Built on the nuances of language, DeepL is a translation model that helps people translate documents, speech, and images into over 100 languages. Steve Rotter, CMO at DeepL, shares how the tool, and its agent DeepL Agent, can provide new capabilities to marketing teams, while maintaining accuracy, security, and quality.
How do marketers use DeepL?
We just had a recent customer who used our tech – their use was all around the language for selling a new book, their marketers had to figure out ways to hyper-personalise for specific GEOs and segments. That message isn’t anything new, but the technologies that enable people to do that at scale are finally here.
For years, we’ve always said “Speak in the local language of the market you’re serving.” But it was tough, and you would spend thousands of hours and millions of dollars trying to get that language right. Whereas now with technology like DeepL’s, for example, getting that level of personalisation at that granular GEO level is possible.
What are some brand use cases for DeepL?
If you think about a typical brand that has to launch something in 20 markets, you have two choices as a CMO. You may launch them all at the same time, which means you have to get your product, all the content, and your website ready for that market; and you have to do that in 20 different markets. There’s no doubt you’re going to have delays, because maybe you’ve got it in English, French, German, and Spanish, but not in Japanese, Thai, or Vietnamese. Those delays are potentially launch-killing, because translation for launching a product is often very time sensitive.
DeepL dramatically compresses that time to market, so that it can be done in days and weeks instead of months and years.
How does the DeepL agent fit into this?
Then we’ll zoom out and see where the DeepL Agent fits in. It’s a general purpose agent that can do things on your behalf – like a smart intern. You give it a task – it can open up a browser, search for information, extract information from websites, and update records.
For example, let’s say you were launching a new product in the southeast United States market. You wanted to find all of the records in your CRM that had a certain number of states, research the companies, and tailor your launch message to those companies with a somewhat nuanced value proposition based on your launch.
DeepL Agent can do that. You give it a plain language instruction, and it can create an email outreach based on the information it finds. There’s also the ability to have a human in the loop, to review its steps when you’re ready.
Were there any benefits in building your agent, being a language company?
We spent the last eight years obsessing about language precision. When you invent an agent that has to be able to interpret natural language instructions and then take action based on that, you’ve got to be really good at language, beyond just building on an LLM that already exists.
The other thing we realised is, even if you just focus on the translation market, there are hundreds of tasks that a localisation team has to do, beyond just the simple process of translating. We found that no technology addresses those.
A lot of the language tech that’s out there touches on the core capabilities of language: routing, translation, and memory. But it doesn’t address those 100 plus other tasks that a language or localisation team has to do. So that’s what we set out to solve first, and we’ve seen it open up a lot more opportunities for other use cases.
What is unique about DeepL compared to other translation tools?
It’s accuracy and security. To say we are going to give all our workers access to a tool that can do things in our systems on their behalf is a scary proposition. One of the things that DeepL has built over the last eight years is a brand reputation for precision – it’s a brand people trust. And as companies consider their options for AI investments, trust is a really big factor.
The core thing all translation technologies have to have is quality. If you just start with that, the ability to take a word, sentence, paragraph, an entire document, and translate that with the right context and language – quality has to be there. Based on our recent quality benchmarks, probably 70 to 80 percent of the time, DeepL is going to have a higher quality of output than competitors. What that translates into is reduced post editing, reduced errors, reduced risk – all of those things.
The second pillar we’ve invested a lot into is the idea of personalisation, because other tools are based on a core translation set, but it’s not based on your brand, your style, your tone. Within the DeepL platform itself, you can personalise and create style guides, you can create glossaries, and you can create specific language that would be unique to your business – no one else really has that level of personalisation.
What ROI do customers see?
I think measured ROI is going to take a little while, and it’s going to come on two sides of a coin. One is just simple productivity: you could do the math and say, this would have taken four hours and it took DeepL 10 minutes. I think the harder part with these technologies is testing what to do with that extra time. Maybe you freed up an hour a day for something you never thought you would be able to do, which lets you analyse your customers better, understand how your products are working in the market better, or look at your media plan and see how it’s performing in a way you never really had the time to.
