For Alaska freight, trucking & barge operators
Alaska freight runs on paperwork that arrives as a photo, a fax, or a text from a driver with one bar of signal. We build agents that read that paperwork, put it in your TMS correctly, and flag the exceptions a human actually needs to see.
Built in Anchorage · we publish our source code · no long-term lock-in
Written by Kristian Baer, Northtek · Anchorage, Alaska · Updated 2026-08-21
The short answer
By removing the retyping. In a typical Alaska carrier, a load touches a bill of lading, a rate quote, a dispatch board, a proof of delivery, and an invoice, and a person copies the same fields between all five. An agent reads the inbound document - a PDF, a scanned BOL, a photo a driver texted from a gravel lot - extracts the fields, validates them against your rate table and your TMS, and writes the record. Humans stay on the exceptions: a piece count that does not match, a lane with no rate, a POD that came back short. The revenue effect is usually not the labor saving. It is the accessorial and detention billing you stop dropping because nobody had time to chase it.
The operating reality
The road system reaches a fraction of Alaska communities. Everything else moves by barge on a seasonal window, by air on a weight-and-balance sheet, or by Bypass Mail. Any automation that cannot handle that reality is a demo, not a system.
Western Alaska barge service runs a narrow open-water season. A missed cutoff is not a late delivery, it is a customer waiting until next year. Your systems need to reason about a calendar, not just a transit time.
Drivers, agents, and shippers send paperwork as camera photos at an angle, in bad light, with a thumb in the corner. Extraction that only works on clean PDFs solves the easy half of your problem.
A driver on the Dalton, a tender in Bristol Bay, and a village agent may all be offline for hours. Workflows have to queue, sync late, and never lose a record when the signal comes back.
Detention, demurrage, reconsignment, wait time, hazmat, and remote-delivery surcharges are all billable and all routinely forgotten. Capturing them consistently is often worth more than any headcount you might save.
Six workflows we build
Each of these is a real handoff we have watched break. We build them one at a time, in the order that pays for itself fastest.
01
02
03
04
05
06
First 30 days
For most Alaska carriers that is the inbound BOL or manifest. It is high volume, it is structured enough to validate, and the error cost is visible immediately.
We sit with dispatch and billing and trace ten real loads end to end, including the ones that went wrong. We count the retypes and the dropped accessorials rather than asking you to estimate them.
We test extraction on your actual past documents, including the bad photos, and we report the accuracy per field. If a field is not reliable enough to trust, it stays with a human and we tell you which one.
The agent runs alongside your current process for a week with nobody depending on it. You compare its output to what your team produced before anything is switched over.
What you own at day 30
A working intake agent inside your TMS, an accuracy report by field from your own documents, the source code and configuration in a repository you own, and a runbook your operations manager can follow without calling us.
What we built, in the open
The question that matters in this industry is not whether a system read the bill of lading. It is whether you can prove what it read, six months later, when a customer disputes a weight and the original photo is the only evidence either side has.
Our agents are written in a language we built for the purpose, so every action one takes - which field it pulled, which rate table it applied - is recorded as a step you can read back.
A gate that sits between the model and your TMS. It scores whether an extracted field is actually supported by the source document, and we publish how often it lets a bad one through.
Your agents read email and attachments from outside your company. We wrote a scanner for the injection attacks that come with that, because a dispatch inbox is an untrusted input.
Nothing here ships into your operation as-is. It is the reason we can promise an audit trail rather than describe one.
Scope, stated up front
Automation that overpromises on freight paperwork gets caught the first time a customer disputes a weight. Here is where ours stops.
It will not fix a bad rate table
If your lane rates are inconsistent or out of date, automation will quote that inconsistency faster and more confidently. We will find this in week one and tell you.
It will not read every photo
Some field photos are unreadable by any system, human included. The honest target is high accuracy with a clean escalation path, not a claim of perfection.
It does not replace a dispatcher
Judgment about which customer gets the last slot on the barge is a relationship decision. We remove the typing around it, not the call itself.
Where your data goes
Freight data is commercially sensitive - your rates, your lanes, your customer list. Four things we put in writing before the first line of code.
Usually yes. We integrate through whatever your system exposes - API, database, flat-file drop, or emailed reports - and if it exposes nothing, we work through the interface a person would use. We check this in the first week and tell you the honest answer before you commit to anything.
Work queues locally and syncs when connectivity returns. Nothing is lost and nothing silently overwrites a later update. We test this deliberately, because on the Dalton and in the bush it is the normal case rather than the edge case.
You are, the same as if a clerk misread it, which is exactly why we do not build blind automation. Extraction shows its confidence per field, anything uncertain routes to a person, and every record keeps a link to the source image so a dispute can be settled in seconds. We report field-level accuracy from your own historical documents before go-live.
No, and we would push back if you suggested it. Replacing a TMS mid-season is a bigger risk than any workflow it would fix. We build around what you run.
Whatever is being retyped the most, which is usually inbound manifest and BOL data. It is measurable in week one, it does not touch customer-facing communication, and it makes the accessorial capture work possible later because the underlying record is finally clean.
Both. A four-person operation with one dispatcher retyping everything often gets a bigger proportional gain than a large carrier with an established back office. The threshold is task volume, not company size.
Aviation & Air Carriers
Charter quoting, load manifests, and maintenance document tracking for Part 135 operators.
Read →
Seafood & Processing
Fish tickets, crew settlements, and regulatory reporting through a compressed season.
Read →
Resource Development
Permit documents, HSE intake, and procurement for remote industrial operations.
Read →
We will tell you what can be automated, what cannot, and what it is worth. Sixty minutes, no cost, and a written answer either way.
Anchorage, Alaska · info@northtek.io · (907) 903-4353