AI assistance
Describe what you want; the result is verified before it is offered.
You can describe a calculation in plain language and have a formula proposed for you.
How it works
The description, plus the shape of your model, goes to a language model. What comes back is not trusted:
- it is validated against the real function registry,
- every field and measure it references is checked to actually exist,
- types are checked the same way a hand-built formula's are.
Anything that fails is rejected rather than shown. The principle throughout is AI proposes, code verifies — a suggestion that references a column you do not have is a wrong answer, and it is better discarded than displayed.
Whose credits it uses
A generated formula costs one credit; a generated dashboard page costs ten.
On a dashboard you own, they come from your account. On a dashboard shared with you as an editor, they come from the owner's account — the formula lands in their dashboard, so they are the one billed, and you never need your own subscription to help build it. If the owner has run out, you are told so rather than asked to upgrade.
Calculated columns belong to a dataset rather than to a dashboard, so generating one is available to the dataset's owner only.
What it is good at
- Getting the shape of an unfamiliar calculation started.
- Naming things reasonably.
- Turning "revenue per active customer this quarter versus last" into a structure you can then adjust.
What to check
The proposal is a draft. Read what it built before saving — particularly which column it chose when several have similar names, and whether a filter you assumed is actually there. It is quick to check because the result is blocks, not text: you can see the structure at a glance.