My mechanics hate typing up repair orders. What are the options?

There are four, and they are not equally good. Give the writing to a service writer. Let techs dictate — free on any phone, or into the shop system you already own. Build out canned jobs so most of the typing never happens. Or use AI that turns a spoken note into a structured repair order — separate jobs, labor hours, parts — as a draft a person approves. Start with canned jobs: it costs nothing and you already pay for the tool that does it. And know the trap before you spend money. Plain dictation gives you a paragraph, and a paragraph is not a repair order. Someone still has to break it into line items.

It is rarely about typing speed. A repair order is a structured document: jobs separated, labor hours against each, parts with numbers, and a cause-and-correction line that has to survive a warranty claim or an argument with a customer six months later. Producing it means stopping work, walking to a screen, and thinking in a format that has nothing to do with the repair. On a flat-rate pay plan, those minutes do not bill. So notes get written short and late, the writer sends them back for detail, and the tech learns that a good note creates more work than a bad one. That loop is the actual problem. A tool that speeds up the typing without changing the loop leaves the tech doing the same disliked job, faster.

The oldest answer, and for plenty of shops the correct one. The tech talks for two minutes at the bay and a service writer builds the repair order properly. The cost is a wage, or the reassignment of someone already on the floor. If you run two or three techs and a light board, this is the entire solution and no software improves on it. Where it fails is volume and concentration of knowledge: the writer becomes a bottleneck mid-afternoon, the repair order is only as good as a hurried conversation over a running engine, and when that person leaves, the shop's documentation quality leaves too.

The keyboards on iOS and Android dictate for free, as do Windows and macOS; paid tools such as Dragon add a custom vocabulary, which is where you teach it your part numbers and acronyms. Cost runs from nothing to a modest per-seat license. This genuinely helps: a phone in a pocket beats a keyboard when your hands are covered in grease, and a tech who will not type will often talk. Two limits. Bay noise — air tools, compressors, a lift cycling — degrades general-purpose speech models, and DTC codes and part numbers come back mangled more often than plain English. And the output is a block of prose in a notes field: it makes the tech's half of the job easy and leaves the writer's half where it was.

Canned jobs — also called job packages, saved services, or menu items depending on the vendor — are prebuilt line items with labor times and parts already attached. Most shop management systems have some version of this; check what yours calls it. An afternoon spent building the twenty jobs your shop repeats most costs nothing beyond what you already pay, and removes more typing than any AI product will, because the text never has to be produced at all. Brake service, cooling system flush, the diagnostic you sell four times a week: none of those should be typed from scratch again. A shop that skips this and buys software has bought a faster way to do something it did not need to do. Check too whether your vendor has added a voice or AI note feature to the tier you already pay for. Included beats bought.

Shops discover this after buying a recorder or turning on dictation: the transcript is not the work. A tech says: found the right front wheel bearing howling on a road test, sway bar links are shot, front pads are down close to metal. That is a good note. It is not a repair order. Someone still has to decide that is three jobs and not one, put hours against each, pull the part numbers, price it, and rewrite it in language a customer will approve over the phone. That restructuring is where the service writer's day actually goes. So the useful test of any voice product is not how accurately it hears speech, but whether what comes out the other end is line items or a paragraph with better grammar. A tool that only fixes transcription has moved the cost from the bay to the front counter.

The category worth evaluating parses a spoken note into a draft estimate rather than transcribing it: jobs separated, hours against each, parts as line items, in the system your writers already work in. Test it with your own audio, recorded in your own bay, on a job you did last week — a demo recorded in a quiet office proves nothing. Then ask what happens when the parse is wrong. A vendor who has thought about it describes an approval step and an easy edit; one who has not quotes an accuracy number. Otto is our attempt at this shape, built against a real two-location shop. A tech speaks the note; Otto writes the repair order and pushes it to TekMetric as a draft estimate for the service writer to approve. Nothing reaches a customer without a person signing off. If you are not on TekMetric, say so before anyone spends time on a trial. Pricing is quoted per shop rather than published, so that conversation starts at contact.

Work in this order. Write down the four things every tech note must contain — what you found, what it needs, which parts, and how long — and tape it inside the bay; some of what looks like an unwilling tech is an expectation nobody ever wrote down. Fill out your canned jobs — the prebuilt line items your shop management system already offers, labor and parts attached. Turn on phone dictation for the notes field. Then measure the number that decides whether you have a software problem at all: the minutes your writers spend each week turning tech notes into line items. Count it for two weeks, then pick a line — an hour a week is a defensible place to put it. Below that, keep your money; you have a habits problem, and a purchase will only annoy people. Well above it, that is a real recurring cost worth testing a parsing tool against.

Last updated 25 July 2026.