Can you dictate a site report on a phone?

Yes. The person dictates at the end of the job, an AI transcribes the sound, sorts the text into findings, actions and next steps, then fills your template. The person reviews, corrects and signs before sending.

Is transcription reliable on a noisy site?

Not always. Accuracy drops near machinery. Numbers, names and references are the costliest errors: test on your own sites and have every number reviewed.

Do we have to change our tools to dictate reports?

In principle no. The report is filed in your storage and your CRM, and follow-up visits are added to your calendar.

What happens to the audio recording?

It is personal data of the person speaking. Decide how long to keep it, inform the team and check where transcription runs. Your legal counsel decides what applies to you.

From what volume is it worth it?

Below about ten reports a week, setup often exceeds the time saved. Above that, or when reports are typed in the evening or over the weekend, the numbers become favourable.

Site reports dictated by voice: from voice memo to document

What is a dictated report, exactly?

A dictated report is a service or site-visit report that the person on location speaks into their phone, and that an AI transcribes, sorts into three parts (findings, actions, next steps) and drops into your document template. The person reviews, corrects and signs; the report then goes to the client. Dictation replaces evening typing, not responsibility for the content.

The starting problem is familiar to anyone running a field business: a technician, site manager or surveyor finishes the day, then types reports from memory, late. The client waits, the internal file stays incomplete, and the detail that mattered at 3 pm is gone by 7 pm. The studio's automation approach starts from that daily friction. This article answers other questions: what dictation really changes, what makes it fail, and how to test it without committing to a project.

What existing articles say, and what is missing

We read three well-ranked pages on the topic: a field report from a construction software vendor, a priced guide from an automation provider, and a transcription vendor's page. They agree on three points: speaking is faster than typing, noise degrades transcription, and a trade vocabulary improves results. Two limits stand out.

Almost every figure comes from a seller. The software vendor claims a defect description takes 30 to 40 seconds to dictate against 90 to 120 seconds to type, without citing a measurement. The provider's guide quotes 90 to 95% accuracy with a glossary and a lapel microphone, about 80% near machinery, and a report taking 60 to 90 minutes cut to 5 minutes of review. The transcription vendor says "up to 99%" on clear recordings. These are commercial orders of magnitude, not independent measurements.
Human control is mentioned, never described. "Review" appears everywhere, without saying who reviews, what, with which warning signals, or what happens when the transcript gets a number wrong.

Keep the direction, not the percentages. Your real gain is measured on your own dictations, in your own noise conditions.

From voice memo to sent document: the steps

Dictation on site. The person speaks into the phone at the end of the job, while it is fresh. They follow a simple thread: what I found, what I did, what remains.
Faithful transcription. Sound becomes text, with no rewording. Trade terms (references, brands, part names) are supplied to the tool as a glossary.
Structuring. The text is sorted into findings, actions, next steps. A follow-up visit goes into the calendar.
Formatting. The text enters your report template, with photos filed in the right folder.
Review and signature. The person who dictated corrects and approves. Nothing leaves before that.
Sending and filing. The report reaches the client and joins the internal file, the same day.

Step 3 is the decisive one. Structured text can be reviewed in two minutes; a raw transcript of five minutes of speech cannot. Structure is what makes review realistic, and therefore control effective.

The dictation thread to give the team

Free dictation produces a story told out of order. A thread of three questions, learned in a minute, is enough to get text that is already almost structured.

What did I find? The state on arrival, what is wrong, with numbers said slowly and only once.
What did I do? The actions in order, the parts fitted or replaced, named by reference.
What remains to be done? What needs planning, for whom, and before which date if known.

Two habits help more than any setting: state the client's name and the location at the start, and spell out unusual references. A technician who spells a part reference avoids the most frequent correction at review time.

What the AI transcribes well, and what it gets wrong

Dictated element	Reliability	What to do
Ordinary sentences, describing a finding	Good in calm conditions	Quick review
Trade terms and brands	Average without a glossary, better with one	Supply and maintain a term list
Numbers, measurements, references	The costliest errors	Highlight every number for targeted review
Names of people and places	Fragile	Take names from the job sheet
Speech near machinery or a noisy site	Clearly degraded	Dictate away from the noise, or afterwards from the vehicle
Accents and mixed languages	Variable	Test with the team's real voices

The rule that follows: the AI proposes, the technician decides on every number. A misheard amount, dimension or reference slips into a clean, well-written report, which makes it more believable, and so more dangerous, than a clumsy draft.

The conditions that make it work or fail

Noise. A published field report puts the accuracy drop around 70 dB, a level reached near a concrete mixer or a cutting saw. That is a vendor figure with no published test: do not treat it as a threshold, test on your own sites.
Connectivity. On a site with no signal, dictation must work offline and upload later. Ask before choosing a tool.
Voice recordings. A colleague's audio is personal data. Tell the team, decide how long you keep the sound once the text is approved (the cautious answer: as briefly as possible) and check where transcription runs. For the Swiss framework, see our article on the Swiss FADP and data sent to a model. This is not legal advice: your counsel decides what applies.
The dictation habit. Talking to a phone in front of a client or colleague bothers some people. Provide a three-sentence dictation thread and let the team choose the moment, on site or from the vehicle.

How to test before you launch

Take ten reports already written: five simple jobs, five complicated ones.
Have each job re-dictated by the person who did it, on their phone, in real conditions.
Compare the resulting text with the original report, checking every number, name and reference.
Classify the gaps: misheard word, unknown trade term, something the person forgot, error already present in the original report.
Time the dictation and the review, not just the transcription. The person's time is what costs.

This test often shows that the reports typed in the evening already contained omissions.

A calculation to redo with your figures

Here is an example with figures set for illustration, not measurements. Four technicians each produce ten reports a week. Each report takes twenty minutes to type in the evening. That is thirteen hours a week. If dictation and review take eight minutes per report, the total drops to about five and a half hours, so seven and a half hours saved per week.

Two corrections apply. Subtract the time to maintain the glossary and the report template. Add what the calculation forgets: the client who gets the report the same day, the detail no longer forgotten, the end of reports written over the weekend. Replace our assumptions with yours.

When this is not the right choice

You produce few reports. Below about ten a week, setup often exceeds the time saved.
Your reports are checklists. If most of it is ticking boxes, an entry app is simpler than dictation.
The site is too loud to speak and nobody can dictate away from it. Keep typing, or dictate afterwards.
The report commits the company heavily. For a finding with contractual weight, dictation can prepare the draft; the final content belongs to the person who commits the company and, if needed, to your counsel.
Nobody will review. Without review, you send transcription errors to clients with confidence.

If your difficulty is producing the quote that follows the job, that is another subject: see the safeguards for a generated quote.

What we do not promise

We promise neither error-free transcription nor a quantified time saving before we have heard your dictations. We promise a method: guided dictation, structured text, flagged numbers, review by the person who did the job, and a test on your own reports before any rollout.