# Microsoft reports 41% fewer fields selected for targeted document review

> Microsoft described a document-processing evaluation with an insurance organization in an October 14 technical report. More than 22,000 field values from over 1,000 production documents were scored against human-checked answers. On a held-out test set, 41% fewer fields were selected for targeted review than when selecting every field, while the selected set contained more than 80% of known extraction errors.

- Canonical URL: https://newsverum.com/en/news/microsoft-document-confidence-targeted-review-evaluation-2026
- Publisher: Newsverum
- Section: Technology
- Published: 2026-10-15T21:30:20+00:00
- Updated: 2026-10-15T21:30:20+00:00
- Source review period: 2026-10-14
- Other editions: [Türkçe](https://newsverum.com/tr/news/microsoft-document-confidence-targeted-review-evaluation-2026.md), [العربية](https://newsverum.com/ar/news/microsoft-document-confidence-targeted-review-evaluation-2026.md), [Français](https://newsverum.com/fr/news/microsoft-document-confidence-targeted-review-evaluation-2026.md), [Español](https://newsverum.com/es/news/microsoft-document-confidence-targeted-review-evaluation-2026.md), [Deutsch](https://newsverum.com/de/news/microsoft-document-confidence-targeted-review-evaluation-2026.md), [Português](https://newsverum.com/pt/news/microsoft-document-confidence-targeted-review-evaluation-2026.md)

*The source does not state the event date; the period shown is the source review period.*

This does not mean staff time fell by 41%. People continued checking every field in the sensitive insurance workflow; confidence scores prioritized attention. Thresholds were calibrated by document type and field, and Microsoft does not present the results as a guarantee for other documents or model versions.

## Data

### Microsoft reports 41% fewer fields selected for targeted document review

Microsoft reports 41% fewer fields selected for targeted document review

Methodology: Reported research results are attributed to the source; no independent performance measurement was made.

| date | label | description | status |
|---|---|---|---|
| 2026-10-14 | Publication of the research report | Microsoft described a document-processing evaluation with an insurance organization in an October 14 technical report. More than 22,000 field values from over 1,000 production documents were scored against human-checked answers. On a held-out test set, 41% fewer fields were selected for targeted review than when selecting every field, while the selected set contained more than 80% of known extraction errors. | occurred |

Sources: <https://devblogs.microsoft.com/foundry/content-understanding-confidence-grounding-hitl/> · Download data: [JSON](https://newsverum.com/api/v1/articles/microsoft-document-confidence-targeted-review-evaluation-2026/data/research-publication?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/microsoft-document-confidence-targeted-review-evaluation-2026/data/research-publication?lang=en&format=csv)

## Sources

- [Microsoft Foundry Blog](<https://devblogs.microsoft.com/foundry/content-understanding-confidence-grounding-hitl/>) (Source checked: 2026-10-15; Published: 2026-10-14)

---

Cite the canonical URL above. [Editorial standards and corrections](https://newsverum.com/en/editorial)