Applied AI

Look beyond a single average score

Editorial edition · 16 November 2026

The idea

A strong average can conceal a weak result for a particular language, document type or request category. Slice analysis groups outcomes by meaningful conditions. It helps teams discover where a system needs a narrower scope or better supporting data.

Design choice

Choose slices before inspecting every result, and report their sample sizes. Very small groups produce unstable estimates. Investigate differences without assuming that a descriptive gap establishes a causal explanation.

Worked example

An assistant performs well on short English instructions but poorly on Dutch tables. Separate measurements reveal that table extraction, rather than translation alone, needs investigation.

Before you ship

  • Report sample size per slice
  • Include relevant languages and formats
  • Avoid causal claims from descriptive gaps

Examples are illustrative. Adapt the checks to your own environment.

Further reading

Model evaluation, scikit-learn

Search the library