Categorize data with Stackby AI

Applying an AI field template

Use a ready-made categorization template to get consistent tags in minutes. Pick the source field (e.g., “Feedback text”), paste a comma-separated list of allowed labels, and optionally add 1–3 examples to guide edge cases. Enable auto-run so new or edited records categorize themselves.

Examples of label sets:

• Support: Bug report, Feature request, Billing, Login issue, Outage

• Product: Pricing, UX, Performance, Docs, Integration

• Research: Positive, Neutral, Negative (sentiment)

Adjusting AI field settings

• Generation: Turn on “auto-generate” to recategorize whenever the source field changes; keep a manual override view for audits.

• Model/quality: Favor accuracy-optimized models for production; use faster options during setup and testing.

• Randomness: Set to 0 for deterministic outputs—ideal for categorical tasks where “creative” answers aren’t desired.

Using AI field with custom prompts

Switch from the quick template to a freeform prompt when the workflow needs more control. Reference multiple inputs, define label rules, and specify single vs multiple selections.

• Multiple inputs: “Read {Feedback} plus {Product area} to pick one label from [Pricing, UX, Performance, Docs, Integration]. Return only the label.”

• Described labels: Add definitions to reduce ambiguity (e.g., “Performance = speed, crashes, memory; UX = navigation, layout”).

• Multi-tag mode: Allow 1–3 labels and require a short rationale for audit trails.

Using AI in Automations

Trigger AI categorization when records are created, edited, or enter a view, then route by outcome.

• On create: Categorize feedback and set Severity if label is “Outage” or “Billing.”

• On update: If label changes to “Feature request,” create a linked Idea; if “Bug report,” open a ticket.

• Weekly digest: Summarize label counts and top examples, post to the product channel for triage.

Try Now Stackby AI

• Start simple: Add an AI field to a Feedback table, choose Categorization template, set Content = {Feedback}, and labels = “Pricing, UX, Performance, Docs, Integration.” Run on 10 records to validate.

• Add control: Convert to custom prompt, include label definitions, and set randomness = 0. Test multi-tag output and add a “Rationale” text field.

• Operationalize: Turn on auto-generate, create a “Needs review” view for “Other/Low confidence,” and wire automations to create tasks or tickets based on labels.