Work Orders: The Information Gap That Costs You

A technician drives ninety minutes, finds the unit is a different model than the one on the ticket, does not have the part, and drives back. Nobody did anything wrong on site. The job was lost at intake, by a form that never asked which model it was.

What is actually missing

Ask a dispatcher why a job needed a second visit and the answer is rarely about skill. It is about a detail that could have been collected in ninety seconds on the phone and was not.

The same handful of gaps account for most of it, and each one has a predictable cost.

Missing at intake	What it causes	Where to fix it
Model or serial number	Wrong part on the van	Asset record, not the phone call
Access details	An hour waiting for a key	Site record, updated after each visit
Who can approve spend	Work stops pending a call back	Customer record, with a limit
History on that asset	Diagnosing something already diagnosed	Attached to the order at dispatch
The symptom in the caller words	Fixing the wrong thing confidently	A free-text field nobody summarises

Leaving the caller description in their own words matters more than it looks. A dispatcher who rewrites it as compressor fault has made a diagnosis from a phone call, and the technician now arrives looking for a compressor fault rather than listening to the customer.

The close-out is where the money is

Unbilled hours almost never come from technicians hiding their time. They come from close-out notes that cannot support an invoice two weeks later when a customer queries it.

A note reading unit serviced, working now is a receipt for nothing. It does not record the two hours of diagnosis, the part fitted, or the extra work the customer verbally approved on the spot. All three happened. None can be charged with confidence.

This is the one place to be strict. A work order should not be closable without time on site, parts used, work performed, and anything the customer approved while the technician was there. Those four fields are the invoice. The same logic applies to any checklist people have to complete under time pressure, which is that it has to be fast enough to actually get done.

Where AI earns its place

Nobody standing in a mechanical room with dirty hands is going to fill in eight fields on a phone. They will talk for forty seconds. Turning that into structured fields is the single highest-value thing to automate here, because it removes the reason the data was never captured.

From the same note you can draft the customer-facing summary, which is normally written badly or not at all, and which is what the customer actually judges the visit by.

Two more are worth having. Pulling the history on an asset into the order before dispatch, so the technician is not rediscovering last March. And flagging orders missing a field that has historically preceded a return visit, which is a pattern your own closed jobs can tell you, in the same way finding the real bottleneck starts with the data you already have.

What to keep away from it

Priority is the obvious temptation and the wrong one. A system that reorders the queue by predicted urgency will eventually deprioritise the customer who never complains, and you will not notice until they leave. Let it surface the information and let a dispatcher decide.

Diagnosis from a symptom description is the other. A suggested cause is useful next to the history. A suggested cause presented as the answer produces technicians who stop looking, and the predictive maintenance caveats apply here for the same reason.

Do not start by buying software. Decide which five fields are mandatory, enforce that for a month, and measure first-time fix rate before and after. Most of the available gain is in that decision, and it costs nothing. A platform purchase afterwards is far better informed, which is the argument in where field service software earns its place.

Frequently Asked Questions

What makes a work order fail?

Almost always something missing at intake rather than something done badly on site. The model number was not recorded, nobody noted that the unit is on a roof behind a locked hatch, or the person who called is not the person who can authorise the spend. The technician discovers each of these after arriving, when every remaining option costs money.

What should be mandatory before a work order is dispatched?

Exact location and access details, the specific asset including model and serial where one exists, a description of the symptom in the caller words, who can approve additional work and up to what amount, and the service history on that asset. Five fields, all answerable at intake, each of which prevents a specific type of return visit.

Why do so many hours go unbilled?

Because the close-out note does not support the invoice. A note that reads fixed leak tells your billing team nothing about the two hours of diagnosis, the part fitted or the work the customer approved while the technician was there. The hours happened. The evidence for charging them did not get written down, and a week later nobody can reconstruct it.

Where does AI actually help with work orders?

Turning a spoken close-out note into structured fields, drafting the customer-facing summary from that note, pulling the relevant history on an asset before dispatch, and flagging work orders that are missing a field that has historically predicted a return visit. The scheduling decision and the technical judgment stay with people.

Do you need dedicated work order software to improve this?

Not to start. Most of the gain comes from deciding which fields are mandatory and enforcing that, which you can do in whatever you already use. Software makes the enforcement easier and the history searchable, and it will not fix an intake process that never asked the right questions in the first place.

How do you measure whether this is working?

First-time fix rate and the proportion of logged hours that make it onto an invoice. Both are unglamorous, both move within a quarter of changing the intake form, and both translate directly into money. Count of work orders closed is the metric that looks busy and tells you nothing about whether the second visit was avoidable.