Case Study: Zendesk achieves 75% lower translation costs and scalable multilingual support with Lilt

A Hybrid Human + Machine Translation Approach Keeps Zendesk’s Customers Happy

Zendesk, a global customer service software company, faced the challenge of translating a large, continuously updated library of English support articles into five target languages without excessive cost or manual administration. They needed a scalable, customizable solution with API access and the ability to handle varying article priorities, so Zendesk chose a hybrid human+machine approach and selected Lilt (adaptive neural machine translation, a centralized translation memory, and an API) to meet those requirements.

Lilt implemented an end-to-end pipeline that connects to Zendesk’s CMS, uses a KPI filter to route high‑priority content to human translators (augmented by Lilt’s adaptive NMT) and lower‑priority content to raw MT, while real‑time learning from the centralized TM improved quality and throughput. The result: over 75% cost reduction, Lilt’s neural MT outperformed phrase‑based MT in 44% of test segments, faster translation cycles, improved translation quality, and immediate uptake of translated content by users—enabling Zendesk to scale multilingual support effectively.

Zendesk

Melissa Burch

Manager of Online Support