How DeepL became a $2B AI translation company by moving fast

Executive overview

Most companies face a language barrier when expanding across European markets — hiring bilingual staff or doubling sales headcount to address each new region. DeepL solved this with a specialized AI translation model built before AI was mainstream.

Jarek Kutylowski launched with a free, no-login product to validate demand, shifted to B2B monetization once signals were clear, and reached profitability by 2018 through a product-led growth model.

The core insight: being first to market with a working AI translation solution created a word-of-mouth flywheel that later entrants could not replicate.

The problem and early bet

• Language barriers cost businesses: the global translation industry is ~$60B.
• European companies selling across borders must hire local speakers or risk losing customers who want to be addressed in their own language.
• In 2016–2017 it became clear in academic circles that neural networks could transform translation.
• Kutylowski had personal experience with language barriers, having moved from Poland to Germany as a child speaking no German.
• The founding team knew the problem was large; the unknown was whether the technology would be good enough to beat incumbents.

Go-to-market: free first, monetise second

• Launched a bare-bones free service with no login — lowest possible friction to adoption.
• Used raw usage numbers as validation signal rather than qualitative research.
• Introduced paid features and paywalls in 2018 only after confirming willingness to pay.
• Early decisions were gut-driven; quantitative data replaced intuition as the product scaled.
• Shifted focus toward B2B and enterprise buyers once the consumer base was established.

Speed as the defining advantage

• Being first with a neural-network translation product created the early-adopter base that spread via word of mouth.
• Kutylowski's view: arriving six months to a year later would likely have changed the outcome entirely.
• The company reached profitability in 2018 — a direct result of PLG growth requiring minimal sales headcount.
• The main regret as a first-time CEO was moving too slowly; a second attempt would focus on compressing decision cycles.

Managing change at high growth

• Fast growth means nothing stays the same year to year — teams experience constant change.
• Communicating why change is happening is as important as executing the change itself.
• Helping employees understand the rationale reduces friction and keeps the organization aligned.

Specialized AI vs. general-purpose models

• Generalist models (e.g., large language models) can handle a wide variety of tasks but often underperform on specific business use cases.
• Specialized models are quality-tested for a narrow problem, delivering consistent output at lower cost and higher speed.
• For translation at scale — across a wide range of inputs, not just one-off emails — consistency matters more than breadth.
• Specialized solutions typically come with the full product layer: UI, integration, and user experience, not just the model.

How to adopt AI in a business

• Start with the business problem, not the technology.
• Identify where your company underperforms or moves slowly, then look for AI solutions to that specific problem.
• Avoid applying AI broadly for its own sake — optimize what is actually worth optimizing.