Creating use cases

A use case is the top of a chain: the use case is linked to a reply, and the reply to the flow that handles it.

A use case describes a situation in your own words so the agent recognises it. It is the top of a chain: the use case is linked to a reply, and the reply to the flow that handles it, whether that is a dialogue or a generative procedure. Writing the use case well is what makes the rest possible.

Create one

1. In the AI agents workspace, select your AI agent, click Content in the sidebar and select Use cases.
2. Add one and describe the situation.
3. Give examples of how customers raise it.
4. Under Reply method, choose how it should be handled.
5. Test with phrasings you did not use in the examples.

The agentic capabilities behind these flows, including multi-step procedures and external API calls, are included across Suite and Support plans; they no longer need a separate add-on.

Describe the situation, not the words

"The customer wants to know where their order is" works. A list of exact phrases does not, because customers will use the phrase you did not think of. The examples are there to show the shape, not to be matched.

Use their words in the examples

Take them from real conversations rather than writing what you imagine. Real messages are shorter, vaguer and worse spelled than anything you would invent, and that is what the agent has to cope with.

Keep them apart

Two use cases describing nearly the same situation make the agent choose between them, and it will sometimes choose wrongly. Where two overlap, merge them or sharpen the boundary until it is a distinction you could explain to a new colleague.

One outcome each

A use case that leads to three possible outcomes is three use cases, or one flow with branches. Keeping it to one makes the behaviour predictable and the transcripts readable.

Deciding the reply method

A dialogue gives you a lot of control and needs more setup and maintenance. A generative procedure needs less setup and gives up some control over fine detail. Choose per use case rather than for the whole agent: a refund path and a "where is my order" path do not need the same rigidity.

Start with your top five

The five most common contacts. Get those right, watch the transcripts, then add. Twenty use cases written on day one are twenty things you cannot yet tell are working.

See also

• Creating generative procedures
• Creating a dialogue
• Importing historical data to train the agent