The translation tax: when half the work is understanding what the client meant

Marina Shmayger, who founded Structed, once tracked a month of her consulting hours. She charged £60 an hour. Her real rate, she found, was about £20.

The work clients paid for took an hour. Around it sat the unpaid part: messages to work out what the client needed, writing the right questions, turning the answers into something usable, then the follow-ups. She raised her rates several times. It didn't help, because the unpaid part grew with every client.

It runs in both directions

The same thing happened when she was the client. She tried to hire a scriptwriter and got a wall of questions in a messenger. With an editor, a pleasant forty-minute call ended with her realising she'd have to write it all down anyway. Each time she did the work herself, because explaining it cost more than doing it.

So the freelancer loses hours and the client loses patience, and the work still comes out wrong.

Why a form doesn't fix it

The obvious answer is a questionnaire. People skip it, or fill it with what they think you want to hear. A form can't ask a follow-up when an answer is vague, and it can't tell that "something modern" means three different things to three clients.

An expert can. That is why the translation is expensive: it needs judgement, and judgement doesn't scale.

What an AI interview takes on

The repeated part. The agent asks one question at a time in the client's own words, follows up when an answer is vague, records "I don't know" instead of pushing, and hands the expert structured requirements instead of a transcript.

What it should not take on is the judgement about which questions matter. That stays with the expert, who edits what the agent asks without a developer. Otherwise you've automated the questionnaire, and you're back where you started.