In the world of predictive analytics, I frequently see the word “real-time” being misused. For some it means seconds, and for others it means hours or even days. By their very nature, some industries require true real-time predictions—with data coming back in seconds or less—while others may only need predictions back in hours.
Let’s look at the most common use cases for real-time predictive analytics:
When predictions are needed in SECONDS…
- Fraudulent transactions. This is the most common scenario where true real-time predictions are essential. In order to prevent fraud, the predictive model must decide if the transaction needs to be accepted or rejected at the exact moment the transaction is happening.
- Dynamic product pricing. When a customer is trying to get information on a product or service, the predictive system must determine the best price at that exact point in time. This information may be dependent on supply and demand, day and time of the week, competitors’ pricing, and sometimes demographics/purchase patterns.
When predictions are needed in MINUTES…
- Store product offers. Once a customer enters a shopping mall or store, the retailer has a small window of opportunity—usually a few minutes—to get their attention. That’s where the predictive system helps out by identifying the best product or service to promote. The same scenario applies to online shopping, though the amount of time people spend on websites is usually even shorter than in brick-and-mortar stores.
- Customer service. Most calls to customer service last several minutes. During that timeframe, the predictive system must provide the support representative with context-sensitive recommendations on what actions to take based on the specific customer request. Ideally, this is an interactive model where the input can be changed based on specific customer interactions, and the predictions will automatically adjust as new data is entered.
When predictions are needed in HOURS or DAYS…
- Predictive maintenance. Machine downtime can cost companies thousands of dollars per hour. That’s why manufacturing shops want and need the ability to predict when a machine will go down so they can schedule predictive maintenance hours or days ahead of time. From a consumer perspective, this is similar to when a car predicts yearly service or maintenance checks and an indicator light shows up on the dashboard.
- Call center staffing. In the event of a big product launch or other event where a company is expecting a surge in call center volume, predictive analytics can be incredibly useful to determine staffing. Once the predictive system determines the number of personnel needed in the coming days or week, the call center can be staffed appropriately to keep wait times at an acceptable minimum.
Whether predictive data is needed in seconds, minutes, hours, or days, the model remains the same. The key is to take historical data and mine it to create appropriate rules. Once you’ve established a model, it can be applied to any data to predict future outcomes, and you can then leverage the results to optimize your actions and decisions.