For Alaska electrical, mechanical & plumbing contractors
MEP shops lose margin in the gap between what a technician did and what the office could bill for. We build agents that turn field notes, photos, and timestamps into complete tickets before the memory of the job goes cold.
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
In ticket completeness and dispatch speed. A technician finishes a call and writes three words in the app. Two weeks later the office cannot bill the trip charge, the after-hours premium, or the parts because the ticket does not support them. An agent prompts the technician with the specific missing fields while they are still on site, reads their photos to identify equipment and part numbers, and assembles a billable ticket the same day. On the dispatch side it reads inbound service requests, matches them to the right technician and truck stock, and drafts the response. Neither task requires judgment. Both leak money every week they stay manual.
The operating reality
A no-heat call in January is a different business than a controls retrofit at a school in July, and most shops here do both with the same crew and the same paperwork system.
A no-heat or frozen-line call has to be triaged in minutes. Dispatch decisions made under pressure are where documentation gets skipped and revenue quietly disappears.
If the part is not on the shelf in Anchorage it is days out at best. Knowing at the point of dispatch what is actually in truck stock changes the callback rate more than any scheduling tweak.
School districts, municipalities, and state facilities carry certified payroll obligations. Different classifications on different jobs in the same week is a recurring error source.
The technician who knows which boiler at which building has the intermittent fault is a single point of failure. Service history that is searchable in plain language turns one person’s memory into shop knowledge.
Six workflows we build
Ranked by how directly each one shows up in a month of billing.
01
02
03
04
05
06
First 30 days
It is the highest-frequency event in the shop, it happens under time pressure, and the cost of an incomplete ticket is measurable in your own invoice history.
We pull a month of closed tickets and identify which ones were short-billed or unbillable because of missing documentation. That number, from your own data, is the business case.
We build the close-out assistant against the way your technicians actually work, on the phones they actually carry, with offline queueing for mechanical rooms with no signal.
Two technicians run it for two weeks against a control group. If their ticket completeness does not measurably improve, we say so rather than expanding the pilot.
What you own at day 30
A close-out assistant live in your field app, a before-and-after ticket completeness measurement from your own data, the source code and configuration, and a fifteen-minute training doc your technicians will actually read.
What we built, in the open
Most AI tooling assumes a desk, a browser, and a connection. Field service in Alaska assumes none of those. The parts of our stack that matter to an MEP shop are the unglamorous ones: local queueing, sync that never clobbers a newer edit, and extraction that survives a photo taken in a dark boiler room.
We wrote our own memory server, including an offline mode. That is why a close-out completed in a basement with no signal is held locally and reconciled correctly later instead of being lost.
A language built so an agent's actions are inspectable. When a ticket gets billed and a customer questions a line, the path from photo to charge is readable.
Verification between the model and your field service platform, so a part number read off a blurry nameplate is checked against the source image rather than trusted.
None of it ships to your technicians as a product. It is why offline is a normal case for us rather than a caveat.
Scope, stated up front
A tool that oversells itself in the field gets abandoned by technicians in a week. Three things this will not do for your shop.
It cannot diagnose equipment
It reads the nameplate and recalls the history. Deciding what is wrong with the boiler is your technician’s work, and anything that blurs that line is a liability we will not build.
It will not fix technician adoption by itself
If your field team already ignores the app, adding AI to it changes nothing. We will tell you when the real problem is process and not tooling.
Takeoff still needs an estimator
Drawing extraction is an accelerator with an error rate, not an autonomous estimator. It gets reviewed, and we design the review step in from the start.
Where your data goes
Service history is the most valuable asset in a shop and the easiest to lose. Four commitments we put in the agreement.
No. ServiceTitan, Housecall Pro, Simpro, and similar platforms stay as your system of record. We build the layer that makes what goes into them complete. Replacing a field service platform is an expensive way to solve a data quality problem.
Because it asks for less, not more. Instead of a form with fourteen fields it reads their photos and asks two specific questions. In pilots the adoption argument that works is the one that shortens their day, not the one about billing accuracy. If it does not shorten their day, they are right to ignore it and we would rather find that out in a two-technician pilot than after a shop-wide rollout.
Close-out works offline and syncs when the phone gets a connection. Nothing is lost and nothing overwrites a later edit. This is a normal case in Alaska buildings, not an edge case, and we test it deliberately.
Yes, as a pre-submission check. It splits hours by classification, verifies fringe and overtime arithmetic, and flags what will not pass review. It is a second set of eyes before filing, not a substitute for your compliance process.
We answer that with your own numbers rather than a generic figure. The month-one ticket audit tells you how much revenue is currently being lost to incomplete documentation, and the two-technician pilot tells you how much of that comes back. Both numbers come from your invoice history, so you are looking at a measured payback period rather than a projection, and we size the engagement to it.
Yes. Controls, fire alarm, low voltage, and building automation shops have the same close-out and service history problems, often worse because the systems are more complex and the institutional knowledge is concentrated in fewer people.
Construction & Contractors
RFIs, submittals, and change order packaging for GCs and heavy civil.
Read →
Utilities & Energy
Outage communications, work orders, and regulatory filings for co-ops.
Read →
Real Estate & Property
Maintenance intake and tenant communication for property managers who call you.
Read →
We will show you what was short-billed and why, from your own data. Sixty minutes, no cost, and you keep the analysis regardless.
Anchorage, Alaska · info@northtek.io · (907) 903-4353