The rise of the FP&A forecasting machine 🤖! Forecast automation using machine learning enters the mainstream.
One area I’ve been thinking about is how to make planning easier on FP&A – and one of the angles is to use automation to populate the first pass of your budget or plan.
In my book The Nine Principles of Agile Planning, I focus on differentiating between what you can control and letting the technology drive the routine. In a perfect world, you want your Finance team to plan on what will move the needle – e.g., changes to pricing, volume, customer retention, new products, new marketing campaigns, staffing changes, new equipment investments - and so on.
Use the technology to forecast the routine parts of your financial forecast.
New forms of automation and machine learning are coming to our budgeting & forecasting tools, and are being democratized so that anyone can use them. This will allow everyone to automate the routine parts of their forecast and focus on those important initiatives that change the business trend.
Oracle’s EPM Planning recently launched its first big release of democratized machine learning tools that can help forecast your existing run rates… and anyone with basic finance & stats knowledge can set this up by clicking through a wizard.
What did Oracle add?
Beginning with Oracle’s November 2021 (21.11) Oracle EPM Planning release, subscribers automatically received Oracle’s Intelligent Performance Management (IPM). IPM has several features of interest:
Auto-Predict (available in Standard and Enterprise editions)
IPM Insights (only available in the Enterprise edition)
Bring Your Own Machine Learning model (only available in Enterprise Edition)
Auto-predict is where I want to focus today….
Auto-prediction magic!
In a planning application, if you want to generate a forecast based on historical data automatically, take a look at Auto-Predict.
I see two use cases for Auto-Predict:
1) If you have a stable revenue or expense trend in certain parts of your P&L, try the machine algos to generate their prediction, and then have your planners attack the exceptions: e.g., the “new-stuff” – like a new incremental marketing campaign or introduce a new product.
2) Use the algos to forecast underlying driver volumes and pricing rates that feed into your driver-based planning math.
How do I use Auto-Predict?
Oracle EPM Planning Auto-Predict wizard.
Configure auto-predict wizard: specifying the scenario/version where your source data lives, the range of data (e.g., cost-centers, accounts), and the target scenario/version (e.g., scenario=forecast, version=auto-prediction).
Set up data-screening options: adjust outliers and fill in missing values to smooth a dirty data set.
Setup seasonality rules: Determine how many time periods constitute a cycle in a weekly model (probably 52). In a monthly model, it’s probably 12. But, this should be set to the ebb & flow of your business activity.
Oracle EPM Planning’s Auto-Predict wizard allows you to control the sensitivity and tune the algorithm.
Once it’s configured – give it a run.
Auto-predict works best when you have a good base of time-series historical data to inform the future. The manual states, “prediction results are more accurate the more historical data you have. There should be at least twice the amount of historical data as the number of prediction periods.”
Each time auto-prediction runs, Auto Predict generates a summary report that helps you understand the predictions, the methods used, the confidence intervals and if any values were guessed.
Running Auto-Predict in Oracle EPM Planning generates a summary report.
Auto-Predict generates a summary report outlining the guessing process, methodologies, and accuracy.
What are the limitations of Auto-Predict?
One drawback is Auto-Predict is single-variate.
Auto-predict can interrogate and predict a single measure at a time based on historical activity. For planning & forecasting use cases, history can be a good predictor of future activity… at least some of your revenue, cost, or expense lines likely follow historical patterns. As I said before, I am a big proponent of divorcing your routine activity (e.g., run rate) from how you plan to change your business (e.g., initiatives). This can be as simple as letting auto-predict generate a trend-based prediction and then layering on a 10% uplift in sales from a new product introduction.
Since Auto-Predict is single-variate, that’s where the other prediction feature of IPM Insights comes in, “Bring Your Own Machine Learning model.” BYOML allows you to have multi-variate prediction models. This is where things get interesting…!
With BYOML the sky is the limit. You can take data from within the traditional finance planning world and combine it with data from the real-world. As an example, if you are ski area, you could predict customer visits for next month using historical and predicted weather data.
Wrap-Up
If you try Oracle EPM Planning's new Auto Predict in your forecasting process, let me know if it reduces work for your forecasting team.
About David Pabst CPA & Oracle ACE Pro
David Pabst CPA/CITP & Oracle ACE Pro is a thought leader and solutions consultant focused on empowering startups to Fortune-100 orgs to employ the Oracle EPM Cloud to make their organizations nimbler. David recently wrote The Nine Principles of Agile Planning: Create Nimble and Dynamic Forecasting in Your Organization, available at major booksellers.