On a Friday afternoon I promised a client their transcripts would be ready by the following Friday. Three files were waiting in my inbox on Monday morning: a 45-minute interview with one clear speaker, and a three-hour panel discussion recorded in a room that sounded like a washing machine, with four people talking over each other. One hour of audio takes about four hours to transcribe — that is the rule everyone quotes, the 4-to-1 rule — so I did the headline maths in my head, nodded, and promised. By Wednesday I knew the promise was fiction.
The 4-to-1 rule is true on average and wrong for your specific files, because it says nothing about who is speaking or how the recording sounds. A clean one-to-one interview and a noisy four-person panel do not cost the same, and blending them into one average hides exactly the file that decides your deadline. The honest way to price the job is file by file, with multipliers. Manual transcription starts at four times the audio length. Two or three speakers add a quarter. Four or more, or overlapping voices, add half. Average quality adds a quarter, poor quality adds half. Multiply them together and the picture changes completely: one hour of poor-quality audio with four overlapping speakers is not four hours of work. It is nine.
Run my Monday inbox through that lens and the week rearranges itself. The interview — 45 minutes, clear, one speaker, manual — is three hours. The panel — three hours, poor quality, four-plus overlapping speakers — is 3 × 4.0 × 1.5 × 1.5, which is 27 hours. Together that is 30 hours of work for 3 hours and 45 minutes of audio, and the panel alone is nine-tenths of it. At six working hours a day, Monday to Friday, that is five full workdays — and the Friday I promised was already gone before I started, because the plan needs the whole week and finishes on Friday evening, not Friday morning.
The other mistake I made was counting calendar days instead of working days. Weekends do not transcribe themselves, and the last day is usually a partial session rather than a full one. A finish date that ignores that is fiction with a date on it. The fix is to split the total work into daily sessions of your real working hours, walk forward from your start date counting only the days you actually work, and let the final day land as a partial session when the work does not fill it. That is a plan you can quote to a client, because it survives contact with your calendar.
There is a faster route, and it is worth pricing honestly too. An AI draft with a genuine human review pass runs at about one and a half times the audio length instead of four — but the saving shrinks on difficult audio, because heavy overlap and poor recordings need much more correction. On my panel, the AI route would still be 3 × 1.5 × 1.5 × 1.5, which is just over ten hours. Faster, yes. Still the file that decides the deadline.
I run the numbers now with this free transcription time planner before I promise anything: list each file with its length, speakers and quality, pick the method, set the working hours and days, and it hands back the total work hours, the workdays needed, the finish date and a day-by-day session table: https://toolbrewlab.com/transcription-time-planner/
The client got their transcripts on Friday, a week later than my first promise and exactly when the plan said they would. The 4-to-1 rule is a fine starting point. A file-by-file plan is what keeps the promise.
Top comments (0)