Navigation: » No topics above this level«
Bizagi Artificial Intelligence - Alpha version
When you have a subscription in Automation Service, you can use a range of Cloud applications provided by Bizagi.
These Cloud applications let you go beyond process execution and use your data to explore fields such as artificial intelligence (predictive analysis), integrate reporting tools of your choice, and create portals to provide a richer user experience.
Artificial Intelligence for Automation Service, is an application which lets you explore machine learning capabilities directly in your Bizagi processes, by letting the processes use a predictive analysis service that is based upon the reliable data that you have in a Bizagi Dataset.
Throug predictive analysis, you can train models and carry out experiments that rely on renowned machine learning algorithms to determine a given outcome based on stored data with a given degree of certainty.
For such models, you can easily set your Bizagi processes to present a prediction once certain data has been provided. You can use that prediction as a default value, or offer it for en usaer approval.
You can configure all of this without being an expert data scientist.
Machine learning algorithms
Employed algorithms include:
•Decision tree (C45)
•Decision tree (ID3)
•Linear SVM Classifier
•Multiple Linear Regression
Bizagi Artificial Intelligence chooses the best algorithm for your specific use case statement, though you can explicitly set which algorithm you want to employ.
For instance, when predicting attributes/variables of categorical type, the Logistic Regression algorithm it is often employed.
Similarly, Multiple Linear Regression would be applied to continuous attributes/variables.
Examples of use case statements could be:
•For a credit application process where a customer requests a loan:
Based on credit amount, customer type and credit type, we could predict whether the process requires additional documentation (i.e predicting a categorical variable).
•For a vehicle insurance underwriting process:
Based on insurance amount, customer type and estimated price, we could predict an estimated duration for which to approve the insurance underwriting (i.e predicting a continuous variable).
•For a help desk support process where a customer submits a ticket:
Based on severity, operating system, and ticket type, we could predict whether the ticket is likely to be resolved by first-level support, or whether it will needs to be escalated to second or to third-level support (i.e predicting a categorical variable).
Before you start
Artificial Intelligence relies on Bizagi Datasets as its data provider.
Before moving on, make sure you are familiar with Bizagi Datasets, and have created a Dataset to use with the Artificial Intelligence application.
For more information about Datasets, refer to Bizagi datasets.
Getting started with Bizagi Artificial Intelligence
The following steps summarize at high level how you work with Artificial Intelligence:
1. Create an AI model and define its data source.
Set it to connect to a Dataset, that holds data.
2. Explore the results of different experiments, and choose a configuration that provides results that satisfy your use case statement.
You can evaluate possible combinations between input parameters and algorithms.
Once the experiment is satisfactory and final, publish it.
3. Connect your Bizagi processes to consume the predictive analysis service so that each new case can make use of a preliminary prediction for an attribute/value.
Configure the invocation to use a RESTful service and that's it.
1. Always create models and experiments based on real data (e.g, by using a Dataset from the production environment).
2. Only publish an experiment when it is satisfactory.
Publishing an experiment allows you to use one service endpoint for testing, and another one for real use (processes' production environment).
For the service endpoint used in testing, you may define fixed rules so that your AI service returns predictable outputs and you can easily work in all of the different processes' paths.
3. Constantly train the model.
By default, new data stored in the Dataset is not considered by an experiment until you explicitly train the model to re-evaluate the data.
When you do this, make sure to re-publish the experiment so that the same service endpoint being used by processes is updated.
You can update an existing experiment which has been published and is under current use, while taking proper precautions.
4. When an experiment in current use, changes its input or output definition, you need to create a new experiment.
We strongly recommend this as a best practice to maintain the stability of ongoing processes.
To learn how to get started with Artificial Intelligence, refer to Creating AI models and experiments.