KinderCare Learning Companies · Digital Products

AI decision review and QA

One tool reviews a decision to use AI and tells you which precautions are missing. The other walks you through the five-step QA before anything is released. Both work on how sensitive the information is, not on a list of approved tools.

Run the review before the work starts, and the QA before it ships. The review sets the depth each QA step needs, so doing them in order saves rework.

·

1. Review the decision
2. QA the output
Reference

The four groups

Group 1 · Productivity Public and general business information

Drafting an internal email, summarizing a public document, brainstorming, tightening wording.

Group 2 · Confidential Confidential company information

Financials, unreleased strategy, contracts, roadmaps, center performance data, internal reporting.

Group 3 · Sensitive Regulated or personal information: PHI, claims, HR, credentialing

Health information, claims data, employee records, background check and credentialing files, child records.

Group 4 · Consequential The AI decides or acts on its own

Changing records in a system, sending communications, approving or routing work, altering what a family sees with no person in between.

The consequence scale

LowA clumsy slide. Someone notices it is off and it gets fixed.
MediumA bad decision. Someone acts on wrong information and real time or money goes into undoing it.
HighA family or center is affected.

What can go where

Do and don't

Do
  • Start with trusted source materials.
  • Ask AI to cite sources when possible.
  • Use a standard review checklist.
  • Maintain version control and document approvals.
Don't
  • Publish AI output without review.
  • Trust numbers or requirements without validation.
  • Let AI content bypass governance.
  • Treat AI output as the source of truth.

The four groups, the consequence scale, and the five QA steps are KinderCare's. The precautions are deliberately high level: we do not yet have a definition of which tools are approved for which data, so this asks whether the data is sensitive or confidential and whether anyone has been asked, rather than pointing at a list that does not exist. Technology, Privacy, and Legal own that definition when it comes.