Predictive analytics uses data to make predictions or solve clearly defined analytical problems. It can forecast outcomes, detect anomalies, classify information, and support recommendations. Its value is not only a prediction: it can turn large amounts of data into structure and patterns that people would struggle to find unaided. section 3.2.1 section 4.0
Start with the job, not the most prominent AI tool. Predictive AI analyzes data; generative AI creates content; agentic AI carries out multistep workflows and may use predictive models. When the question is narrow—such as assigning a sentiment category to a review—a small predictive model may be a better fit than a general-purpose generative model. section 3.2.1 section 3.4 section 3.0
Before choosing a predictive approach, check the practical fit:
- Define the outcome clearly. Classification, regression, forecasting, anomaly detection, and recommendations address different questions. section 3.2.1
- Check the data. Many predictive applications need labeled examples, either already available or annotated for the task. section 3.2.1
- Judge the product trade-off as well as accuracy: cost, speed, hosting, privacy, and whether the resulting insight improves decisions or the customer experience. section 3.0 section 4.0
Choosing a model for review sentiment
A movie-streaming platform wants better ratings and recommendations. Sending every review to GPT-4o with a lengthy few-shot prompt produces a popular sentiment feature, but it is expensive and raises concerns about sending individual data to a cloud model. For this narrower task, a logistic-regression model is presented as a small, fast, fairly accurate alternative that can run on the company’s own infrastructure. The lesson is to compare solutions against the actual analytical task and operating constraints, rather than assume a larger model is necessary. section 3.0
