Skip to content

AI Managed Checks in Practice

Qualytics AI takes care of the heavy lifting — proposing rules that fit the shape of your data. Most of the time, working with AI Managed checks is reviewing them, monitoring them, and stepping in only when a specific tuning is needed. This page walks through real-world scenarios with concrete examples that show how an AI Managed check appears to you and what the right next step usually looks like.

The examples below use simple, recognizable data (clinical values, phone numbers, regions) so the pattern is easy to follow regardless of your industry.

Scenario 1 — A clean rule that fits out of the box

The situation: You ran Profile against a lab_results table that holds clinical measurements. Qualytics AI created a Min Value check on glucose_level with Value = 40 and Coverage = 100%. Looking at the rule, you confirm it: glucose readings below 40 mg/dL are clinically rare and would signal a real problem.

What the check looks like in the UI

Field Value
Rule Min Value
Field glucose_level
Value 40
Coverage 100%
Owner AI Managed
Active Anomaly 0

What to do: Leave it alone. The next Profile run will keep the rule up to date for you, and the next Scan will evaluate it. No action needed unless anomalies start appearing.

This is the most common scenario — between 80% and 90% of AI Managed checks in a typical deployment fit this pattern.

Scenario 2 — Tuning coverage for a known exception

The situation: Qualytics AI generated a Matches Pattern check on phone enforcing the international E.164 format with Coverage = 100%. After looking at recent anomalies, you realize about 0.3% of phone numbers use a legacy in-house format that hasn't been migrated yet — and management has accepted that gap until the migration is complete.

The problem: With Coverage at 100%, every legacy record produces an anomaly, drowning out genuinely bad phone numbers in the noise.

What to do: Open the check on the Edit an AI Managed Check page and lower Coverage from 100% to 99.5%. Save with Update.

What changes after saving:

  • The badge in the Summary section flips from AI Managed to Authored.
  • Future Scans tolerate up to 0.5% non-matching records before flagging an anomaly.
  • Profile no longer touches the rule — your tuning is preserved.

Why the conversion is helpful here

The conversion-to-Authored is exactly what you want in this case: without it, the next Profile run could reset coverage back to 100% and your tuning would be undone. By converting, you take responsibility for the rule and Qualytics gets out of your way.

Scenario 3 — Refining a rule that is almost right

The situation: Qualytics AI generated a Matches Pattern check on email using a strict RFC 5322 regular expression. Most addresses match, but you discover the AI's pattern rejects valid addresses that contain apostrophes (for example, O'Connor@example.com), producing false-positive anomalies.

What to do: Open the check, edit the regular expression in the Properties section to allow apostrophes in the local part, and save.

What changes after saving:

  • The check is converted to Authored because rule properties changed.
  • Profile no longer manages the regex — your refined version is preserved across runs.
  • Future Scans use the new pattern; old anomalies remain in History but no new ones are produced for the previously affected records.

When to use Edit vs. Clone

If the original rule was almost right and you just need to tweak it, edit it — that converts the AI Managed check to Authored. If you want to keep the original AI Managed check intact and add a separate variation alongside it, clone it instead. Clone always creates a fresh authored draft without touching the original.

Scenario 4 — Watching the AI adjust a rule over time

The situation: A Volumetric check on the transactions table monitors expected daily row count. Looking at the History panel, you see entries showing Qualytics adjusted the lower and upper bounds during the last couple of Profile runs. These adjustments are normal: Profile retrains the volumetric bounds based on observed traffic. As long as the new bounds match what your business expects, no action is needed.

What to do:

  • If the bounds look right — leave the check alone. The AI is doing its job.
  • If the bounds drift in a direction that doesn't match your knowledge of the business (for example, after a known traffic spike that should not be the new normal) — edit the volumetric properties to set bounds that match your expectations. The check converts to Authored and Profile stops auto-adjusting.

History is your audit trail for AI Managed checks

The History panel on the right side of the check overview is the most reliable way to spot when an AI Managed check has been edited (by you or anyone else) and when Profile last adjusted it.

Scenario 5 — Cloning to scope a rule narrower

The situation: An AI Managed Min Value check on glucose_level enforces Value >= 40 across the whole lab_results table. You want a stricter threshold (Value >= 70) but only for records from the ICU department, while keeping the original rule for the rest of the data.

What to do:

  1. Open the AI Managed check.
  2. Click the 3-dot menu at the top of the check details and select Clone (the duplicate icon). Qualytics opens a new check pre-filled with the same rule and target. The clone starts as Authored and Draft.
  3. Add a Filter Clause like department = 'ICU'.
  4. Change the Value in Properties to 70.
  5. Run a Dry Run to confirm the new rule catches what you expect.
  6. Validate, then Update to save the cloned check. To activate it later, open the Update dropdown next to the button and select Activate.

The end result: Two independent checks living side-by-side — the original AI Managed one covering the table-wide rule, and the new Authored one applying a stricter threshold inside the ICU.

Scenario 6 — Letting the AI handle most of the work

The situation: You have a 200-table datastore. After the first Profile run, Qualytics AI generates around 1,500 AI Managed checks covering completeness, uniqueness, value ranges, format patterns, and basic distributions.

The temptation: Open every check and review.

The pragmatic approach:

  1. Trust that the AI Managed checks are reasonable starting points — most will be.
  2. Run a Scan and let anomalies surface. The checks that produce real, actionable anomalies are the ones worth investigating first.
  3. Tune those specific checks (Scenarios 2 or 3). The rest can stay AI Managed.
  4. Layer Authored checks on top for business rules that statistics alone can't capture (for example, "every order with status = 'shipped' must have a tracking number") — see Authored Check.

This is how most teams use the AI: 80–90% of rules stay AI Managed and self-maintain, and human attention focuses on the 10–20% of business-specific Authored checks.

Quick reference — when to step in

If you see... Do this... Result
A clean AI Managed rule with no anomalies Nothing Profile keeps it up to date
Many false-positive anomalies on a small fraction of records Lower Coverage Converts to Authored, tolerates known exceptions
The rule's logic is slightly off (regex too strict, threshold too low, etc.) Edit Properties Converts to Authored, preserves your fix
You want a stricter or differently-scoped variant Clone, then edit the clone Both checks coexist independently
Profile keeps overwriting your changes Make an evaluation edit (Properties, Coverage, Filter, or Field) Converts to Authored — Profile leaves it alone
The rule is fundamentally wrong for your data Delete it Profile may recreate on the next run if the data still suggests it; otherwise it's gone

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