Translators: measuring hours per thousand words to price the next job
· 6 min read
Translation is usually priced per word, which is convenient for quoting and useless for knowing whether the work was worth doing. Two jobs of five thousand words at the same rate can take five hours or fifteen, and only one of them was profitable.
The figure that resolves this is hours per thousand words, and it takes a few weeks of measurement to establish.
Why the per-word rate hides the answer
A per-word rate converts effort into a number at quoting time and then never checks the assumption. The variables that actually determine effort are invisible to it:
- Subject matter you know against subject matter you have to research
- Clean source text against a badly written or badly scanned original
- Formatting that carries across against formatting you have to rebuild
- A client with a glossary against a client who has opinions but no glossary
- Revisions, and whether they were scoped
None of that appears in a word count. All of it appears in your week.
The calculation
hours per 1000 words = total hours worked on the job / (words / 1000)
effective hourly rate = fee / total hours worked
Total hours worked means everything: research, the actual translation, revision, formatting, correspondence about terminology, and the final check. Not just the hours that felt like translating.
Do this for a dozen jobs and patterns appear quickly. You will usually find that your rate per word is fine for some categories of work and quietly loss-making for others, and that the loss-making ones are the ones you say yes to because they seem similar to the profitable ones.
Getting an honest figure for hours
The denominator is the hard part, because translation work is fragmentary. An hour here, forty minutes there, a stretch in the evening. Reconstructing it on Friday produces a figure that omits most of the small pieces, which is precisely where the time goes.
Automatic tracking on the Mac fixes this specifically. It records which application was in front and for how long, without you starting or stopping anything. For translation work the categories map cleanly: the translation environment or word processor, the browser for research and terminology, mail for client correspondence, a PDF reader for the source.
That last split is worth having on its own. Most translators have never measured how much of a job is research rather than translation, and for technical work it is often the larger share.
What the numbers change
Which work to take. Once you know that legal work runs at four hours per thousand words and marketing copy at two, a per-word rate that treats them identically is obviously wrong. Either the rate differs by category or you decline the slow one.
How to quote a new client. A sample of the source, checked against your own figures for similar material, gives you an estimate with arithmetic behind it. How to estimate hours for a quote covers building that from records.
Whether tooling pays for itself. Translation memory, terminology management and a better source format all cost money or time to set up. The only way to know whether they helped is to have measured before and after. A drop in hours per thousand words is a real return; a feeling that things are smoother is not.
When a client is expensive. A client whose work reads well and arrives clean is worth more per word than one who sends scanned PDFs and revises twice. The rate can reflect that, and it cannot until you have measured it.
Where Punchcard fits
Punchcard sits in the menu bar, records which application is frontmost by name, and prints a receipt of the day at your closing time: one line per application, with a total.
It reads no window titles and no document names, which for translation matters more than it might seem. Source documents are frequently confidential, and their file names often identify the client or the subject. A tool recording window titles would be building a list of what you are working on for whom; this one cannot, because it never looks.
There is no account and no networking code, so nothing about your work leaves the Mac. Why most time trackers want your window titles covers what the alternative involves.
What it does not do is know which job a block of work belonged to. There are no projects or tags. Assigning the day’s blocks to jobs is a two-minute step you do daily with the receipt in front of you, and doing it daily rather than weekly is the difference between a real figure and a remembered one.
Setting it up
Week one. Just collect. Do not change how you work; you want a baseline of a normal week.
From week two. Each evening, note which blocks belonged to which job. Two minutes.
At the end of each job. Total the hours, divide by thousands of words, note the figure and the type of material.
After a dozen jobs. Sort by hours per thousand words. The pattern in that list is your pricing structure, arrived at empirically.
The Sunday roll covers keeping the weekly habit going.
Questions
Should I switch to hourly billing? Not necessarily. Clients generally prefer per-word because it is predictable, and predictability is worth something. The point is to set the per-word rate from measured hours rather than from habit.
What about machine translation post-editing? That is exactly the case where the per-word rate misleads most, since effort varies enormously with the quality of the raw output. Measuring hours per thousand words per client is the only way to price it sensibly.
My work is split across two machines. Each Mac keeps its own record, so you would total both. Tracking time across two Macs covers it.
Does this account for thinking time away from the desk? No, and for translation that is a real omission. A difficult passage worked out on a walk is genuine work and invisible to any desk-based record. Note it by hand.