A pipeline you can audit
Every processed email becomes a run in the Inbox view: what arrived, what the agent parsed, which historical routes it used, what it drafted, and who sent it. When a price was edited before sending, that signal feeds pricing feedback — a drafted-but-unsent quote never trains the calibration.
The Monitor view shows recent pipeline events, and Stats aggregates pricing performance, so you see drift before clients do.
Risk stays gated
Documents with grounding alarms, sanctions hits, or low confidence never auto-file — they wait in an approval queue with the reason stated. A sanctions hit requires an acknowledged operator override, which is logged.
Team members get scoped access: assignment-based visibility for drivers and managers, and a capability matrix that keeps clients and drivers away from agent chat entirely.
Rolling it out to a skeptical team
- Connect one shared inbox first; the agent only drafts, so nothing client-facing changes on day one.
- Let the desk compare drafts against what they would have written for a week or two.
- Turn on Telegram approvals and the Deal Watcher once trust is established.
- Enable 4logist auto-create last, when the team wants analysis filed into the TMS automatically.
Frequently asked questions
Can I see why the agent priced a quote the way it did?
Yes. Each run stores the analysis: referenced routes, filter level, and the reasoning block operators see before sending.
What happens when the agent is unsure about a document?
It leaves fields blank and gates the packet for human review instead of inventing values. Blanks gate; inventions get filed — so the design refuses inventions.