Evaluating a business process as an Agentforce use case
Key Takeaway
Evaluate each idea before you build an agent. Work that follows clear rules passes the decision-making test more easily than work that needs a person's judgment at each step.
An Agentforce use case is a way of applying AI where an AI agent takes an action or series of actions to accomplish a goal or job-to-be-done for your employees, customers, or organization. Some business processes suit autonomous AI and some don't, so each idea needs evaluation before you build.
Use case ideas and the evaluation questions
A use case starts as an idea. Ideas can come from these sources:
- Crowdsourcing
- Business process review
- User research
- Market analysis
You evaluate each idea by testing it against questions in these five areas:
- Value: Ask why you are delegating the work to an AI agent. A good candidate has a clear payoff, such as the agent being faster or more accurate, or giving a better experience.
- Work: You should be able to describe the work the agent will do. You should also fully understand all the business processes involved in that work.
- Decision-making: Ask whether the decisions and steps can be completed without direct human input or judgment. Also ask whether there are well-defined policies, rules, and constraints the agent can follow alone. Work that follows clear rules passes this test more easily than work that needs a person's judgment at each step.
- Risk: Ask whether the agent can operate within the security, legal, ethical, and regulatory requirements that apply to the work.
- Data: Ask whether your data can support the work the agent will do. Agents perform best when they run on trustworthy, high-quality data that is relevant to the business context.
For example, imagine a company with a written returns policy that states which orders qualify for a return. An agent can follow that policy alone, so the work passes the decision-making test more easily. Now imagine work with no clear rules, where a manager must use judgment at each step. That work doesn't pass the test as easily.
To answer these questions well, you need to understand what agents can do. Often the easiest way to get started is to build an agent and see how its capabilities match your goals.