Most freight quote requests don't arrive through a form. They arrive as long email threads, in many languages, with packing lists attached. We built an agent that works the inbox the way a good sales coordinator would.
YearJun 2026
My roleAI/UX design lead, outcome owner
ChannelCustomer email threads
StackLLM agents, rate and location APIs
StatusProof of concept
Vibe-coded UI The thread inbox: every quote request with its mode, lane, status and whether the agent or a person owns it.
Outcome
What moved.
85%
Less time to quote
Any language
Replies in the customer's own
100%
Threads with a full audit trail
Before the work started
What we believed, what stood in the way, and where the opening was.
Assumption
Customers will use the quote form.
Most don't. They reply to an old email, attach a packing list and write in their own language and vocabulary.
Challenge
Long threads, missing fields, risky cargo.
Threads run 10 to 15 emails. Fields are scattered across the body, signatures and attachments, and some cargo must never be auto-quoted.
Opportunity
Meet customers in their inbox.
Let an agent read the whole thread, ask only for what's missing, quote in the customer's language, and step aside when a human should decide.
The problem
Quote requests arrived as emails, not form submissions. Sales coordinators read each thread, pieced together origin, destination, cargo and dates from the body, the signature and the attachments, re-keyed it into the rate tools, and replied hours later. Every follow-up email started the process again.
The goal: get a trustworthy, priced reply back while the customer is still at their desk, without losing the human judgement that some shipments need.
The process, on request
Interested in how this works, and the process behind it?
Discovery, structure, how it works, how we built it and what we learned, with 8 visuals. This part is encrypted and opens with a password.
2025The dropdown that was mis-routing every lead94% leads routed right
2024Migrating a 5,000-page legacy marketing site92% faster publishing
2017Enterprise UX for the Crown Prince Court, Abu Dhabi60% faster processing
2016A legacy banking app, rebuilt around real behaviour70% fewer errors
Agentic AI
Feb 2026Photo-to-volume moving estimator80% more accurate quotes
Agentic AI
Aug 2026Agentic invoice reconciliation for finance70% confidence score in the PoC