How to use machine learning to make smarter decisions with your data

AI automation can help finance professionals make the most of their time, diverting effort from administrative work toward meaningful tasks. But realizing that potential requires using the right tools for the right tasks, which is easier said than done.

GenAI, which includes popular tools like OpenAI’s ChatGPT and Anthropic’s Claude, has become extremely popular among businesses and consumers alike. But fewer people understand the fundamental science behind GenAI: machine learning.

This article will break down how to implement machine learning so you can leverage your data more effectively.

Machine learning vs. generative AI

First, it’s key to understand the difference between these two concepts. Machine learning (ML) and generative AI (GenAI) are both forms of artificial intelligence, but they serve different purposes in business.

Christian Wattig, a veteran finance professional who now heads up the FP&A certificate program at the Wharton School, gave an overview of the key differences in our recent AI webinar.

ML is the science that teaches computing systems to recognize patterns in historical data. It uses this data to make predictions or decisions without being explicitly programmed. The system “learns” from the data, and its accuracy improves as it processes more data.

The power of ML lies in its ability to process large, complex datasets and surface insights that guide smarter, faster business decisions. For example, when an ML model analyzes your financial documents, it can identify subtle patterns like seasonal spending cycles, customer payment behaviors, and vendor terms that a human analyst might overlook. You can use ML to:

• Create more accurate forecasts for revenue, costs, and liquidity.
• Automatically detect anomalies and flag unusual transactions or compliance risks.
• Reconcile accounts faster by categorizing transactions automatically.

GenAI, on the other hand, uses ML to create new content (text, images, code, formulas, and calculations) based on the patterns it has learned. For example, GenAI can produce spreadsheets and presentations that visualize the forecast your ML model predicted.

Ways to implement ML in finance

In the webinar, Wattig shares how he used ML to develop a working capital forecast at Unilever. Here’s the guidance he provided.

The first step is to decide how you’ll create your ML model. There are three primary ways to do this:

1. Build your own ML model. This is the best option if you have a capable in-house data science team. It will allow you full IP ownership and control of the model. You’d work with them directly to build and train the model using an enterprise ML solution like Google Vertex, Microsoft Azure, or AWS Sagemaker.
2. Hire a third-party data science provider. Outsourcing ML implementation to a third party will help you get hands-on, customized support. This is likely the most expensive option, but if you have a large or complex dataset, it’s helpful to have dedicated data science expertise. This is the route that Wattig took at Unilever.
3. Purchase software with pre-built, “off-the-shelf” ML models. Commercial ML SaaS solutions like Datarobot and Datarails allow you to train an algorithm on your company’s data, as well as public econometric datasets, without a data scientist. This will almost always be the least expensive option, but if you go in this direction, it’s prudent to understand core ML concepts first.

5 ML concepts you need to know

1. Backtesting: In ML, backtesting refers to testing a model on historical data to evaluate how well it would have performed in the past. This is especially common in finance (e.g., trading algorithms) but applies broadly anywhere predictions matter. If the model performs well in backtests, it suggests potential usefulness in real-world scenarios—but it’s not a guarantee. Beware of “data snooping bias:” if you tune too much to past data, you may fool yourself into thinking the model is better than it really is.
2. Feature selection: Feature selection is the process of choosing which input variables (features) should be used to train the model. In practice, not all available data points improve performance — some may add noise, redundancy, or even harm predictions. Good feature selection reduces complexity, prevents overfitting (see #3 in this list), and often improves interpretability. Techniques range from domain expertise (e.g., “customer tenure is more useful than signup date”) to algorithmic approaches.
3. Overfitting: Overfitting happens when a model learns the noise in the training data rather than the underlying pattern. The model may achieve near-perfect accuracy on training data but fail on new, unseen data. This is like memorizing answers to past exam questions without learning the subject. Overfitting is one of the biggest challenges in ML, especially with flexible models. Common remedies include regularization, cross-validation, simplifying the model, and adding more diverse data.
4. Statistical significance: While ML emphasizes predictive performance, statistical significance helps assess whether observed patterns are real or just random chance. For example, if adding a new feature seems to improve accuracy by 1%, is that a meaningful improvement or just noise? In ML, this ties closely to robust validation: ensuring results hold up across multiple random samples or datasets, not just one lucky split.
5. Correlations vs. causation: ML models are excellent at finding patterns that suggest two values are correlated, but that doesn’t mean one causes the other. For example, an ML model might find that customers who buy sunscreen also buy bottled water, but sunscreen doesn’t cause water purchases. Mistaking correlation for causation can lead to misleading conclusions and poor business decisions. Causal inference methods, experiments (like A/B testing), or domain expertise are needed to go beyond “what is associated” to “what drives change.”

With a better understanding of machine learning, your team can take your data to a whole new level.

To learn more about AI, watch the complete webinar and explore our Applied AI in Finance series.