Skip to content
Optimal Lab

AI Automation · interactive

Confidence thresholds and fallbacks

A model extracts totals from 24 invoices and says how sure it is. Above the threshold it posts them automatically; below, a person checks. Move the threshold and trade automation for mistakes.

0.99✓0.98✓0.98✓0.97✓0.97✓0.96✓0.95✗0.95✓0.93✓0.92✓0.91✓0.90✗0.88✓0.86✓0.84✗0.82✓0.80✓0.77✗0.74✓0.70✗0.66✓0.60✗0.55✗0.50✗

Solid: posted automatically. Dashed: sent to a person. ✗ means the extraction was wrong.

Automated
50%
To a person
12
Errors caught
6
Errors posted
2

At 0.90: 12 of 24 posted on their own, 2 wrong ones among them. 12 go to a person.

  1. 1

  2. 2

  3. 3

    A wrong payment costs more than a person's two minutes, so finance sets the line high. Tagging a support ticket can go lower.

    The threshold is a business decision.

  4. 4

In production

Tools and technologies

Human review queues are a standard part of document automation.

Not tied to one language — use these names when you search docs or design a system.