Overview

An AI skill is a design object that enables you to build, configure, and train an artificial intelligence (AI) model using Appian low-code capabilities.

This page discusses the basic features and functionality of the AI skill design object.

Create an AI skill

To create an AI skill:

1. In the Build view, click NEW > AI Skill.

   

2. Select the skill type you want to create.

   

3. Configure the following properties:

   Property	Description
   Name	Enter a name that follows the recommended naming standard.
   Description	(Optional) Enter a brief description of the AI skill.

4. Click CREATE.
5. On the Review AI Skill Security window, configure security.
6. Click SAVE. The AI skill opens in a new dialog or window.

Skill types

As you create an AI skill, you'll select the type of skill that suits your purpose.

• Document Classification: Identify a document's type based on certain traits. You’ll tell us about the types of documents you want to classify in your business process and provide samples of each type. For example, this skill can help you classify and route invoices, purchase orders, and receipts that you receive through one channel.
• Email Classification: Identify an email's type based on certain traits. You’ll tell us about the types of emails you want to classify in your business process and provide samples of each type. For example, this skill can help you classify and route inquiries, feedback, and help requests that you receive through one inbox.
• Document Extraction: Extract data from a document. You’ll tell us about the structure of the type of document you want to extract data from, such as invoices. For example, this skill can help you extract data from fields commonly found in certain documents. For invoices, this might be the invoice number, due date, and total.

You'll configure your skill based on its type.

Design guidance

The AI skill design object is not versioned. However, as you train your skill, you can review the results and create a new model using additional sample data. During this refinement process, each model is preserved in the training summary.

You can access the training summary to view details on model creation history and how the skill was refined to its current state.

Using an AI skill in a process

After you train the skill and determine a model's predictions are suitable for your purposes, you'll publish the model. The published model is what the corresponding smart service uses when it calls the AI skill in a process model.

You'll call an AI skill in a process model using the smart services in the AI Skill menu of the palette:

• Classify Documents
• Classify Emails
• Extract from Document

Delete

Deleting an AI skill prevents users from further viewing or editing it or the models it contains. Additionally, any smart service configured to use the AI skill will no longer work as intended.

AI skills can be deleted by users with Administrator permissions to it.

To delete an AI skill:

1. In your application, go to the Build view.
2. Select the checkbox of the AI skill you want to delete.
3. Click Delete.
4. Review any object dependents and click DELETE.

Security

Actions	Administrator	Editor	Viewer	Deny
Execute the skill	Yes	Yes	Yes	No
View the skill	Yes	Yes	Yes	No
Update the skill	Yes	Yes	No	No
Save the skill	Yes	Yes	No	No
View the security	Yes	Yes	No	No
Create a model in the skill	Yes	Yes	No	No
Publish a model in the skill	Yes	Yes	No	No
Configure the smart service to use the skill	Yes	Yes	No	No
Update the security	Yes	No	No	No
Delete the skill	Yes	No	No	No

Deployment behavior

You can deploy the AI skill alongside other design objects in your Appian application. This section describes what is deployed with the object and what isn't.

	Deployed	Not Deployed
Document extraction AI skill	Document structure Learned mappings from reconciliation tasks	N/A
Document classification AI skill	Document types Published model Training metrics for published model	Documents used to train the model Unpublished models
Email classification AI skill	Email types Published model Training metrics for published model	Emails used to train the model Unpublished models