The FP&A Trends Webinar: Mastering Analytical Transformation with FP&A Trends Maturity Model
Click here to view details and register
The FP&A Trends Webinar: Mastering Analytical Transformation with FP&A Trends Maturity Model
Click here to view details and register
By Asif Khan, Global FP&A at PayU
The democratization of technologies is underway. Tools like machine learning (ML), which were confined to universities, hedge funds or investment banks just until a decade ago, are now finding their way into industry-wide applications. The finance function is set to reap the benefits of this democratization wave.
In PayU's finance function, we decided to test ML’s effectiveness in forecasting the financials in 2017.
Below I share our thought process about WHY and HOW we went about implementing ML-based models for forecasting.
The first step of the path towards implementing ML in finance function was to understand the WHY? As Simon Sinek puts it in his hugely viral video “Start with Why”.
Before starting with ML-based forecasting, we were doing bottom-up forecasting on a monthly basis. To arrive at a forecast, finance teams of different regions/countries had discussions with commercial teams, performed pipeline analysis and seasonality checks, etc.
However, the results of the forecast were suboptimal due to four key factors:
Hence, our WHY to implement ML-based models was to improve the accuracy levels and speed of forecasting while reducing the time needed to come to forecast.
We took a very agile approach to ML model implementation in our company and did not go all in on ML with a lot of resources. The idea was to come with prototype versions of ML models and test if there is some real merit in using these models. The criterion of success of the prototype was very simple: Can the prototype models forecast better than bottom-up process?
Step 1. Forming a small team. Having engineers in finance team helped on the journey to a prototype. We also sought help from tech team members on a part-time basis. Since machine learning was a new area for us all, we took self-learning classes from some of the fantastic courses available on leading online learning platforms.
Step 2. Data preparation. We had data sitting in silos. Connecting all the data sources, cleaning it, removing the outliers and then making a useable and clean data repository was a crucial step.
Step 3. Implementation. Finally, we managed to start implementing ML algorithms on the processed data sets. We divided the data into 2 sets: (i) Training data: to train the ML models, and (ii) Test data: To test the trained models on actual historical data. This helped us to increase our confidence in the predictive power of the models.
Results: It took us around 3 months to have a prototype version of the ML models. This timeframe really depends on the complexity of each individual companies, and the purpose of usage.
Figure 1: Architecture supporting ML Project for forecasting
In our case, using ML models for forecasting lived up to the expectations.
We used these models to forecast revenue, gross margin and EBITDA. These models also provided several benefits across accuracy, speed and flexibility:
Our initial foray into using ML capabilities for forecasting has been successful. ML models are data hungry, so we have continued to enrich our models and are still testing them to get better accuracy levels. We will also be creating new models for products which were not in scope for the prototype version.
Our experience with implementing ML models for forecasting has encouraged us to also look at other processes within the finance function where ML can be used to full advantage.
The concepts of artificial intelligence (AI) and machine learning (ML) are not new. They are relatively...
In July 2017, I presented at the 2017 AICPA FP&A Conference in Las Vegas over AI...
In this article, the author explores top 3 impacts of AI & ML on FP&A people, processes and...
We will regularly update you on the latest trends and developments in FP&A. Take the opportunity to have articles written by finance thought leaders delivered directly to your inbox; watch compelling webinars; connect with like-minded professionals; and become a part of our global community.