Decision Trace
A Decision Trace is a complete, step-by-step record of every action an AI agent takes: what triggered the workflow, what data was gathered, what intelligence informed the decision, what action was taken, what the response was, and what the outcome was.
Why It Matters
When your board asks 'how is AI making decisions in our supply chain,' you need an answer that is specific, auditable, and exportable. The Decision Trace provides exactly that.
The FourKites Perspective
FourKites captures the full decision trace for every agent action across 2.5 million executions per month. These traces are not just audit records. They feed back into the Graph, teaching the system what works and making the next decision better.
Frequently Asked Questions
What does a Decision Trace record?
Seven elements: trigger event, data gathered, intelligence consulted (from the Graph), decision made (with reasoning), action taken, response received, and outcome recorded. Every element is timestamped and exportable.
How is this different from an audit log?
An audit log records that something happened. A Decision Trace records why it happened, what intelligence informed it, and what the outcome was. It is the difference between 'Tracy emailed the carrier' and 'Tracy emailed the carrier because the Graph showed a 78% success rate for direct dispatcher contact on this lane pattern.'
How do Decision Traces improve the system over time?
Every trace feeds back into the Graph. When Resolution Path B consistently outperforms Path A on a specific carrier-lane combination, that pattern becomes the default. 2.5 million traces per month, each one teaching the system something new.
Related Concepts
Related Outcome PAGES
Related PLATFORM PAGES
See how this concept powers autonomous operations.