Deadline analysis may become a basis for important notifications that can predict if the deadline will be missed even unfinished real-time timelines.
Let's imagine the created project already has some actual data displaying how processes can develop in various situations and what time is required for steps to finish. The program allows emulating this behavior assuming various parameters from timelines. Consequently, while uploading real-time data including chunks of process instances that may end tomorrow or in a week, the program will check those timelines in progress and make an assumption if the deadline will be missed and send a notification.
How to build predictive alerts
Using a trained prediction model, you gain an opportunity to forecast with some accuracy whether the deadline will be missed for unfinished real-time process instances and build an alert notifying about these cases. That means that every time you upload a new chunk of data to the project, the program will scan timelines and send a prediction if a timeline is likely to break the deadline even if the process instance is still in progress.
The procedure for creating predictive alerts can be divided into 2 general steps:
-
Prepare and train a model to become a basis for a future predictive alert.
- Make sure you have enough project data for training.
The project should include at least several thousands of timelines obtaining both starting and expected events, defined by the deadline.
- Configure a deadline and save it using the upper-left menu.
- Train a model.
Click Start button in the Training section of the deadline window. After some time, the program will build a resulting predictive model. You can check the processing status by clicking 
at the bottom-left corner. Once the model is trained, it shows a diagram with different accuracies for validation sets. The accuracy shows how well the model was able to learn from the data provided. As a final accuracy, the highest number will be chosen.
-
Configure an alert using deadline definition.
In this part, you create an alert and make some configuration adjustments to base it on the obtained model.
- Create a basic alert using instructions from this article: How to Create and Execute Alert.
- Select to base alert on Deadline and choose pre-saved deadline configuration with performed training.
- Choose Send alert on prediction and specify Threshold value. It means, that relying on the selected trained model in N% of cases prediction should be correct. At the same time, the higher Threshold you determine, the more data about the process should appear. Thus, the program will wait for more data to upload and get closer reach to a possible deadline.
- Select and specify a condition to notify recipients about violations. This is needed because Timeline builds the alert based on the model and not the configured deadline itself.
Use one of the following options:
- Deadline is defined by: Time interval
Along with this, in the Alert recipients section enter the time interval for the deadline. If the expected event doesn't appear in this time gap after the starting one, the deadline will be missed, and an alert will be triggered.
- Deadline is defined by: Cutoff time
Along with this, in the Alert recipients section enter the time by which the deadline should be reached. For example, you may assume the expected event to occur before 2:00 PM, otherwise, the deadline will be missed, and an alert will be triggered.