Applies to version: 2019.1.4.x and above; author: Anna Puka
In WEBCON BPS 2020, two new standard reports – Anomalies and Suggested are available in the “Insights” section, which allow you to quickly find important information in your processes.
The Suggested report shows instances that can be the most important for the user. These instances are designated based on the date of the last modification and the modifying person. In the report we find all instances which have been assigned to the user and have been modified by the user. The shorter time since the last modification on a given instance, the higher it will appear in the report. The Suggested report is automatically created and does not require any additional configuration.
The Anomalies report shows atypical instances in processes for example: an order for an excessive amount compared to other orders, or vacation requests spanning a very long period time. Launching analysis of the anomalies will automatically create a process description – a set of AI rules based on real data entered by users.
A set of AI rules is created by using artificial intelligence. A random sample of data is selected for the process, based on current and historical instance values (for each workflow instance, form type and step), and then rules are created which encompass the most common (covering over 75% of cases) values or ranges of values for all numerical and categorical fields.
Moreover, dependencies between various fields on the form are created, and norms based on statistics about editing a given form field (whether the field should be completed or can be modified) are also designated. After creating a set of rules for the process, each instance registered in it is verified to what extent it matches the designated norms. Based on this match, the instance is classified as more or less typical on the anomalies report.
The resulting set of AI rules creates a description of current data entered in the process and thus can also be used to verify the assumptions in the implementation of the workflow (e.g. additional validation of the entered values). With appropriate configuration, we can define which of the detected rules are more important for us when determining atypical instances, as well as define what their further processing should look like.
Launching the AI analysis
The process of configuration consists of a few steps:
- Enabling AI analysis in website roles (similar to OCR option, the AI option is available after activating the AI Framework license).
- Selecting the processes to be included in the AI analysis
How to do this? In each process, in the “Settings” tab select the “Include in AI analysis” option.
- Creating a set of rules for the processes
In the “Report” -> “Basic reports” section select “AI analysis”, then expand the “Add” button and select the “Build AI model and perform analysis” option. This option will allow you to create sets of rules for the processes and determine the atypical instances (anomalies).
You can also:
- Clean AI analysis – cleans the rules created so far and the anomalies report on the WEBCON BPS Portal
- Build AI model – creates a set of rules for each processes/steps included in the AI analysis
- Perform AI analysis – updates the anomalies report and determine the atypical instances for each of the analyzed processes based on the currently existing sets of rules
- Selecting the processes for which AI analysis is to be performed
The process selection window for which we want to perform the AI analysis will appear on the screen. All processes with the “Include in AI analysis” option selected will be available on the list. We have also the option of indicating a specific data range from which the instances for analysis will be collected, and indicating the priority of analysis for each of the processes.
After saving the changes, the AI analysis will be performed.
Important! The minimum of 500 instances in the workflow are required (current or historical) to perform the analysis. If the processes do not have enough data, the analysis will not be performed. We recommend building this model on a process with minimum of 1000 instances and values dating back at least one year.
Results of the analysis
After building the model, the obtained set of rules will be visible on each of the analyzed processes on the “AI rules” tab.
For each type of form, workflow, and for each step, a set of rules is built. Additionally, a set of rules for the entire workflow (for all steps simultaneously) will be created.
The rules are designated for all: dates, choice fields, integer numbers, floating-point numbers, checkboxes (yes/no) and number of versions (modifications) of the instance. The designated rules defining:
- The most common values or range of values for the field
- The difference and relation between values in the form fields
- The fields that are not usually modified in a selected step
- Typical number of versions for a selected step
- The fields that should not be empty in a selected step
Parallel to the set of rules, the Anomalies report will appear on the WEBCON BPS Portal in the “Insights” section of the navigation panel.
On the report, instances will be visible if they are considered atypical according to the defined rules. For each of them, two parameters will be designated:
- Standards matched – percentage score of the designated set of rules fulfilled by an instance
- Rules violated – number of rules an instance does not meet
In the upper menu the symbol will appear in the form of these instances which allow you to check the results of the analysis.
The resulting set of rules is the basis for determining anomalies in the processes. There are several ways to modify the AI analysis to better suit our process:
A) Changing the rank of the rule
For each of the generated rule, we can define how important or relevant it is supposed to be. Depending on the rank, they will have a greater or lesser impact on the norm execution level of the instance.
After changing the ranks in order to update information on the Anomalies report, from the level of “AI analysis” report, run the “Perform AI analysis ” option in WEBCON BPS Designer Studio.
B) Changing the analyzed form fields
For each of the analyzed form fields we can define “AI analysis mode” on the “Style and behavior” tab:
- Ignore – a form field will not be included during the analysis
- Per step – default option, the rules for each form field will be designated both in individual steps and for the whole process
- Per instance – only general rules for each form field will be designated
After modification you must re-activate the “Build AI model and perform analysis” option to get a new set of rules in the process.
C) Changing a date range of the analysis
Example of uses
The obtained results can be used to manage the atypical instances in the process. Based on the generated rules you can for example: display additional warning message/information for the users, show additional path or automatically redirect the instance for additional approval.
Three new business rules have been added with the introduction of the AI analysis:
- AI ANALYSIS RULE COUNT FOR INSTANCE – indicates the current number of rules for a given instance
- AI ANALYSIS BROKEN RULE COUNT FOR INSTANCE – indicates the number of rules that a given instance does not meet
- AI ANALYSIS SCORE FOR INSTANCE – calculates the norm compliance level for a given element
Below is an example of a simplified invoice approval workflow.
Using the AI analysis you can redirect instances that contain atypical values to additional approval (e.g. with a different currency than usual).
How to do it? Just add a technical field on the form,
Just add a technical field on the form, will enter the level of deviation from the norm of the instance. For this purpose, you can either use the current compliance with the instances norm or the number of unfulfilled rules. In the example above, the following action has been added to the Approve path from the Approval step:
Next on the “Workflow control” step a simple rule has been added which checks the number of broken rules, and depending on the result, can direct the instance for additional approval or straight to the archive step.
Similarly, you can build an inverse scenario to the one mentioned above, where you build a rule based around the AI analysis score (e.g. only instances whose AI analysis score > X won’t need approval).
The AI analysis mechanism in processes helps you easily catch instances that have been potentially incorrectly filled out, or non-standard instances that may require additional attention or verification.