Why is my per-word rate dropping if my quality hasn't changed?

Because the comparison your client makes has changed. Machine translation plus a quick edit pass is cheap and often good enough for generic text, and that's the new benchmark your client uses. That's market pressure on the offer, and it says nothing about your translators' quality.

Should I start offering machine translation myself to compete?

That's a positioning decision, and it's yours to make. Some agencies deliberately move into the generic-text segment; we don't build that layer. Our work sits in the admin around the translation process. What we do see: agencies that keep competing on quality need the margin from the admin side to sustain that position.

Where is the margin actually recoverable, if not in the rate?

The margin sits in the hours between quote and delivery that nobody bills for: answering intake emails, prepping and counting files, sending status updates, filling in QA checklists. That work is separate from the translation itself; it's the administration around it. If AI takes over that layer as a draft a human checks, the unbillable hours drop without touching translation quality.

Will AI replace my translators?

No. We automate the administration around the translation: intake, file prep, status messages. The translation and the final edit stay entirely with your translators. The distinction between converting language and judging it stays human work; a model can't check a source text for cultural nuance or context that isn't written down in it.

Will our client messages start sounding like a bot?

They shouldn't. Every draft message the system produces is reviewed and only sent after a human reads it. Nothing goes to a client automatically. If a draft reads generic, that's a reason to adjust the instructions given to the model, not a reason to remove the review step.

Translation agency margins under AI pressure: what actually helps?

Q: What can Orellis concretely do here?

We start with a free 30-minute audit: where do the unbillable hours in your intake-to-delivery process actually go? That produces a written assessment with three opportunities, ranked by time saved. No system access needed for the audit. If a pilot follows, we build one workflow at a time, with a human reviewing every draft before it goes out the door.

An honest answer for translation agencies · Orellis

Per-word rates are under pressure because machine translation made the entry tier cheaper. That's market pressure, and it says nothing about your quality. The margin that disappears there is often recoverable in the admin next to it: intake, file prep, and status updates that currently cost unbilled time. That's exactly where operational AI makes a difference.

Framing this as an "embrace AI" or "ignore AI" question misses the distinction that matters: there are two different kinds of AI at play here. One makes the offer cheaper (machine translation, competitors quoting faster and lower). The other makes your admin cheaper (the unbillable hours between quote and delivery). The first squeezes your margin. The second can recover it. Agencies that don't separate these often end up solving the wrong problem.

Why your rate is under pressure

Large language models have pushed machine translation for generic text (manuals, product descriptions, internal communications) close to "good enough," at a fraction of the cost. Clients who once needed a translation agency for everything now compare your rate to a tool that does 90% of the work for free. That's a real market shift, not something to argue away.

For agencies competing on quality, nuance, and liability (legal translation, medical documentation, marketing copy that has to land in tone), a market still exists. But the margin in that market increasingly has to come from the operation, because the rate itself has less room than it did five years ago.

Market pressure on the offerMachine translation lowers the price expectation for generic text. This hits the rate you can charge, and no workflow fix solves this part.

The unbillable hoursIntake emails, counting and prepping files, status updates, QA checklists: work that never appears on the invoice but still costs the clock.

Where AI hands margin backWhen a draft of that admin step is produced by AI and checked by a human, the unbilled hours drop without touching translation quality.

What this looks like in practice

A translation agency with five translators often spends more coordination hours on a project than translation hours. A quote request needs answering, the source document needs counting and prepping, the right translator needs scheduling, the client needs updating mid-project, and a QA checklist needs signing off at delivery. None of those steps is the translation itself. All of them cost time that isn't separately billed.

What we build there: a draft reply to a quote request based on comparable past projects, automatic word counts and file prep, a draft status update at fixed points in the process. Everything stays a draft until a human reads it, adjusts where needed, and sends it. The time saved comes from removing the blank-page moment; the check stays in place.

Where AI does not help here, and we'll say so

The honest version of this has to name where AI isn't the answer, or even frames the problem wrong.

AI doesn't fix a pricing problem by automating harder

If the business model depends on the entry-level rate for generic translation, automating the admin around it won't change much. That's a positioning question: which segment do you serve? We solve the workflow question, and leave the positioning call to you.

We don't build the translation or final edit itself

Converting language and judging it for nuance, register, and cultural context is exactly the work your client hires a translator for. What we automate is the administration around it. The translation and the quality check on the text stay entirely with your translators.

When nobody has time to review drafts

The whole benefit of a draft step is that a human reads it before it goes out. If there's structurally nobody to do that, the answer isn't automation, since that just moves the unreviewed risk faster. A smaller starting point is the honest answer there.

This is an operational estimate. It isn't market research, and it doesn't guarantee a result. How much margin automating the admin recovers depends on your current process and volume. The audit maps that concretely for your situation, without assumptions upfront.

How Orellis approaches it

We start with a free 30-minute audit: a conversation about where the unbillable hours in your intake-to-delivery process actually go. No system access, no data needed upfront. That produces a written assessment with three opportunities, ranked by time saved, including where automating is not a good idea. A pilot then builds one workflow at a time, with a human reviewing every draft before it's sent.

We review everything that ships. This page was drafted with our own AI stack and reviewed by a human before it shipped. That review discipline is the product.

Map your own unbillable hours

Tell us what your intake-to-delivery process looks like now, and we'll tell you honestly where AI gives time back and where it doesn't. No system access or data needed, just a human reply.

Tell us where the time goes or book a free audit