Using Blue Prism Decision
Blue Prism® Decision provides Blue Prism users with the ability to train and deploy intelligent machine‑learning decision‑making models within their Digital Workforce, without requiring any data science expertise. Blue Prism Decision's simple and intuitive user interface allows users to:
- Define a model – Add a title, description, decision type, and decision variables.
- Train the model – Use active learning based on defined input variables specified by the model service, without needing any data science knowledge to produce statistically accurate machine‑learning models.
- Calibrate the model – Review model inputs and outcome predictions, and overwrite model decisions if required.
This guide outlines the functionality and usage of the Decision plugin available in SS&C | Blue Prism® Hub.
This guide in intended to be used by anyone who wants to use Decision to create a model that can be used in an automation and any users who create processes in Blue Prism that will use these models.
The Decision plugin requires access to the Blue Prism Decision Model Service – the Decision plugin's Machine Learning API. This is available as either a Windows service or packaged as a Linux container. For information on installing the Decision Model Service, see Installing Blue Prism Decision.
Blue Prism Decision is compatible with Blue Prism 6.4 or later.
It is your organization's responsibility to implement Decision in accordance with laws applicable to your organization. For example:
- Your organization may be subject to laws that prohibit or restrict it from making solely automated decisions that have a legal or similarly significant effect on an individual (such as a decision about access to credit or shortlisting for a job). You can build in human involvement in reviewing decisions by building a process with human interaction by adding a human‑in‑the‑loop after the decision is made, in the exact same way they would do now in a process using tools like Blue Prism® Interact, to review the decision if necessary.
- Certain data privacy laws require that an organization using a statistical model to make a decision/prediction about people ensure that the model is sufficiently statistically accurate and avoids discrimination in order for the processing of personal information to be fair. The accuracy of the model is continually updated as the user continues to train the model, and it is solely at the discretion of the user to continue to train a model (by supplying enough samples) to the sufficient accuracy they wish to get to, before utilizing the decision capabilities in a production process.
- To comply with the principle of transparency, certain data privacy laws require that an organization informs people about how it processes personal data in an machine learning system, which includes being able to explain the basis for any decisions. Decision is built on the principles of simplicity and auditability. Once the model is utilized in a process, you can view an audit log that records every input, output, confidence score, and output decision made. This information is available in the prediction list.