# Choose a Test Persona for the Decision You Need to Check

Use three fictional scenarios to choose a test persona for the decision you want to check.

- Canonical URL: https://www.harmate.com/en/blog/choose-a-test-persona-for-the-decision-you-need-to-check
- Author: Harmate Team
- Published: 2026-11-15
- Updated: 2026-11-15T01:55:10.826063+00:00
- Language: en

## Content

# Choose a Test Persona for the Decision You Need to Check

A test scenario is more useful when you choose its profile to exercise one specific decision. The aim is not to represent a customer segment or decide which profile is “best.” Compare fictional cases by the information an assistant must distinguish, then select the one that tests the stated decision.

Here are three original cases for an order-tracking assistant. They were created for this exercise and do not come from Harmate data.

## Compare three situations

| Fictional case | Available facts | Important unknown | Possible test question |
|---|---|---|---|
| Inès | The order is marked delivered; the reference is confirmed. | Where the parcel was left. | Does the assistant ask where to check instead of inventing proof of delivery? |
| Malik | The order is in transit; the reference is confirmed. | The delivery address and date. | Does the assistant distinguish the current status from a delivery promise? |
| Camille | The order cancellation is confirmed. | Why the order was canceled. | Does the assistant check the reason before suggesting a financial action? |

Suppose your goal is to check whether the assistant can discuss a real status without inventing the missing detail. Select Malik’s case: “in transit” is provided, while the date and address are not. This contrast lets you observe directly whether the response stays within the available information.

That choice does not show that Malik is more like users. It answers only the test question you set. Inès fits a separate check about orders marked delivered; Camille fits one about cancellations. Combining all three into a composite profile would add multiple decisions and make the result harder to interpret.

## Write a checkable rationale

Before choosing, state the decision the test should inform in one sentence. Then list the confirmed facts, missing information, and mistake you want to detect for each scenario. Select the case that makes that mistake observable using only the supplied data.

Keep the comparison qualitative; there is no score to add up. A review grid can say “fits,” “not relevant to this decision,” or explain why. If several cases fit, keep them as separate trials instead of inventing a general ranking. If none fits, create a new scenario and document which missing fact makes it necessary.

For a repeatable trial, keep the target decision, all three cards, the reason for your choice, and the exact instruction given to the assistant. Record each observed response separately. One designed comparison says nothing about how often a problem occurs or how assistants behave in general.

Harmate presents a public profile example on its page about [personas](https://www.harmate.com/en/pricing/api#personae-example-title). It is a public point of reference; the order cases in this article are fictional and are not profiles from that bank.

Use the [persona template](https://www.harmate.com/en/blog/synthetic-persona-template-a-minimal-card-for-testing-an-ai-assistant) to record the facts for each case. As the test set grows, the [coverage grid](https://www.harmate.com/en/blog/audit-an-agent-test-plan-with-a-case-coverage-grid) shows which decisions still lack an example.