Localization and translation agency

The simple explanation

Every business has jobs it needs done well but doesn't want to hire for: marketing, design, bookkeeping, ads, content. An agency packages one of those jobs into a service, sells it to multiple clients, and delivers it reliably. It is the most direct business model that exists: find someone with a problem, solve it, invoice them. That is why it is usually the fastest path to first revenue.

A simple hypothetical example

Illustrative — invented to show the shape of the Services & Agencies pattern. No real company is named, and no figure in it is data. The real, sourced companies for this model are on the Examples tab.

A landscaper is booked solid in summer and empty in winter, and their website looks like 2009. You redesign it, set up their review flow, and run a small local campaign for a monthly fee. Their phone rings more; your invoice is a fraction of the extra revenue. Word spreads to the plumber and the roofer, and you have an agency.

A closer look at localization and translation agency

You sell a word and you buy a word, and almost everything interesting happens in the gap. Illustrative only: a 20,000-word product manual quoted at $0.22 a word bills $4,400. A freelance translator at $0.10 takes $2,000, an independent reviewer at $0.04 takes $800, and file engineering and project management take roughly $400, leaving about $1,200, just over a quarter, before you have paid yourself. That is a normal, survivable margin, and it is why the job is scheduling and vendor management rather than translating. The trap is that the tooling you must use erodes your own revenue by design. A translation memory stores every segment you have ever delivered for that client, and standard industry pricing discounts repeats. Take that same manual next year with 12,000 of its words matching memory at a quarter rate: 8,000 words at $0.22 plus 12,000 at $0.055 bills $2,420, not $4,400. Your best, longest-standing accounts get cheaper every year. Nobody tells you that at the start.

Machine output then moved the floor under the whole price. Most commercial work now arrives as machine translation post-editing, where the software drafts and a human fixes, and it is paid at a fraction of a from-scratch rate. The numbers in the cards above are the two halves of the same story: RWS, the largest listed pure-play, wrote off £88.0m of goodwill in 2025 citing weaker performance in core localization, while TransPerfect grew 7% to $1.32 billion with its co-CEO openly noting that more translation workflows are becoming automatable and its technology licensing line up 18%. Volume of content is rising. The price of the underlying unit is not. An agency that plans to grow by selling more words at last decade's rate is planning against both data points.

So the margin has migrated to the places where being wrong is expensive. Clinical trial documentation, medical device instructions for use, patents, pharmacovigilance, financial filings, contracts, and anything requiring a certified or sworn translator all carry liability that a raw machine draft cannot absorb, which is why they still carry human rates and process certifications. Live interpreting (courts, hospitals, conferences) is a different business with different economics and has not deflated the same way. Choose one such vertical and learn its file formats and its regulator; the generalist who will translate anything into anything is competing directly with a free browser feature.

Two operating realities decide whether the agency survives its own growth. The first is cash: linguists expect payment on roughly 30 days and enterprise clients pay on 60 or more, so every new account you win is a loan you extend, and a big first project from a slow payer is how well-run agencies die solvent on paper. The second is that you cannot personally check the work. You will approve Japanese you cannot read, which means your real product is a vetting-and-review system: test translations, a second independent linguist on anything regulated, and a glossary the client signs off before the first job. A single mistranslated dosage or contract clause ends the account and possibly the company.

How money moves through this model

Who pays: Businesses that value the outcome more than the fee

What they pay for: An outcome they lack the time, skill, or desire to produce in-house

What creates profit: Fees minus the labor cost of delivery, yours at first and a team's later

• Customer
• Offer
• Localization
• Costs
• Profit

What makes this model hard

The honest difficulty: you are the product. Early on, every dollar is bought with your hours, and growth means either working more or hiring and managing people, a completely different skill from delivering the work. Client concentration is the silent killer: two big clients feels like success until one leaves.