AI classification (coming soon)
Coming soon. Configurable AI classification is planned for end-user journeys. A service builder will define the available categories and describe what belongs in each one, so supplied content can be classified for a particular service purpose.
This page describes the intended approach and gives illustrative configuration examples. The final Builder controls and runtime contract will be documented when the feature is released.
Define classification buckets
The core configuration is a set of buckets with a name and a description. The description should explain inclusion criteria and distinguish the bucket from its neighbours. “Other” or “Needs review” can provide an intentional destination for content that does not clearly fit.
For a facilities-reporting service, the builder might define:
| Bucket | Example description |
|---|---|
| Heating and ventilation | Problems with room temperature, radiators, air conditioning or ventilation. |
| Water and plumbing | Leaks, blocked drains, taps, toilets or loss of water supply. |
| Access and doors | Broken doors, locks, access cards or entry mechanisms. |
| Needs review | Reports that are unclear, describe several unrelated issues or do not fit the other categories. |
These are examples of builder-defined categories, not built-in Govform classifications. Avoid labels with overlapping meanings unless the intended handling is explicit.
Choose the source and result
The intended pattern is to select the content to classify, provide the bucket descriptions and use the resulting category in the journey. For example, a problem description could produce a suggested category for the respondent to confirm or a derived value for later workflow.
Decide whether your service needs one category or several, whether the person can change the result, and what should happen when the input is empty, ambiguous or unavailable. These are design requirements to resolve before configuring a classification step; supported options will be confirmed at release.
Use classification in the journey
A facilities service might collect a description, suggest “Water and plumbing”, let the person confirm it, and then use the confirmed value to select follow-up questions or route a notification. Preserve the original description so staff can assess the report in context.
If routing depends on an AI-generated category, provide a route for uncertain or failed results. Do not make an unresolved classification prevent someone from submitting a request. Keep exact eligibility rules in ordinary conditions and validation.
Prepare examples for testing
- Clear examples belonging to each bucket.
- Reports spanning more than one category.
- Short, vague, misspelled and unfamiliar wording.
- Empty content and descriptions outside the service's scope.
- Cases where the suggested category needs correction.
Review the results with the people who will handle the submissions. Update bucket descriptions when recurring examples expose overlap or a missing category. Define the handling for unsuccessful processing before relying on classification for routing.
For another planned runtime feature, see AI summarisation. For capabilities you can use in Builder now, see AI capabilities overview.