Skip to content

How AI Managed Checks Work

This page walks through what happens behind the scenes when Qualytics AI creates an AI Managed check, from the moment a Profile operation finishes to the point the check appears in your Checks list.

The Profile Operation Is the Starting Point

AI Managed checks always come from a Profile operation. When you run Profile against a datastore, Qualytics:

  1. Walks through every container (table or file) you selected.
  2. Collects field-level statistics — minimum and maximum values, mean, standard deviation, distinct count, distribution shape, completeness, and others — for each field.
  3. Hands those statistics over to Qualytics AI.

Because AI Managed checks depend on real statistics, the more data Profile sees, the better-fitting its checks tend to be. You can re-run Profile periodically to keep AI Managed checks aligned with how the data actually behaves.

Qualytics AI

Qualytics AI is the part of the platform that decides which rules to generate. It does this in four steps:

  1. Receives the metadata generated by the Profile operation for each field in the datastore.
  2. Splits the data into a training set and a testing set so it can validate the rules it proposes against data the AI has not directly used to build them.
  3. Applies a number of machine-learning models and statistical techniques to the training set, looking for rules that fit the observed values well — for example, "this column is always positive", "this column always matches an email pattern", or "this column's volume stays within a predictable range each day".
  4. Filters the candidate rules by running them against the held-out testing set. Only rules that hold true above a confidence threshold survive. Those surviving rules are surfaced in the UI as AI Managed checks.

The result is a set of checks that are statistically grounded in your data, not generic templates. The depth of analysis is controlled by the AI Effort setting on the Profile operation — see Profile Operation for details.

Coverage of Your Rules

AI Managed checks typically cover 80–90% of the data quality rules a team needs day to day. They are designed to be a strong starting point you can rely on without writing any rules manually. The remaining 10–20% — for example, business-specific rules that don't show up in raw statistics — is where Authored checks come in.

Maintenance Over Time

AI Managed checks are kept up to date by future Profile runs:

  • The AI still finds the same rule — the AI Managed check stays as it is.
  • The AI's confidence drops below the threshold or it no longer detects the rule — the check is deprecated and removed during that Profile run.
  • You edited the check in a way that changes how it evaluates the data — it has become Authored, and Profile will not touch it. See AI Managed Checks Introduction for the conversion rules.

This means an AI Managed check has two ways to leave your active set: either Qualytics removes it because the data no longer supports it, or you take ownership by editing it. Profile never modifies or removes an Authored check.

Where AI Managed Checks Appear in the UI

AI Managed checks show up in the same Checks list as Authored checks, but with two visual cues:

  • The Check Summary header displays a purple AI badge (four-point star icon). Hovering it reveals the tooltip "Authored by AI and is continuously tuned to your observed data."
  • The check's source (the Profile operation that created it) is referenced in the check details.

You can filter by check type — AI Managed or Authored — from the Checks tab to focus on one set or the other.

When to keep an AI Managed check

If an AI Managed check fits your business rule out of the box, leave it alone. The next Profile run will keep it up to date for you. Only step in to edit it when you have a specific tuning the AI cannot capture on its own.