Capture roughly in the field with voice and camera, structure it later at a keyboard.
Do now: Record a two-minute voice note about a real task, then turn it into a structured record.
Say the numbers and names slowly and twice. Those are what transcription gets wrong.Most people's most useful raw material — site visits, conversations, observations, receipts, whiteboards — is generated away from a keyboard. Capturing it properly at the moment is worth more than any amount of processing afterwards.
Voice capture that survives transcription
Speak in short, complete sentences and state structure aloud: 'Decision one.' 'Action for Peter.' 'Question, unresolved.' You are giving the later processing something to work with, and it takes no extra time.
Say numbers, names and places slowly, and repeat them. These are exactly what transcription gets wrong, and a repeated figure gives you a second chance to catch the error rather than discovering it in a report.
Start every recording with the date, the place and who is present. Later, that opening sentence is what turns an unidentifiable audio file into a record, and it costs five seconds.
State the date, place and people first. Say numbers twice. That is most of the skill.
Camera capture, and what not to point it at
Photographing a whiteboard, a form, a meter reading, a damaged part or a receipt is faster and more accurate than typing it, and text can be extracted later. Photograph straight on, in the best light available, and get closer rather than zooming.
Take two photographs of anything containing numbers, from slightly different angles, so a misread can be checked. Confirm the extracted values against the image before using them, because numbers from photographs are the values most likely wrong and least likely questioned.
Then the boundary. Do not photograph other people without permission, other people's children at all, screens containing personal data, documents belonging to someone else, or anything in a clinical or private setting. Check the background before you take it, not after you have uploaded it — a photograph of a form on a desk often includes three other forms.
Capture rough, structure later
Do not try to produce finished work on a phone in the field. Capture completely and untidily, then structure at a keyboard where you can see the whole thing. Trying to write neatly on a phone in a crowded room produces less information, not more.
Batch the processing. One structured session per day converting a set of voice notes and photographs is far more efficient than processing each one as it arrives, and on limited data it costs a fraction as much.
Keep the originals. The voice note and the photograph are the record; the structured version is an interpretation. Where a decision or a quantity matters, being able to return to the original is what makes the record trustworthy.
The recording is the record. The tidy version is an interpretation of it.