How Unbabel reduced their number of management emails by 55% with n8n

Unbabel, a leading language operations platform, used n8n to build an advanced peak management system that enables the team to quickly engage more editors to validate automated translations. The system was built in just a couple of weeks – from conception through to production – to run alongside the main product. The system built with n8n helped Unbabel reduce by half the number of Communities with significant volumes of time-consuming translations (1 hour +)

Delphine Pessoa is the Senior Product Data Scientist at Unbabel, an AI-driven, human-refined translation solution delivering automated, quality translations to small, mid-size, and enterprise businesses that include Microsoft, Logitech, and easyJet.

Unbabel is able to automate translations to a high degree of quality, but it also layers in editors and professional translators to improve meaning, context, or localization when appropriate – not only creating greater understanding, but also improving the underlying algorithm.

And that’s where the main challenge surfaced.

How did Unbabel cross paths with n8n?

Initially, n8n was introduced to Unbabel by André Silva, Senior LangOps Engineering Lead, who by nature is driven by innovation and new technologies. He was considering integrating automation tools at Unbabel after reading a Reddit thread titled “Automating yourself out of a job”.

After a quick installation, André launched n8n locally to automate a few basic tasks such as sending sprint planning notifications via Slack, and generating document templates.

“We considered using Node-RED and Zapier, however, the former didn’t have an appealing interface, and the latter had a strict limit on the number of steps we could generate,” said Silva.

No available engineers? No need with n8n!

When deciding which initiatives to push forward, Delphine’s team discussed working on setting up a peak management system that would detect delays in translations, by notifying Community Managers, who are responsible for the smooth operations of a language pair Community, when certain language pairs were experiencing demand peaks, when the currently active translators were insufficient to handle the current workload.

At the time, it was a highly manual process, as Community Managers had to transition between multiple dashboards to compare the different variables that determine the current workload in translations.
The main issue was that the Community Managers didn’t have a unified and easy way to be notified when a translation queue was experiencing demand spikes and might generate excess workload. A streamlined notification system was needed to allow the Community Managers to focus more on core functions.

Operational efficiencies done easily with n8n

A peak management system is used to detect an imbalance between the demand and supply in any given digital transaction and, in turn, invite more market makers to participate to fulfill the demand. In the case of Unbabel, its premise was to enlist more editors to support the demand for translation language pairs.

The peak management system is triggered every hour, as n8n checks the current translation backlog, and in case of excess workload, a slack notification is sent to Community Managers prompting a message “this language pair might be having issues” and an email notification to editors saying “there’s work available on the platform”. This enables Community Managers to quickly intervene.

“n8n opened up numerous possibilities of running exciting R&D projects, which wouldn’t be possible due to lack of engineering resources,” said Pessoa.

unbabel-workflow-2.png

With n8n, you can set up queries using HTTP requests, so it allowed Delphine’s team to quickly pull in data and generate outputs for production systems. Without having n8n’s UI, these operations would be extremely manual and time-consuming, given the team had no engineering resources available.

A new feature saving their customers’ time

After successfully deploying the workflow, the Unbabel team managed to help Community Managers reduce the number of peak-related emails they had to send editors by 55%. Additionally, none of the primary language pairs had more than 10% of translations taking over an hour’s time.

“This wouldn’t have happened without n8n, especially in the timeframe we were able to achieve it. We were developing a portal and internal apps at the same time, so getting engineering capacity would have been impossible."

Discover how other users are saving engineering resources with n8n workflows.

Without n8n, it most likely would have taken half a year at minimum before this initiative would have been greenlighted.

n8n_image

Delphine Pessoa

Senior Product Data Scientist at Unbabel