoughts for those anxious about AI:
If you'd like to learn more about the key challenges for any CFO that is looking to automate, read our whitepaper here.