Ir al contenido

Acceptance criteria

De Scrum Manager BoK
Esta página es una versión traducida de la página Criterios de aceptación. La traducción está completa al 100 %.
⏱ 5 min  ·  📅 2026
Acceptance criteria define the minimum requirements a user story must meet to be considered complete and acceptable to the customer. They are the explicit agreement between the Product Owner and the team on what it means for a story to "work correctly." Unlike the Definition of Done, which applies equally to all stories, acceptance criteria are specific to each story.

Characteristics

  • They are usually presented in list or scenario format.
  • They can be written in natural language: what matters is that they clearly reflect the intent behind each criterion, not that they follow a formal syntax.
  • Although the Product Owner may have prepared them in advance, they are finalized as a team during refinement or Sprint Planning. The conversation this process generates is as valuable as the written result.
  • It is the Product Owner's responsibility to ensure that the criteria cover all the aspects the team needs to estimate and develop the story properly.
  • In the Sprint Review, the Product Owner checks whether each story meets the defined criteria in order to accept it or send it back.

How they are written

Natural language

The simplest form: a list of conditions that the system must meet.

Example for the story "As a customer, I want to search for products by name":

The search field is visible on the home page.

When a name is typed and "Search" is clicked, a list of matching products is displayed.

Each result shows at least the title, the price, and an image.

If there are no results, a clear message is shown to the user.

BDD / Gherkin format

BDD (Behavior-Driven Development) is an approach that centers development on understanding the expected behavior of the system. Gherkin is the language it uses to describe scenarios in a way that is readable by both the technical team and stakeholders.

The structure is: 'Given' (context) / When (action) / Then (expected result).

Story: As a customer, I want to perform an advanced product search so that I can easily find what I'm looking for.

Scenario 1: Search by name

Given a customer has accessed the store.

When they type the name of a product in the search field and click "Search".

Then they see a list of products matching the name entered, with title, price, and image.

Scenario 2: Search by category

Given a customer has accessed the store.

When they select a category from the drop-down and click "Search".

Then they see a list of products in that category, with title, price, and image.

Acceptance Criteria for AI Stories

When a story involves AI-generated output (text, images, recommendations, classifications), traditional acceptance criteria are not enough. The team needs to explicitly define:

  • Output quality criteria: when is the result "good enough"? For example: "The generated summary does not exceed 150 words and covers the key points of the original document."
  • Human oversight criteria: "Every AI-generated response has been reviewed by a human agent before being shown to the user."
  • Edge case behavior criteria: "If the model cannot generate a response with sufficient confidence, it displays a handoff message instead of an incorrect response."
  • Privacy criteria: "No personal user data is sent to the external model without explicit consent."

Without acceptance criteria specific to AI behavior, the team has no way of knowing when a story with an AI component is truly done. "The model responds well" is not an acceptance criterion: it is a subjective impression.

Common mistake

Confusing acceptance criteria with test criteria. The criteria describe the observable conditions that will show the need has been met; test cases are the technical verification that those conditions are fulfilled. The criteria emerge from the conversation between whoever has the need, whoever builds the solution, and whoever verifies it, with the Product Owner accountable for making sure they are clear; attributing their writing to a single role turns the conversation into a handoff.

When the behavior does not allow for a single correct answer, as with features based on language models, the criterion stops being binary: a rubric of what makes an answer good is defined, along with a set of evaluation cases and a threshold over that set.

Resources

🏦 Skill User storiesState of the art and development guide · Skill Arena

📄 Scrum Master en equipos con IA v.1.0 [Scrum Master in Teams with AI v.1.0, in Spanish]Free download · Scrum Manager

📄 Criterios de aceptación: ejemplos para elaborarlos [Acceptance Criteria: Examples for Writing Them, in Spanish]Scrum Manager Blog · Mar 2023

See also

Want to grow in agility? You can search for course and exam calls or go at your own pace by becoming a member of the Agile Club. This membership includes exclusive resources, e-learning classrooms, and access to Skill Arena: a space to practice and measure your agile skills at your own pace.