A Tale of Workflow Automation, LangOps and Experimentation with Edwin Trebels (Part 2)
Edwin Trebels
Translations Director, Philadelphia Church of God
40 Hours a Week Saved Through Workflow Automation
Edwin Trebels, Translations Director at Philadelphia Church of God, returns for Part 2 with a detailed look at the LangOps system his team has built. Using Blackbird as a workflow orchestrator, they have created a component-based tech stack that ties together DeepL for machine translation, Dropbox for file handling, Trello for Kanban-style process management, and Slack for triggering translations with a simple emoji flag.
One microservice demonstrates the speed: drop a blog post into a Dropbox input folder, Blackbird downloads the glossary, translates via the DeepL API, renames the file, adds the word count, and uploads the translated file to an output folder. All within 15 seconds. Across all their microservices combined, the team saves approximately 40 hours per week, the equivalent of a full-time role. That person has been retrained into more creative, meaningful work rather than being made redundant.
Strategically and Tactically Independent
Edwin's team was given the freedom to build rather than buy. Larger localization suites exist and are the right choice for many companies, but they are not as customisable as Edwin's team needs. By assembling their own stack from individual components connected through Blackbird, they get exactly the tools they want at a lower entry cost, with full autonomy over their workflow.
The system uses Trello's Kanban boards for process management, mirroring how the car industry moves products from step to step. When a file is dropped into Dropbox, Blackbird automatically creates a project card on the correct Trello board using the right template based on the file name. Translations can even be triggered via Slack by dropping a document into a channel and adding a language flag emoji.
Two-Thirds Accuracy from Generic Machine Translation
A recent quality estimation analysis of a full year of Spanish content found that 27.3% of DeepL's output was published untouched by human linguists. Another third required only light editing. The final third needed heavier editing. That is with a generic model. The next steps are custom neural machine translation models trained on their own translation memories, followed by large language model integration.
Experimentation Should Not Be a Dirty Word
Edwin's strongest message: experimentation should not be discouraged by leadership. He describes conversations with people at large companies who experiment in secret because they fear it will appear in their performance reviews as evidence they do not know what they are doing. If you cannot experiment, you cannot know whether something is worth adopting. His advice: start small, measure the change in isolation so you can see its actual impact, and be willing to discard what does not work. A culture of experimentation is the only path to continuous improvement.
He also revisits the Gemba Walk concept from Toyota: senior leaders walking the production floor, treating workers as knowledge workers, and thanking them for pulling the cord when they spot a problem. The flow of information from top to bottom and bottom to top is where real organisational learning happens.
Beyond Translation: The Native Experience Podcast is produced by LEXIGO, Australia's trusted translation and multicultural communication agency.
Beyond Translation: The Native Experience Podcast explores multicultural communication, translation, and culturally diverse engagement in Australia and beyond. Each episode features experts sharing real stories and practical insights on topics from multicultural campaign strategy to CALD community engagement and localisation best practices. Produced by LEXIGO, Australia's trusted translation and multicultural communication agency with triple ISO certification and NAATI certified translators across 171 languages.
Edwin: Experimentation should not be a dirty word and it should not be something that leadership discourages. I think that comes down to the culture of your organisation. Is it okay to experiment? Because that is really important. If you cannot experiment or do experimentation, then how do you know if something is good to adopt or not?
Brian: Today we are back with Edwin Trebels, Translations Director at Philadelphia Church of God. He is working on a number of new and exciting projects. Edwin, welcome back to The Native Experience.
Edwin: Thanks, Brian. Thanks for having me on.
Brian: What are you nerding out on right now?
Edwin: I am nerding out on Blackbird at the moment, blackbird.io, which is a workflow orchestration tool.
Brian: With over 2,000 language technology tools out there, how do you sift through them?
Edwin: You have to take it one step at a time. Think of one single use case and say, this is what I would like to accomplish. Then narrow down, evaluate with a few companies, rely on peers, ask around, look at what is popular and what seems to be working for other people. Consultants definitely have a role to play within this massive boom of language technology.
Brian: Tell us about the LangOps workflow you are building.
Edwin: LangOps is exciting for us because we can create a component-based tech stack. We use Blackbird as the backbone of our workflow for a whole variety of use cases. Tools connect to Blackbird as an app based on API connections. We have something like two dozen experiments running at the moment.
Edwin: One example: we have a Dropbox folder where you put a file ready for translation in an input folder. Blackbird is notified, downloads the glossary we have prepared for a target language, imports that glossary into DeepL, translates the document through the DeepL API, changes the file name, adds the word count, and uploads the article back to an output folder. All within 15 seconds.
Edwin: Even if you did that manually on DeepL Pro, uploading, clicking, waiting, downloading, that takes a few minutes. If you did it a couple of thousand times, you are saving hours daily and weekly.
Brian: How many hours are you saving?
Edwin: All of our microservices combined, that is probably a savings of something like 40 hours a week. It is literally a full-time job that is automated. This person who was doing this work is being retrained and doing other things. Overall the output is much higher with that kind of savings.
Brian: How accurate are the translations?
Edwin: We had an analysis done recently by a company specialising in quality estimation. They found that when we use a generic model like DeepL, on average over a whole year of Spanish content, 27.3% is untouched by human linguists and gets published as raw machine translation. Another third is almost untouched with very light editing. Then another third is more heavily edited. That is with a generic model. There are opportunities for custom or adapted machine translation models, which is our next step.
Brian: You have been told you are strategically and tactically independent. What does that mean?
Edwin: There is always a discussion between buying versus building. Larger suites are available and they are the right choice for many companies. But they are not as customisable as we would like and they do not necessarily use the exact tools we are already using. When you make your own suite, you are more independent. You have more autonomy. You can choose what tools you want to use.
Edwin: We use Trello with a Kanban style process management, which is how the car industry uses it. A product moves from list to list, step to step. With Blackbird, once our microservice is built out, putting a file in Dropbox will also trigger the creation of a project card on our Trello board automatically and select the correct template based on the file name. We can even trigger translations through Slack by dropping a document into a channel and adding a language flag.
Edwin: It creates this almost magical system with exactly the tools we like using. That is what makes us strategically and tactically independent. It is specifically customised for us and the entry costs are much lower.
Brian: How long does adoption take?
Edwin: It depends on where you are in your maturity as an organisation. The architecture tends to reflect the complexity of the organisation. Complex ones can take years to build systems. We operate like a startup in some ways even though we are a larger organisation. We move very quickly. Within a matter of months of research and adopting technologies, we are already building things that make a direct impact.
Brian: How far is too far with process management?
Edwin: That is always the balance. You do not want to make it too complicated or have something that breaks all the time. You have to weigh it in the balance and then experiment. Experimentation should not be a dirty word and it should not be something that leadership discourages.
Edwin: I was talking to people at large companies recently who said they were experimenting in secret because they felt it would come across as if they did not know what they were doing and would come up in their evaluation. That is sad. You need a culture where it is okay to experiment. If you cannot experiment, how do you know if something is good to adopt? You have to try small things and be willing to get rid of what does not work.
Brian: What advice do you have for companies navigating all of this?
Edwin: You need to carve out time for this. Somebody who understands the needs and use cases needs to have freedom and time. I was talking to somebody at a company with 2,000 employees and they said nobody has time for this kind of thing. Everyone is running around constantly busy with daily stuff. A lot of it has to come from the top. Leadership has to allow for it. Sometimes it needs to be a senior person doing it because there are a lot of ramifications including financial, staff, and division of labour. You need buy-in from leadership.
Edwin: This reminds me of the Gemba Walk concept. In a car manufacturer, senior folks walk the manufacturing floor. They consider the workers knowledge workers. There is literally a cord hanging above them called an Andon cord. They can pull it for any kind of improvement or problem. And they are thanked for pulling it. On the Gemba Walk, leaders walk around and see how things are really going. That is when you see the real truth of what the problems are. When you have that flow of information from top to bottom and bottom to top, you have real education within your company. You need to get out of your office and see what is actually happening.
Brian: Edwin, thank you. This was great.
Edwin: Thanks, Brian. Thanks for having me again.
Brian: Edwin Trebels, Translations Director at Philadelphia Church of God. Remember, always strive for authenticity and embrace the power of native experiences.