Interestingly enough- currently at my work we are running an innovation challenge in which teams need to submit ideas to for uses of AI that can improve our finance teams and help them become more efficient. This brings me back to the conversation we had in class (also discussed in the book) of the best ways to encourage teams to be more digital is to allow them to explore AI (and other digital transformations on their own) about learn from them and how to best use. Similar to in class when we talked about using the method of a “hack-a-thon”.
So I decided to kill two birds with one stone and write about AI in FP&A and the different use cases and trends we predict to see in the future. The first being in forecasting.
Today in order to build accurate forecasts and plans FP&A teams at my company take a bottom up approach. Working directly with markets to roll up a country forecast which feeds a regional forecast that then leads into a business unit forecast which then flows into a total company view. Through this process many teams are making different decisions and making assumptions on the inputs.
Many companies have begun to use machine learning in place of this process in which needs to utilize at least 4 cycles of data (4 years to understand seasonality trends). Results are accurate for the next month or quarter- the longer this is forecasted the less accurate forecasts become. For an income statement this type of forecasting can be very accurate this is because expense accounts are easier to predict than revenue accounts. Expense accounts- since they follow similar trends and can be planned & predicted easier can be estimated at 95% accuracy using AI. On the other hand revenue accounts are 85%-90% accurate because there is more fluctuation year over year. A few different models can be used to do this:
- “Regression: Predictive model which investigates the linear relation between output and time.
- SARIMAX: SARIMAX forecasts future values based on past values, that is, lags and the lagged forecast errors.
- LightBGM: LightGBM is a gradient boosting framework that uses tree-based learning algorithms.
- Recurrent Neural networks (RNN): Recurrent neural networks (RNNs) are a class of neural networks well known for processing of time-series data and other sequential data.”
This is on interesting idea I am brining up to my team since we currently do a bottoms up forecast each month that a AI could be used to save time and resources to focus on other tasks. Another benefit of using this type of simulation is that you can easily adjust drivers and get results instantaneously- maybe you want to add a driver to look at your plans or update an existing one. For humans using spreadsheets this would take a lot of time and manual work where using an AI model this can be done and analyzed in a matter of seconds.
Another use of AI in FP&A is data evaluation:
Today at my company we are currently working to be able to access more data- this means being able to access information from our local operations on data such as claims and distribution. To do this we are currently building out a data lake that feeds from various local systems to provide information on specific claims, policy holders and determine the trends driving these in each month. This currently includes a lot of manual testing and back and forth with different teams to understand the data. I think this would also be a good use case for AI. Data validation is something that can be done with machine learning. If an algorithm is built- machine learning can be utilized to quickly sort through millions of data sets to identify outliers and data items that do not match between systems. This will save a lot of time for people who are currently testing and matching each data set in excel. This will help turn the raw data into clear, meaningful data to help business users understand the next actionable setps. Using AI for this can also help identify patterns where this data might be unmatched that might be hard for the human eye to see- either due to input error or the way a system is intaking the data that is creating a miss in correlation.
The last item I explored for AI use in FP&A in through Business Insights:
While doing some research on this topic I stumble upon another insurance company Pacific Specialty that is currently using Avanade (an AI analytics tool created by Microsoft and Accenture) to drive business insights. This system has the ability to take in data from many different platforms on policy holders / policy information to identify trends and insights into different types of policy holders. This can help drive growth within a company and form better plans based on insights gained from the system. I went over to the Avanade website and saw that many other insurance companies are also using this system to drive real time insights and make business decisions.
I’d love if anyone has anything to add on how AI can help FP&A and if your company has implemented it in some way that is making processes more efficient & accurate!