All posts
How teams are actually using public data to assess risk
2 May 2026Practice6 min
A survey of how risk, trust and safety teams changed their thresholds over the past year, and what happened to their false-positive rates when they did.
- 01
Thresholds moved down, not up
The pattern we saw repeatedly was teams lowering their auto-approve threshold rather than raising it, once they could see the factors behind a score. Explainability bought them the confidence to automate more, not less.
- 02
The review queue changed shape
Total review volume fell, but the share of reviews that ended in a decision reversal rose. That is the outcome to aim for: fewer cases reaching a human, and the ones that do actually being ambiguous.
- 03
Calibration is not a one-off
Every team that held its thresholds fixed for a year drifted. The data underneath a score keeps moving, so a threshold set once is a threshold slowly going out of date.
Recent posts
Tags
- Entity resolution
- Risk signals
- Compliance
- Public data
- Evaluation
- Provenance
Categories
- Research1
- Practice1
- Opinion1