What changed since launch: the booking agent launched in June 2026. Since then it added tiered outbound (your carriers first, DAT as the automatic fallback), rules that inherit from enterprise down to workspace, an auto-award switch, and carrier memory that never re-asks a known carrier for their DOT or MC.
Chain's booking agent handles the carrier side of a load end to end. It answers carrier emails and offers, resolves who the carrier is, runs them through your vetting policy, negotiates inside thresholds you set, and assigns the carrier back in your TMS. It runs on the same load thread as everything Autopilot already did after booking.
One agent, not another bot
Chain has run tracking in production for five years. Autopilot has worked on top of it since early 2025. Carrier sales was the obvious next thing to automate, and the obvious way to build it would have been a second bot with its own inbox and its own idea of what a carrier is. We didn't, because brokerages with six bots end up with six places to look when something goes sideways. The negotiation that covered the load sits directly above the pickup check-in, which sits above the POD. One record.
Outbound: your carriers first, the board last
Outbound works in tiers, on a schedule you control:
- Contract carriers already set up in your TMS.
- Core carriers who've run the lane with you before.
- DAT, posted automatically as the fallback when the earlier tiers pass.
Outreach goes by email and text. Carriers reply with a number or use an offer link, and every offer, from any channel, lands in one view with its source attributed. If a post has no rate, a SONAR range can load onto it.
The thresholds decide. The model writes the words.
You set three numbers: a start rate, a target and a ceiling, from your TMS or the rating tool you already use. Those numbers decide the action:
- At or below target, with auto-award switched on: award.
- Well above the ceiling: decline.
- In between: counter, and never above the carrier's own ask.
- No number in the message: ask for one. It won't invent a counter.
- Anything that would book a loss, or that you haven't authorized: escalate to a person, with a briefing.
Rules that only narrow
Set policy once at the enterprise level. Workspaces inherit it, and they can tighten it but never widen it. A division can exclude hazmat or lower its ceiling. It can't raise the ceiling the enterprise set. Every rule change can dry-run against real loads before it goes live, so you see what it would have done first.
Carrier memory
A carrier who has already given their DOT or MC isn't asked again. History carries across negotiations: this carrier with you, this lane, this facility. Vetting still runs on every booking, before a rate con can exist.
What it won't do
- Counter above the carrier's ask.
- Guess a number when it has no benchmark.
- Auto-award if you haven't switched it on.
- Keep talking when your rep jumps in. One message pauses it on that offer.
- Discuss detention, lumpers or any dollar figure outside the rate.
- Pretend to be a person if a carrier asks.
Every turn writes an audit trace: what stage it reached, whether the outcome came from a guardrail, vetting or the model, and why. Edit a setting mid-negotiation and the conversation in flight keeps the policy it started with.
An honest note on scope
Tracking has run in production at Chain for five years. The booking agent launched in June 2026. It's a start. We'll show it to you on live loads before you sign anything.