The four steps, and the one people skip
Speak. Read. Correct. Save. The third step is where teams quietly fail. Someone dictates a note in a driveway, glances at the screen, sees roughly the right shape of words, and submits. Three weeks later the estimate says twenty-two squares because the recogniser heard 'twenty-two' when they said 'thirty-two'.
Building the review step into the habit costs about eight seconds. Not building it in means every downstream number carries an invisible confidence problem.
Where transcription reliably fails
- Proper nouns: customer surnames, street names, manufacturer names.
- Alphanumerics: model numbers, part codes, unit sizes, serials.
- Numbers spoken quickly, especially teens versus tens.
- Trade vocabulary the general-purpose recogniser has never heard.
- Noisy environments: wind, traffic, running equipment, other voices.
How to dictate so the text comes out usable
- One observation per burst. Tap, say the sentence, stop, check.
- Spell surnames and part numbers, or type those two fields by hand.
- Say units explicitly: 'thirty two squares', 'eight twelve pitch'.
- Turn away from the wind and the compressor before you start.
- Dictate the body; type the numbers if precision matters more than speed.
Who is responsible for the saved note
The person who pressed save. Not the recogniser, not the software. This is the same standard as a typed note, and it is the reason the workflow deliberately keeps a human between the microphone and the record.
If the note later turns out to be wrong, the correct response is a new note correcting it, with its own timestamp and author, rather than quietly rewriting history so nobody can tell what was believed when.