Take the next step with machine learning for predictive analytics in the CFO Office
Modern Enterprise Performance Management platforms provide embedded functionality for predictive analytics through artificial intelligence and machine learning. This makes the technology easily accessible. In the CFO Office, predictive analytics can be applied to recognise and exploit patterns in historical and transactional data to guide decision-making. This is an opportunity for the CFO Office to provide insights across the organisation to drive positive change and increase financial health at the same time. Moreover, such an initiative is an excellent growth opportunity for financial and data savvy talents or even to attract such talents. Unlocking this potential in your organisation can be done via a project-based proof of concept.??
Define at least three use cases with both a financial and business impact to demonstrate and visualise the hidden potential of your data assets. Sufficient data and data quality are necessary conditions to train your predictive model through machine learning. That’s why operational financial data is relatively easy to use, as automated validations are used to secure data quality. Nevertheless, an interesting first use case is to use machine learning in data collection to improve data quality. A second use case is to find hidden patterns, anomalies and outliers to learn from the past. The third use case is to create driver-based simulations.?
1. Improve data quality?
The first use case is about creating the foundation for all business-oriented use cases. To apply machine learning successfully to large data sets, it is important to bring both financial and operational data together. Validated data has to be available in a structured manner. Data from different sources and with varying quality needs has to be brought together and refreshed periodically. How can machine learning help to improve the acquisition and the quality of the data? Established ETL-Tools provide functions like recoding, filtering or pivoting. These are typical inaccuracies which cannot be handled with conventional ETL:?
Outliers have to be detected to train the software based on every field provided from the data source. When sophisticated enterprise performance management software is used for planning, consolidation and reporting processes, validations are used to check the quality of the data. Mostly, this is at an aggregated level for the periodic consolidation. Now with more granular financial and operational data, different validations are needed to make sure the data is correct (meaning registration of transactions is done correctly). Machine learning should be applied to train the software on outlier detection and to automatically apply translation rules to address data to the right spot in the data model needed for predictive analytics?
2. Detect hidden patterns, anomalies and outliers?
Modern advanced analytical software use out-of-the-box statistical models to detect hidden patterns, anomalies and outliers automatically. These models are trained when new data sets are added periodically and provide new insights instantaneously. There is no need to create hypotheses upfront and validate if your assumptions are right. Outcomes are by itself proposed to you and you need to interpret these outcomes and see if these are aligned with your business objectives, strategy or sustainability targets individually and combined. If you want to dive deeper, suggestions about anomalies and outliers are automatically proposed for further automated and human investigation.?
This sizeable and internal data is a good starting point for companies with significant overhead, purchases and/or production costs and revenue from services, premiums or products. Cost controlling can be improved by:??
领英推荐
The sales and revenue forecast will be more accurate and reliable, resulting in:?
3. Create driver-based simulations?
A sales forecast provides a static view of the future and depends on facts and subjective interpretation by the forecaster and receiver of the forecast. Decisions are made and, in hindsight, evaluated on accuracy and reliability. Driver-based simulation assessments can be done beforehand on possible effects and provide insights into whether estimates are too high or too low. Also, it is important to understand dependencies for cause and effect when, e.g. prices of raw materials go up, causing higher sales prices and lower sales volumes, resulting in lower production. What has been discovered in the previous use case 2. Detecting hidden patterns, anomalies and outliers can be used to enrich simulation models. This can be further enriched by external drivers, e.g. a country’s birth rate, demographical factors of the population, inflation rate, economic growth, etc., to evaluate upcoming capital expenditures. The outcomes can be used to improve the management capabilities for accurate decision making.?
Start with a real-life business experiment to gain experience and create awareness about the potential of machine learning for predictive analytics, which is often technically already available in your company. Bringing together the right people to explore this potential with the guiding principles for what to achieve should be a fun exercise as well.?
To learn more about how inlumi can support you with machine learning, take a look at this case study with Autoliv.?
Consulting Director at inlumi?