AI Managed Checks in Practice

AI Managed checks provide broad starting coverage from observed data behavior. Your team supplies the business judgment: choose the right oversight setting, review controls that protect important decisions, document known exceptions, and add Authored checks for expectations that statistics cannot establish.

The scenarios below show when to keep an AI Managed check, when to take ownership of it, and when to add a separate business rule.

Scenario 1: Review a Baseline Rule

Context. You run Profile on an orders table. Qualytics creates a Not Null check for order_id with Coverage = 100% because every profiled record contains an identifier.

What happens. A data steward confirms that every order must have an identifier. If Oversight kept the new check in Draft, the steward activates it. Later Profile runs continue to maintain it as an AI Managed check.

Why it works. The observed pattern and the business expectation agree. Keeping the check AI Managed lets Profile maintain the baseline without removing human review from an important control.

Scenario 2: Account for a Documented Exception

Context. Qualytics creates a Matches Pattern check for phone with Coverage = 100%. Your team knows that a small set of records still uses a legacy format during a planned migration.

What happens. After confirming the accepted exception with the data owner, you lower Coverage to the approved percentage and document the reason in the description or metadata. Saving the evaluation change converts the check to Authored.

Why it works. The Authored check preserves a deliberate business decision. Future Profile runs will not replace the approved coverage setting.

Scenario 3: Correct Rule Logic

Context. An AI Managed Max Length check for product_sku uses a limit of 12 characters. The product team confirms that a new, valid SKU format can contain up to 16 characters.

What happens. You edit the rule property from 12 to 16, document the new format, and save. The check converts to Authored because its evaluation logic changed.

Why it works. Observed history could not predict the approved format change. A person supplies the new business requirement, and conversion ensures that later Profile runs preserve it.

Scenario 4: Let a Behavioral Baseline Adapt

Context. An AI Managed Volumetric check monitors the daily row count for a transactions table. As scheduled observability measurements accumulate, Qualytics adjusts the expected bounds after normal seasonal growth.

What happens. The data owner reviews the updated bounds and compares them with expected business activity. If they still represent normal behavior, the check remains AI Managed. If a one-time event should not redefine the baseline, the owner edits the properties and documents the exception, converting the check to Authored.

Why it works. Measurement history can adapt the observed baseline, while a person decides whether the change reflects normal business behavior.

Scenario 5: Add a Narrower Business Rule

Context. An AI Managed Max Value check confirms that discount_pct never exceeds 100 across an orders table. Partner sales have a separate policy that caps discounts at 25.

What happens. Clone the AI Managed check, add a filter such as sales_channel = 'partner', change the maximum value to 25, and validate the draft. The clone is an independent Authored check, while the original AI Managed check continues to cover the table-wide pattern.

Why it works. The observed baseline remains useful, and the Authored check captures business context for a narrower group of records.

Scenario 6: Govern a Large Set of Checks

Context. A Profile operation creates many AI Managed checks across a large datastore. Reviewing every check with the same depth would hide the controls that matter most.

What happens. Your team uses a risk-based review:

1. Keep new checks in Draft when the datastore requires approval before activation.
2. Start with data used for critical decisions, regulatory reporting, customer-facing processes, or automated actions.
3. Review checks with active anomalies or unexpected changes in History.
4. Activate AI Managed checks that match the intended use of the data.
5. Edit or replace checks that need business-specific logic, and add Authored checks for missing expectations.

Why it works. AI supplies scalable starting coverage, while people concentrate governance effort where an incorrect control or missed issue would have the greatest impact.

Quick Reference: When to Step In

For the full conversion rules, see Introduction. For the screen-by-screen edit walkthrough, see Edit an AI Managed Check.