Machine Learning is far from a new concept. As early as 1959, Arthur Samuel’s self-training checkers algorithm had already achieved “amateur status”, a remarkable achievement for its time. This article aims to clarify the two main types of Machine Learning you’re likely to encounter, especially if you’re considering exploring this space but aren’t sure which approach best suits your business.
The Two Types of Machine Learning
At a high level, Machine Learning can be divided into two categories. Although the terminology may vary, they are most commonly known as static or batch models, and dynamic, incremental, or self-training models. These approaches differ in several ways, which we’ll explore below.
Machine Learning Type 1: Batch Models
Batch models are trained manually and updated at specific intervals, usually offline. This manual process means they generally take longer to develop. These models depend on historical data and are retrained periodically, rather than continuously adapting in real time.
A good example of this is a churn or attrition model, which estimates the likelihood of a customer leaving for a competitor. The amount of effort and time invested in developing these models varies depending on the business context and how long the model is expected to stay relevant and accurate.
In fields such as credit risk, for example, predictive scorecards must perform reliably well into the future. As a result, a significant amount of time is invested to ensure that the input variables (or features) and their patterns are stable and dependable. Depending on the complexity of the problem and the availability of data, developing a batch model can take anywhere from a few hours to several months.
Machine Learning Type 2: Incremental Models
Incremental models, on the other hand, are designed to retrain themselves continuously or semi-continuously with minimal manual involvement. These “online” models are ideal for environments where the data changes frequently and new patterns emerge rapidly.
A typical example is found in call centres, where the data, such as customer behaviour or agent performance, shifts regularly. Setting up an incremental model requires determining the extent to which recent data should influence the model’s evolving parameters. In environments where trends change quickly, the model must heavily weight new data and virtually disregard older patterns. Conversely, in more stable environments, models perform best when they apply only slight adjustments based on recent data.
In a call centre environment, for example, a “Right Time to Call” model can use live dialler data to adapt and continuously improve. In such cases, data from 12 months ago may be far less relevant than the data from the past few days.
What makes incremental models more complex is the need for an automated data feedback loop, which allows the model to retrain itself. Additionally, automated monitoring systems must be in place to ensure newly trained models are robust enough before they replace existing ones. However, once this infrastructure is in place, applying the same system to other areas of the business becomes more straightforward.
Choosing the Right Type of Machine Learning for Your Business
Some businesses find the idea of a fully or semi-automated retraining system particularly appealing. In the context of a call centre, for example, models can help identify which agent should handle a specific customer interaction or determine the optimal time to place a call. These models can be developed relatively quickly, and the automated retraining process can usually be implemented with confidence, especially in areas where the cost of a poor prediction is low.
However, in other areas where mistakes come with greater consequences, a fully automated approach is less advisable. Consider a model used to approve or decline loan applications, or one that determines appropriate credit limits. In these cases, errors can cost your business millions, making reliability and stability far more important than speed or automation.
For high-impact applications like these, more time and effort must be invested in developing models that utilise stable input features and consistent patterns. This approach produces models that are not only more trustworthy but also more resilient over time. If your business already has experience building static batch models, you’re well-positioned to extend those capabilities into more dynamic, self-training environments and unlock the full potential of Machine Learning across other parts of the organisation.
Not Sure Which Type of Machine Learning Is Right for You?
We’ve spent almost two decades building high-performing Machine Learning models for the financial services sector. Our work spans retail, insurance, and telecommunications, encompassing a wide range of use cases, including predicting customer churn, identifying new customer acquisition opportunities, forecasting missed payments, and targeting cross-sell or upsell potential.
If you’re unsure which Machine Learning approach is right for your business, get in touch. Let us know what you’re trying to achieve, and we’ll be happy to help you explore the best way forward.