The AI Suggestions tab of the Pending Tasks page lists changes that AI agents propose for your catalog — filling in a missing description, classifying a sensitive attribute, adding a relation, or creating a new object. A data steward reviews each one before it reaches the catalog. A proposal only takes effect once you approve it; declining it leaves the catalog unchanged. Pending Tasks also has a separate Concept approvals tab for other pending concepts, which this article does not cover.
Understanding AI suggestions
Dawiso calls a single AI-generated proposal a finding. AI agents — for example, the business term suggester, which reviews new data products — scan your catalog and create a finding when they detect something worth fixing or adding. A finding can carry one proposed change, or bundle several related changes together; the list marks a bundled finding with a change count.
Finding types
Every finding carries a type, shown as a colored label in the Finding type column and at the top of the preview panel, naming the category of issue or opportunity the AI detected:
| Finding type | What it flags |
|---|---|
| Missing description | The target object has no description. |
| Weak description | The target object has a description, but the AI judged it too weak. |
| PII classification | The AI proposes classifying an attribute as personal data. |
| Missing owner | The target object has no owner. |
| Missing parent | The target object has no parent. |
| New relation | The AI proposes relating the target object to an existing object. |
| New object | The AI proposes creating a new object, related to the target object. |
The first five types propose a single change to the target object itself, previewed as described in The proposed change. The last two behave differently.
New relation
A New relation finding proposes relating the target object to an object that already exists in your catalog — for example, linking a data product to an existing business term. It previews the same way as any other relation change: the object it would be related to, shown in the proposed-value row with a plus sign.
New object
A New object finding proposes creating a new object — for example, a new business term — rather than only changing the target object. AI agents such as the business term suggester create this kind of finding while reviewing new data products. The new object is created as a concept related to the target object; it only becomes a real object in your catalog once you approve the finding.
For this finding type, the preview panel shows:
- The new object’s type, as the panel’s heading.
- Its proposed name, marked with a plus sign.
- Its proposed value for one attribute — for example, a definition — when the AI filled one in.
- Related to: the object it will be related to (the finding’s target object).
- Location: the space, and, when the AI also chose a parent, the parent it will be created under.
Open in concept opens the proposed object’s concept so you can inspect it in full, while the finding is pending or declined. Once you approve the finding, the same button reads Open in object and opens the object that was created.
Listing pending findings
The AI Suggestions tab shows only findings that are still pending, sorted by severity by default. Each row shows:
| Column | Shows |
|---|---|
| Severity | How urgent the finding is: critical, warning, or info. |
| Finding type | A colored label naming the finding’s category — see Finding types above. |
| Finding | The finding’s title, plus a change count for a finding that bundles more than one proposed change. |
| Target object | The object the finding applies to. |
When there are pending findings, a strip above the list shows how many fall into each severity level, plus a running total. Every column shown above can be sorted and filtered from its column header.
Click a row to open the finding in the preview panel next to the list. Check the boxes next to one or more rows to act on several findings at once — see Approving or declining a finding.
Previewing a finding
Opening a finding shows its severity, finding type, change count, and status at the top, followed by its title and creation date.
Why AI suggests this
When the AI supplied a rationale, it appears first, under the heading Why AI suggests this.
The proposed change
For a finding with a single proposed change, the panel shows a compact before/after: for a change to an attribute, the object’s current value next to the AI’s proposed one; for a change to a relation or a user relation (such as assigning an owner), only the proposed value, marked with a plus or a minus to show whether it is being added or removed.
For a finding that bundles several changes, the panel instead shows an impact analysis: how many objects and changes are affected, and how many of the changes already have a conflicting value. Below that, a list shows the first five proposed changes, with a note for how many more there are. A link opens the full set on a dedicated review screen, where each change can be selected individually before applying.
A button at the bottom of this section opens the affected object. While the finding is pending or declined, the button is labeled Open in concept and opens the concept holding the AI’s proposed values — a full, isolated view of the change. Once the finding has been applied, the same button is labeled Open in object and opens the object’s live page instead.
Approving or declining a finding
The preview panel’s footer shows Approve and Decline buttons only while the finding is pending. Once the finding is resolved, the footer disappears and the panel becomes a read-only record of what happened.
- Approve applies the finding’s proposed change — or, for a bundled finding, all of its still-pending changes — immediately.
- Decline opens a dialog for an optional reason before rejecting the finding. Nothing in the catalog changes.
To act on several findings at once, check the boxes next to their rows in the list. A bar then appears at the bottom of the screen, showing how many findings are selected. It carries its own Approve selected and Decline selected buttons, plus a button to clear the selection. Declining also accepts an optional reason. If some of the selected findings cannot be applied or declined, they stay selected so you can retry them.
Viewing applied findings
The AI Suggestions tab only ever shows pending findings. Click View applied, next to the list’s title, to open a separate, read-only log of every finding that has already been applied. It shows the same severity, finding type, finding, and target object columns, plus when each finding was applied. Clicking a row opens the same preview panel, without the Approve and Decline buttons, since the finding is no longer pending. Click Back to pending to return to the AI Suggestions list.