As a Brit I’m innately lazy with languages, but in the last 6 months this absence of linguistic ability is starting to feel increasingly vindicated. Just within big tech we’ve seen Google announce real-time translation in Google Meets, Meta’s latest AR Ray-Ban’s and Apple’s latest Airpods both embedding live translation. Over in Small Tech start-ups like DeepL, Palabra, Elevenlabs, Sesame, Lokalise, Sanas are working across various vectors to enable the shift from translation as post-processing (Google Translate, Trados) to translation as more ambient invisible infrastructure across states (docs, interfaces, voice, demo flows etc.). You can argue about preferred UI/UX and latency, but the trend seems clear, the language component of localisation is commoditising. The interesting question is what are the second order consequences?
- Increased Social Status. I suspect my suppressed jealousy for my effortlessly multi-lingual friends only increases. In a world where the economic premium for being multi-lingual evaporates, having such a frustratingly difficult but completely pointless skill only increases in social status.
- Cultural Context Becomes The Blocker. OpenAI is still hiring for localisation managers. At least in the near-term, the human skill premium for localising experiences embedded with the necessary tacit cultural fluency and market understanding will remain. It’s worth emphasising that localisation is a multi-faceted problem (language, culture, regulation, reimbursement, integrations etc.), which can be especially acute in the B2B context.
- Emergent Scale. Mr Beast auto-translated his Youtube back catalogue to world domination and Vinted’s use of real-time in-app translation perhaps acts as a bellwether for opportunities to generate overlooked pockets of homogenised scale. From an EU perspective if a key localisation barrier is now falling, does this enable more opportunities for scale parity with the likes of the US? A topic we’re particularly interested w.r.t vertical AI.
- New Labour Arbs. Does the English premium that powered Jack Ma to success disappear or democratise? Especially when humans augmented with LLMs creates a route to greater cognitive parity between the west and lower income regions. Looking at OpenAI’s latest GDPval the opportunities write themselves. Already large BPOs are using Sanas’ software to ‘neutralise’ Indian call agent accents, why not give them Cluely too. Start-ups like Asterix Health are innovating to collapse global access to clinicians and Supersonik provides a glimpse at a future where these capabilities are combined with our emerging capabilities to “sell work”.
New Labour & Geo Arbs?
For founders, the lay-up of emergent scale and new labour arbs is worth pause. For instance, the consensus narrative is that in the AI age BPOs are cooked. Their labour cost + mark-up business model is undoubtedly threatened by AI and stocks like TASK, CNXC and TEP.PA have been torched since c.2022. But at least on topline these incumbents have proven relatively stable and now loosely congregate on a narrative of moving up the value chain to “AI-resistant” work, which tends to be more vertical / specialised (finance, healthcare, insurance etc.) and augmenting their labor with AI tools. Even Sedgwick, subject to over a dozen AI native TPAs seems to be holding firm, for now.
As a BPO owning such workflows may be increasingly valuable given the excitement around scaling RL and the need for environments which can teach the models, navigate longer context tests, use of tools, collaborative environments etc. Scale AI was a wrapper on the Philippines after all. As Mercor explains, the economy will scale as humans teach AI agents to automate every workflow. RL environments for every job, tool, and task in the economy with human labour required to build, verify, and refine them. Does Europe’s localisation hell then become training data, not friction? One for the researchers, but if it’s Mira’s revealed preference then it’s good enough for me. High flying Fyxer provides another interesting example, before starting in 2023 the founders had run an executive assistant agency since 2016. Equipped with 500,000 hours of time-tracking data on an EAs workflow helped them in steam rolling the market with a more agentic solution.
As an ecosystem the EU suffers from the mind virus of Rocket Internet, over extrapolating on hot US series A’s. But perhaps we’re in an era which creates a new surface area for innovation, where the unique market structure of Europe can be a help, rather than a hindrance.
If you’re tinkering around these themes please reach out on jamie@triplepoint.vc