Confidence Levels and Why They Matter

Two scores can both be 4. But one might be backed by five distinct passages of clear, behavioral evidence, and the other by a single ambiguous statement. They're not equally trustworthy — and treating them the same would lead to different decisions than treating them differently.

Savo attaches a confidence level to every dimension score so that distinction is always visible.

The three confidence levels

High — Strong, consistent evidence across multiple passages. The score is well-supported and reliable.

Medium — Sufficient evidence to produce a score, but some gaps or inconsistency in the signal. Treat the finding as directional and worth investigating further.

Low — Minimal or conflicting evidence. A score exists, but it should be held lightly — additional sessions or a more targeted event design may be needed to sharpen the finding.

What drives confidence

Confidence is a function of three things:

Amount of evidence — How many relevant passages did the conversation produce for this dimension? A finding supported by five distinct evidence units is more reliable than one supported by one.

Consistency — Do the evidence units point in the same direction, or does the participant give conflicting signals across the conversation? Consistent evidence produces higher confidence than mixed signals at the same overall quantity.

Relevance — How closely does each evidence unit speak to the dimension being scored? Direct, behavioral evidence is more confidence-building than oblique or inferential references.

Confidence vs. abstention

Confidence levels apply to scores that exist. When there isn't enough evidence to produce a reliable score at all, Savo doesn't assign a Low confidence score — it abstains entirely and records ""No Score.""

This distinction matters: a Low confidence score of 4 means ""we found some evidence pointing to High, but not enough to be certain."" A No Score means ""we didn't find enough evidence to say anything reliable."" These are different kinds of uncertainty and warrant different responses.

How to use confidence in practice

When you read an Insights report, treat High confidence findings as reliable enough to act on directly. Treat Medium confidence findings as strong signals worth investigating. Treat Low confidence findings as hypotheses — they may be real, but you should look for corroboration before making significant decisions based on them alone.

Low confidence and No Score findings can also inform your event design: if a dimension consistently produces weak evidence, it may be that the facilitation guide needs stronger signal anchors for that dimension.

For how scores work, see Understanding Dimensions. For how to inspect the evidence behind any score, see Evidence Units.