AI for translation agencies
Prices per word are under pressure, but the work of managing each project is not getting smaller. We build AI agents that take on the email, files, chasing and paperwork around your TMS, with a project manager approving each step.
PDFs, scans, spreadsheets, web links.
A project manager downloads the files, runs the word count and match analysis, applies the client's rate card, checks the deadline is possible and writes the quote. That is unpaid work, and it may not turn into a job.
For each project, PMs email several freelancers, wait, offer the job to the next person when someone declines, then chase the delivery. Availability, rates and subject specialties often live in a spreadsheet or in someone's head.
The 2025 European Language Industry Survey (ELIS) reported, for the first time, that most freelance translators saw price drops despite inflation (ELIA). When rates fall, the cost of managing a project becomes the part you can still control.
Each job creates a purchase order to the linguist, an invoice back that must be matched to it, and an invoice to the client. Agencies that pay on time keep the best freelancers, but matching and chasing take hours.
We do not sell another platform. You already have a TMS (XTRF, Plunet or similar) and CAT tools. The repetitive work sits between them, and that is where agents fit.
What the agent does
What it never does
The agent reads the incoming email and attachments, identifies language pairs, subject area and deadline, triggers the file analysis, applies the client's rate card and minimum fees, and drafts the quote and project record.
The PM checks and sends. The agent then follows up on quotes the client has not answered.
The agent ranks linguists by language pair, subject, rate, past quality scores and availability, sends offers in order, handles accept and decline replies and books the job. It flags late deliveries to the PM early.
Halfway through a job the agent asks each linguist for words done, and flags anyone behind so the PM can act while there is still time.
It works from progress messages, not from the text itself. Worked example: how to check a translation is on track before the deadline.
The agent reads freelancer invoices from email, matches them to purchase orders, flags mismatches and queues approved payments. When a project is delivered it sends the client invoice and reminders to late payers, synced to your accounting.
Linguists raise questions in the middle of a job. The agent collects them, removes duplicates, sends the client one list, and posts the answers back to everyone on the job.
Illustration, not a client result
A worked example with assumed numbers. Suppose an agency sends 150 quotes a month and each takes a project manager 30 minutes of file handling and analysis. That is 75 hours a month, close to half of a full-time PM, before a single job is won. If an agent does the first draft and the PM only reviews, most of that time comes back, and quotes go out sooner.
These numbers are assumptions to show the arithmetic. On the discovery call we ask for yours.
Illustration with assumed numbers, not a client result.
Admin and operations first. By default, agents handle metadata and admin (emails, deadlines, rates, invoices), not the text of client documents, unless the client allows it.
No training on your client's data, and data kept in the region you need (for example for GDPR).
A person approves quotes, bookings and payments. Every agent action is logged, which supports documented processes such as ISO 17100 and ISO 27001 audits.
Linguists stay central. This is about removing admin, not replacing translators.
Start small. One workflow, a weekly demo and a fixed-price quote before anything is built.
A free call to understand your business and where the time goes.
We map the workflow, measure the baseline and deliver a prioritized roadmap with a fixed-price quote.
We build the first agent, show progress every week and launch with human approval switched on.
Once it works, we add agents and keep everything monitored and tuned.
Those systems have their own automation, and we will not try to replace them. The gap is usually the work around them: email intake, chasing, invoice matching. Part of the value can be setting up what you already pay for.
Not by default. We design agents around metadata and admin, and only process content if your contracts allow it, with no training on client data.
It drafts and proposes, using your rate card and linguist data. A PM approves each quote and booking.
The agent contacts them the way a coordinator would, and the aim is faster, clearer offers and quicker payment. We keep messages short and in your voice.
Every engagement starts with a free discovery call and an assessment that ends in a fixed-price quote for the first agent, so you know the cost before anything is built.
Usually quoting, because it is the step PMs spend the most unpaid time on, or deadline tracking if lateness is the bigger pain.
Worked examples (illustrations, not client results):
Other industries: contractors and home services, editors and publishers, short-term rental hosts and managers.
Book a free discovery call. We will map one workflow, tell you what an agent could take over, and be honest if your TMS can already do it.
Book a free discovery call