General guidelines for deploying an AI workforce

AI workers, like other agentic AI, benefit from strategic planning for testing, deploying, and monitoring performance. Learn guiding principles to help your AI workers achieve their highest chances of success.

Overview of deploying an AI workforce

AI workers use artificial intelligence to perform work, but they don't require extensive knowledge about AI to set up and use. The possible skills and tasks available to an AI worker are configured by Now Assist applications. You can choose which ones apply, test to verify that the AI worker works as you expect, and then activate.

Like all AI, strategic planning for AI workforce implementation can be key to getting the most out of your AI workers.

Questions to ask before you begin activating AI workers might be:

• What problems do I want my AI worker to solve, and why?
• Do I need to use an AI worker to solve these problems, or would they be better solved with a Now Assist skill, AI agent, or agentic workflow?
• What counts as success in terms of AI worker effectiveness, efficiency, and customer satisfaction?
• How do I want my team members to interact with an AI worker?

When configuring an AI worker, there are several important aspects to consider.

Within the AI worker's profile, the sections each relate to different aspects of the AI worker's suitability for certain work:

Skills: what the AI worker can do.

Skills are the domains where your AI worker is best able to solve problems. Skills are pre-selected by the Now Assist application that the AI worker is installed from.

Assignment groups: the work the AI worker is assigned.

The AI worker can't work on tasks not assigned to those assignment groups. If you don't want the AI worker to work on certain tasks, ensure that the correct assignment groups are selected.

Roles: what data the AI worker has access to.

See the following section Data access security for AI workers.

Languages: what languages the AI worker can use.

If the Communicate updates task is enabled, the AI worker can communicate to users using those languages. AI workers won't work on records that are in languages other than the languages specified.

The AI worker's tasks specify the actions that it can take to solve problems within the skill domains. Each task has different options to specify the actions further. For more information about the different tasks available, see Edit the tasks of an AI worker.

Data access security for AI workers

AI workers only have access to the data that you specify. AI workers can be assigned roles to limit their data access to certain scopes. When configuring the profile of an AI worker, verify that the roles granted are enough to allow the AI worker to complete its tasks, but not enough to access data it shouldn't. The roles function the same as the roles for a human user.

Assigning work to the AI worker

Tasks must be assigned to the AI worker. You can assign tasks to the AI worker like any other user. Manual assignment, business rules or other triggers, and Advanced Work Assignment (AWA) all work. AWA allows you to select rules for assigning work so that work can be assigned automatically when the record is created or updated.

The User record associated with the AI worker can be found on its profile. You can use business rules, triggers, or AWA to assign work to the AI worker referencing that User record.

Previewing your AI worker

Previewing your AI worker on multiple records can show you what to expect when the AI worker is activated and running.

Wichtig:

Changes made by the AI worker during previews are actually made to the record.

• Use representative tasks for initial validation. Selecting tasks that mirror the ones the AI worker will handle enable you to check basic functions. You could also create dummy tasks that simulate active ones.
  Hinweis:

  The AI worker won't send any communications to users during tests.
• Test across different incident and case categories. Testing multiple categories allows you to confirm that the AI worker is working effectively in multiple domains.
• Run at least 5 to 10 tests before activating. Ensure that the tests cover all skills assigned to the AI worker. For example, if your AI worker has the Password Reset and VPN issues skills, you can test 2 password reset incidents and 3 VPN issue incidents to get a general idea of AI worker performance.

Improving AI worker performance with feedback

AI workers can make mistakes. With feedback or changes to the sources they access, you can improve their performance next time.

• Service Desk agents, Service Desk managers, and AI administrators can view the in-progress or completed work of an AI worker on a record in the form view in the Service Operations Workspace or the Core UI. All of the individual steps that the AI worker took to reach its conclusions and make changes are available for review.
• AI administrators and Service Desk managers can view an AI worker's activity in AI Agent Studio, you can see the different tasks that your AI worker has worked on. In the list, there is the option to provide granular feedback of either Helpful or Not helpful. You can open the record, check the AI worker's work, and then submit the feedback in the activity list.
• If the AI worker makes errors based on Knowledge articles that contain outdated or faulty information, Knowledge administrators can update the articles to improve AI worker performance.