What can Localization learn from about 20 years of shadow IT?

A few days ago, I was talking with a friend who leads a Localization team at a company whose main product is financial software used by businesses to manage their financial operations.

We talked about a bit of everything and, of course, AI came up. What a surprise, right? :) He told me he was getting quite a lot of pressure to use more AI in Localization.

Well, I told him that this does not surprise me at all. Almost the strange thing nowadays would be the opposite! A Localization leader who is not feeling some kind of pressure to implement AI in one way or another in their processes.

Leadership teams are asking functions across the company how they are using AI. Questions like: Where can we automate? Where can we save money? Where can we move faster? Where can we scale without adding more people? These are becoming pretty normal. So far, nothing unusual.

The combination of AI hype + the fact that AI is genuinely useful in many cases has been the norm for the last 2-3 years. But then he mentioned something that made me think about the topic from a completely different angle. His team had recently come across some ads with translated subtitles.

While they were reviewing them, something did not feel familiar. Nobody remembered working on those subtitles. So they checked internally with the Localization team. Nobody knew where they had come from. My friend kept pulling the thread.

He spoke with the Product Owner, who redirected him to the Marketing team leader, who then redirected him to the Marketing specialist who had been working on that particular campaign.

And during the conversation about those ads, the Marketing specialist mentioned something that I think can worry and demotivate those of us leading internal Localization teams in equal measure.

It turns out they were using HubSpot for their marketing campaigns, and HubSpot now includes AI-powered multilingual translation capabilities. They were a bit tight on time. So they decided to use the feature. Why not?

Ouch!

I do not know if you are familiar with it, but HubSpot allows users to create a version of a page or blog post in another language and automatically translate it using DeepL. Think about what that means from a Localization perspective.

Someone in Marketing can be working on a landing page, decide they need a Spanish version, generate it, and continue working without necessarily entering the company’s Localization workflow at all.

And this is where the problem starts. It is not necessarily that another team wants to bypass Localization. They simply need the Spanish campaign. And there is a feature in the software they see and use every day that helps them do exactly that. Automatically. No need to open a Localization request in Jira or whatever equivalent tool the company uses. No need to explain what they need, when they need it, provide context, and so on. So why would they bother contacting the Localization team?

Uf, scary thought. And I totally understand it. That button gives you a more or less decent translation. Good enough in many cases. And it gives it to you instantly, with almost no friction.

And we humans are very attracted to anything that helps us get things done with less effort :) So that is where this post is going. What can we do in a situation like this?

Shadow Localization Is Here for Localization Professionals

For years, IT teams have been dealing with something that became known as shadow IT: the use of software within a company that is not supported by the organization’s central IT department.

Source TechTarget

It is a term that has always carried a somewhat negative vibe. And the reasons employees use software not approved by IT are very similar to the reasons someone might translate content without going through Localization. One of the main reasons is impatience.

In IT, shadow IT often results from an impatient employee seeking immediate access to hardware, software, or a specific web service without going through the necessary corporate steps. I see many parallels between shadow IT and the early development of what we could call shadow localization.

The good news is that we do not need to reinvent the wheel. Many of the practices IT teams developed over the years to deal with shadow IT, understanding security risks, avoiding unnecessary costs, increasing visibility, and creating approved alternatives, can also help us build a framework for Localization. A framework that can help us bring some light into the shadows created by shadow localization.

View fullsize

So, let’s look at four areas that I think can help us avoid Localization being bypassed.

1. Make the Risk Concrete

One of the big concerns with shadow IT is security and compliance. IT cannot properly protect something it does not know exists, and unsupported applications may not follow the same controls as approved systems. I think Localization might be facing something similar.

The concern is not only that someone used AI to translate a few subtitles. The problem is what might come next. Little by little, the question that starts going around in our heads is:

What other multilingual content is being created without the Localization team's knowledge? Maybe it is a social ad today. Tomorrow it could be product information, pricing, customer communications, or content with legal implications. And that is where the risk starts to become much more concrete.

·      Wrong terminology.

·      Inconsistent brand voice.

·      Incorrect financial information.

·      Compliance issues.

·      Or simply a customer experience that nobody in the Localization team has seen or reviewed.

2. Prohibiting AI Translation Probably Will Not Work

Another lesson we can learn from shadow IT is that blocking a tool does not eliminate the need. I think the same will happen with AI translation. If Marketing, Content, or Product teams already have multilingual features inside the tools they use every day, telling them not to use those features will probably only get us so far. Telling them that AI translation quality is not good enough will probably not work either.

And honestly, what does “good quality” even mean in Localization? In industries like pharma, legal, or life sciences, the quality argument can be very powerful because a translation mistake can have serious consequences. But in many other industries, the situation is different. If the content is not related to health, safety, or legal risk, stakeholders may be perfectly happy with something that is simply good enough for the purpose.

So saying “AI quality is not good enough, therefore you need to come to Localization” is probably not a very convincing argument on its own. We need to be more specific about the risk, the purpose of the content, and the level of quality that is actually needed.

Because the reality is that these stakeholders are not necessarily trying to create risk or bypass Localization. Most of the time, they just need to get something done, and the AI option is right there. Fast and easy.

This is where I think Localization also needs to look at its own process. If someone can translate, review, and publish content directly from HubSpot in a few minutes, while the official Localization route means finding a portal, filling in a request, providing context, waiting for the request to be processed, and then waiting again for the translation to come back, we should not be too surprised if the easier option wins.

And if we simply prohibit that option without offering a realistic alternative, there is a risk that the translation still happens, only with even less visibility for the Localization team. So perhaps the answer is a combination of clear guardrails and a much easier approved path.

We need to define where AI translation is acceptable, where additional review is needed, and where the risk is high enough that the official Localization workflow is still required. If we want people to stay inside the Localization process, convenience has to be part of the solution too.

3. Visibility Before Control

There is another lesson from shadow IT that I think applies really well here. Before you can manage something, you first need to know where it is happening. In my friend’s case, the worrying part was not only that Marketing had used AI translation. It was the Localization team that discovered those ads by chance.

They did not even know this type of multilingual content was being created. And that makes me wonder how many Localization teams really know where translation is happening across their company today.

·      HubSpot

·      ChatGPT

·      CMS platforms

·      Customer support tools

·      AI video tools

·      Local agencies

·      Regional teams

There may be many more paths than we think. So before creating another AI translation policy, it might be worth doing a simple mapping exercise with Marketing, Product, Content, Support, and regional teams.

And I think the most useful question is not: “What are you planning to use?” But: “What are you already using today?” That gives Localization visibility first. Only then can we start deciding where the real risks are and what actually needs to be controlled.

Final Thoughts

After thinking about my friend’s situation, I do not think the answer is to keep AI away from Localization. Quite the opposite. There will be many situations where AI translation is exactly the right solution. But I do think Localization should remain the funnel for multilingual needs. And perhaps this is another lesson we can take from shadow IT. The answer was never simply to stop people from using technology. IT created approved environments, selected the right tools, defined security requirements, and enabled the business to use technology without requiring every employee to become an expert in technology governance.

I think Localization has an opportunity to do something similar with AI. The decision about how content should be localized should sit with Localization. Our role is to understand the content, risk, quality expectations, and timeline, and then choose the method that makes the most sense. Sometimes that will mean a highly automated AI workflow. Sometimes it will require human review. And sometimes a more traditional approach will still be the right one. The challenge for us is to meet the speed the business now expects while still protecting the level of quality and control each type of content needs.

 @yolocalizo