Why a Chrome Extension Is the Right Architecture for AI Dispatch

A dispatcher's day is spent inside a load board. They scan DAT and Truckstop, sort by lane and rate, pull up broker details, and fire off emails — for hours, in the same browser tab. Any tool that wants to help with dispatch has one honest choice to make: meet them inside that workflow, or pull them out of it. Most AI dispatch products pull them out. Numeo Spot doesn't, and that decision drives almost everything about how it's built.

This post is an argument for the boring-sounding answer. A Chrome extension is the right architecture for AI dispatch, not because extensions are trendy, but because the alternative — a separate platform a dispatcher has to migrate into — fights the one habit you can't retrain.

The workflow you can't move

Start with where the work actually happens. DAT's network spans more than 1.7 million trucks and over 500,000 loads posted a day; for a large share of North American carriers it's the first screen they open and the last one they close. A dispatcher doesn't "use a load board" as one task among many — the load board is the desk. Searching, filtering, reading broker names, comparing rates, and opening a quote all live there.

That matters because the freight market is built on small operators with no slack to absorb a tooling change. FMCSA counted roughly 787,000 active carriers as of December 2023, and by ATA's 2025 reporting about 91.5% of them run ten trucks or fewer. The person running dispatch is often the owner, or one of two or three people doing everything. There is no IT department to stand up a new system, no training budget, and no appetite for a workflow that breaks during the switch.

So the expensive thing in dispatch software isn't the software. It's the re-keying. Every product that puts AI in a separate app is implicitly asking a dispatcher to look at a load in one window, then copy the lane, the rate, the broker, and the equipment into another window to get the benefit. That double-entry tax is paid on every single load, all day. It's the worst thing you can do to someone whose edge is speed, and it's exactly what the rip-and-replace platforms charge for.

Meet them where they work

An extension inverts the model. Instead of asking the dispatcher to come to the AI, the AI goes to the board. Numeo Spot runs on top of DAT and Truckstop and layers its logic directly onto the listings already on screen — no second window, no copy-paste, no migration step. The load the dispatcher is looking at is the load the AI is reasoning about, because they're the same DOM.

Concretely, that layering does three jobs at the point of decision, each one in place on the load the dispatcher is already looking at:

• AI ranking. Loads get scored against the carrier's real cost structure and lane history right in the row, so the best option surfaces without exporting anything. The benchmark for "is this rate any good" is grounded — ATRI's 2025 data put the average marginal cost of trucking at about $2.26 per mile in 2024 — and that kind of reference belongs next to the load, not in a spreadsheet.
• Broker safety. Carrier-facing fraud is not a footnote; CargoNet reported roughly $725 million in cargo theft and related losses in 2025. Surfacing what's known about a broker before a dispatcher commits is far more useful inline, at the moment of contact, than buried in a separate dashboard.
• One-click email negotiation. When the dispatcher decides to move, Numeo drafts and sends the broker email from inside the board and tracks the thread for follow-up. Numeo negotiates with brokers primarily by email today, and email is exactly where this workflow already lives — so the AI meets the broker the same way the dispatcher always has.

None of this asks the dispatcher to learn a new place for anything. The information appears in the context where it's relevant, next to the load it describes. That's the whole point: the cognitive map a dispatcher has built over years of staring at the same layout stays intact, and the AI rides on top of it.

Why rip-and-replace loses here

The competing architecture is the standalone platform — its own login, its own interface, its own copy of the data, and an onboarding process to match. On paper it offers more control: one vendor owns the entire experience and can build any screen it wants. In a greenfield market, that's a fine bet.

Dispatch is not greenfield. The board already exists, the dispatcher already knows it, and the broker relationships already run through it. A rip-and-replace platform has to first reproduce the load board the dispatcher already trusts, then convince them to abandon the original, then earn back the speed they lost in the move. That's three hard problems before the AI does anything useful. Meanwhile the dispatcher is comparing a familiar tool they can use blind against a new one they have to think about — and in a job measured in seconds-per-load, thinking is the cost.

The economics make the switch even harder to justify. A median dispatcher earns around $46,860 a year (BLS, 2023); their time is the budget, and a multi-week migration spends it up front for a payoff that's promised later. An extension flips the order. The value shows up on the first load, on the board they already had open, and the cost of trying it is closing a tab. For a market of small carriers who've been burned by "implementations" before, that asymmetry is the entire pitch.

The trade-offs are real

An extension is not a free lunch, and pretending otherwise would undercut the argument. The model has genuine constraints, and they're worth stating plainly.

It's browser-bound. An extension lives in the browser, which means it depends on the board's page structure staying put; when a load board ships a redesign, the layer has to be updated to match. That's an ongoing maintenance cost a self-contained platform doesn't carry, because a standalone app owns its own DOM. It also ties the experience to the browser session — there's no native desktop client, and the work happens where the board does.

It's per-board. Because the layer attaches to a specific board's interface, supporting DAT and supporting Truckstop are two pieces of work, not one. Each surface a dispatcher uses is a surface the extension has to learn. A standalone platform sidesteps this by importing data once and rendering it in a single interface of its own — at the cost, of course, of being that separate interface in the first place.

Both of those are real, and both are the right trade to make. The maintenance burden of tracking a board's UI is the vendor's problem to absorb, not the dispatcher's. The per-board effort is bounded — the freight market concentrates on a small number of boards, not a long tail. What you buy with that effort is the thing that actually moves adoption: zero migration, zero re-keying, and value on the first load. A standalone platform spends the dispatcher's time to save its own engineering; an extension does the reverse, and in this market the dispatcher's time is the scarcer resource.

Where the extension ends

Being honest about the model also means being clear about its edge. An extension is the right shape for the work that happens on the board — evaluating a load, vetting a broker, opening a negotiation. It is the wrong shape for everything that happens away from the board: fleet-wide visibility, multi-load operational dashboards, paperwork and document workflows, the cross-cutting views a dispatcher doesn't want crammed into a load row.

That's not a knock on the architecture; it's a boundary. Some surfaces genuinely call for a dedicated app, and forcing them into an injected panel would be the same mistake as forcing dispatch into a separate platform — just in the other direction. The discipline is matching the surface to the work: per-load decisions belong on the board, where Numeo Spot lives, and fleet-level operations belong in a product built for them. The extension doesn't have to do everything. It has to do the on-board job better than any separate app can, because that's the job no separate app can win.

The takeaway

The architecture question for AI dispatch isn't really technical — it's about whose time you're willing to spend. A standalone platform spends the dispatcher's, asking them to migrate, re-key, and relearn before the AI earns its keep. An extension spends the vendor's instead, taking on the browser-bound, per-board maintenance so the dispatcher gets ranking, broker safety, and one-click negotiation without leaving the board they already live on. The trade-offs are real and the boundary is honest, but for a market of small carriers measured in seconds per load, meeting them where they work is the only architecture that survives contact with the job.

That's the bet Numeo Spot makes: layer the AI onto DAT and Truckstop, change nothing about where dispatch happens, and let the value show up on the first load. See how it works at Numeo Spot.