Back office ops ·

Unbabel builds automated peak management system with n8n to reduce translation backlog

The problem

Unbabel's Community Managers had no unified way to be notified when translation queues experienced demand spikes, requiring manual monitoring across multiple dashboards to compare workload variables.

First attempt

The team evaluated Node-RED and Zapier as automation tools but rejected both — Node-RED had an unappealing interface and Zapier imposed strict limits on the number of workflow steps.

Workflow diagram · grounded in source
1
Hourly backlog trigger
trigger
“The peak management system is triggered every hour, as n8n checks the current translation backlog”
2
Pull data via HTTP
integration
“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”
3
Detect excess workload
validation
“in case of excess workload, a slack notification is sent to Community Managers”
4
Notify Community Managers via Slack
output
“a slack notification is sent to Community Managers prompting a message "this language pair might be having issues"”
5
Notify editors via email
output
“an email notification to editors saying "there's work available on the platform"”
6
Community Manager intervention
human_review
“This enables Community Managers to quickly intervene”
Reported outcome

The n8n peak management system reduced peak-related emails sent to editors by 55% and reduced by half the number of Communities with significant volumes of time-consuming translations (1 hour+), with none of the primary language pairs having more than 10% of translations taking over an hour.

Reported metrics
Communities with time-consuming translations (1 hour+)reduce by half
Peak-related emails sent to editors55%
Primary language pairs with translations over 1 hour10%
Time from conception to productiona couple of weeks
Reported stack
n8nSlack
Source
https://n8n.io/case-studies/unbabel/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The n8n peak management system reduced peak-related emails sent to editors by 55% and reduced by half the number of Communities with significant volumes of time-consuming translations (1 hour+), with none of the prima…

What tools did this team use?

n8n, Slack.

What results were reported?

Communities with time-consuming translations (1 hour+): reduce by half; Peak-related emails sent to editors: 55%; Primary language pairs with translations over 1 hour: 10%; Time from conception to production: a couple of weeks (source-reported, not independently verified).

What failed first in this deployment?

The team evaluated Node-RED and Zapier as automation tools but rejected both — Node-RED had an unappealing interface and Zapier imposed strict limits on the number of workflow steps.

How is this back office ops AI workflow structured?

Hourly backlog trigger → Pull data via HTTP → Detect excess workload → Notify Community Managers via Slack → Notify editors via email → Community Manager intervention.

Related back office ops cases