Every major technology shift creates winners and losers. AI is no different. Companies built around producing more words will come under pressure. Companies built around trusted knowledge, governance and enterprise workflows will become more valuable. Our strategy is built on being firmly in the second category.
The story elsewhere is that AI killed translation. For thirty years the business sold words in, words out, priced by volume, and AI has taken the cost of that raw work to nearly zero. An industry priced by the word has no floor once the word is free. The traditional per-word translation model is becoming increasingly commoditised. Our strategy is not to compete on the price of words, but on the value of trusted, enterprise-grade outcomes.
Here is the part the obituaries miss: the demand did not die with the price. It moved up. And the place it moved to is a bigger, better business than the one that is ending.
What enterprises actually need
Watch what’s happening in Japan. From April 2025, every Tokyo Stock Exchange Prime Market company, roughly 1,600 of the country’s largest enterprises, must disclose in English at the same moment as Japanese. Not translated eventually. Simultaneously.
Nobody with something at stake wants a translation. They want a result they can put their name to: terminology that matches every prior filing, layout that survives the conversion, language a regulator will accept, and a name on the result when it goes out the door in hours, not weeks. That is not a cheaper version of the old product. It is a different product, bought by a different part of the enterprise: not procurement, but the people who answer for compliance, risk, and speed.
Meanwhile the supply side is collapsing on schedule. The specialist translators who served regulated industries are aging out, and the pipeline behind them is thin. Rising demand, shrinking supply, zero tolerance for error. That is not a dying industry that is a market being rebuilt around whoever solves it properly.
Why this is a real business, not a hope
Every leadership team I meet says the same two things about AI in the same breath: it is the defining shift of their careers, and most of what they have tried has not fulfilled the promise. The independent research agrees the large majority of enterprise AI efforts never reach real production value. The models are not the problem; they are remarkable. The breakage happens at the last step, where AI output meets something real, a filing, a claim, a price and at machine speed no one can check every output before it ships.
This is exactly the discipline our work has run on for years: let the machine do the work, and never release what a human standard has not cleared. We did not learn that from a whitepaper. We built a business on it, in the one domain where getting it wrong answers to a regulator, not a proofreader. The rest of the market is now scrambling to buy that capability. We already run it, in production, for clients who cannot afford to be wrong.
Why we hold the position
Raw translation is a commodity. Verified, accountable content is not, because of what sits underneath it: more than a decade of human-approved decisions, gathered into a single store that every output is checked against before it ships. On top of that we run our own models, specialist systems for heavily regulated sectors like financial services, and one whose only job is to judge whether another model’s output is good enough to release.
That store of judgment is the moat. A competitor can rent a capable model this afternoon. It cannot rent the decade of approved decisions that decides what is allowed to go out.
And it travels. The same method carries a speaker into another language in their own voice on video, and reaches upward toward governing what our clients’ own AI is allowed to produce. Translation is not the ceiling. It is where we proved the method everything else is built on.
What comes next
The next phase of this industry won’t be humans using AI tools. It will be models and agents doing the work directly with humans designing, governing and assuring the system rather than working inside it. A customer won’t submit a document for translation; they’ll connect their disclosure, legal or communication workflow to a platform where specialised agents draft, check terminology against every prior filing, verify compliance, format, and flag only what needs human judgement.
When agents and models do the work, the old pricing logic breaks with it. Per-word pricing measures human effort and human effort is no longer what customers are buying. What they’re buying is intelligence and trust: the expertise of the system doing the work, and the assurance that its output can be relied on in a regulated world. Those are different things, with different value, because raw intelligence is becoming abundant, while verified trust is not.
The industry that’s ending was priced by the word. The one emerging is priced by intelligence and trust, and we’re building the platform native to it (more on that soon!!).
David Sowerby
Co-Chief Executive Officer