Prezly's approach on AI, focus on Useful AI™️
Say bonjour to Auto-Translation and Actually Useful AI™️
As you’re almost certainly aware, AI is everywhere. For years now (and moreso these past few months), it has been busy disrupting industries from supply chain management to croissant production, and naturally public relations is no exception.
As a group of techie-types working in SaaS, we got predictably hot under the collar at the prospect of adding AI to our platform in a way that would help our clients do what they’re already doing, only easier and faster. We didn’t jump on the bandwagon instantly, instead started experimenting, researching and writing about AI wherever we could manage.
But there’s a difference between incorporating AI into your software because it’s new and shiny and gets hits on Google, and doing it because it actually improves your product.
When we were thinking of how to best add AI functionality to Prezly, it was important that the end result would benefit our users rather than simply be a gimmick.
To that end, here are the few principles we set out for ourselves:
- AI should build on the capacity of our users, not replace them
- AI should be offered in suggestion mode – there are always good reasons to overrule it
- AI should save our users time in areas they don't like spending time
- AI should not be offered for the sake of offering AI
We even took to referring to this internally as #usefulAI, after a while without cringing.
And although it's been 2023’s talk of the town, AI isn't a new avenue for Prezly. We've been using some sort of background AI and machine-learning processes in the background since around 2015. Prezly search is powered by Algolia, for instance, which is heavy on machine-learning, and our old star ratings feature was very much avant la lettre.
But today, we're ready to announce a new feature within Prezly: auto-translation.
It's something that’s been on our minds for a while now, and we'd tested some options before, but never felt that the quality was good enough to make it worthwhile for our customers. Because in most cases, our users write and share quite sophisticated text that requires significant nuance.
And to be completely transparent, we're still not sure how the new feature will perform with different languages (both source and target), but the only way to find out is to roll it out and get real feedback from the people who use it. Our internal tests are impressive, making the generated translation very usable.
Prezly is a child of multilingual parents. Growing up in Belgium, even our earliest users needed the capability to publish stories in Dutch, French, and English. Making it possible to translate a story quickly and easily has always been a core feature and a primary selling point.
One of the reasons why customers choose Prezly is that it's so easy to use it to run a multilingual newsroom. Looking at our backend stats, on average, for every story published there are two translations. Translations are a big deal for us, because they’re a big deal for you.
But while we worked to make our translation setup useful and straightforward, we always knew the process could be simplified. And you knew it too.
Fortunately for us, the translation of text from one language to another is one of the handful of useful things that AI is getting pretty damn good at.
Which brings us to today.
We’re upgrading Prezly’s existing translations feature to include automated AI translation from one language to another using what is currently the most effective auto-translate service on the market, DeepL.
This means that instead of briefing a (usually external) party on the translation you need, waiting for this to come back to you, proofing it, creating a linked translated version of your story, and finally copying the new text into your layout, you can simply…
… click to translate your story and review the text.
I’m sure you’ll agree, that’s far, far simpler.
Like I mentioned earlier, the service powering the new feature is DeepL, currently the most accurate machine-learning translation tool on the market. You can try it in your browser right here if you like.
We support 30 languages through DeepL, which accounts for the most common use-cases of Prezly customers. At the moment, it lacks a few that we support – such as Arabic – which means that we cannot offer AI translation for every language listed within Prezly. Yet.
Essentially, this new feature has the potential to massively speed up the press-release-creation process for anyone who routinely works with multiple languages. Although you will still need to review and finetune the translated text to be sure that it matches what you need it to say, you’ll no longer need to create the translation itself from scratch or wait around to get it back from your translator.
It’s also about as secure as can be, with any text fed through DeepL deleted from their servers after translation, keeping your privacy in mind.
And of course, you’ll still have the option to ignore using AI translation altogether and create your translation manually if that’s what makes sense for your context. You do you.
The next step? Test it out! We’ve run our own internal trials, but the proof is always in the pudding, so we won’t really know how well it works until you eat try it. Which is why I’m here inviting you to do just that :)
The next time you need to get a story into a different language, click the usual translation button in Prezly and you’ll be offered the option to have the AI translate it automatically. Try it and note down how it goes – what works well, what’s clunky, what the end result was. Then please, let us know! We’ll use that feedback to iterate on the feature and make it better, faster, simpler to use next time.
Auto-translation is completely free for anyone to use while we’re still ironing out the kinks. In future, it’s likely that we’ll make it available as an optional add-on depending on your usage. We’ll let you know when this happens, and never charge you extra for anything unless you specifically ask us to.
So, yes, AI can absolutely be a trendy gimmick that SaaS platforms use to seem down with the kids. But that shouldn’t overshadow the fact that in some cases, it can be genuinely very useful. Give it a red-hot go, and tell us what you think :)
Until next time,