How do we show the value of localization?

"I work for a Localization Department (buyer side), and we’re under constant pressure to cut costs and speed up delivery because of AI. It feels like our department is getting smaller and smaller. I understand there are areas we can improve our workflows, and we should use the latest technology, but how can we get leadership to understand AI isn't good enough on its own yet? I need help. They don't see the value of what we do."

—Anonymous

Belén Agulló García

This is such an existential question, and it unites us all in the industry: language service providers, translators, educators, and localization buyers. We are all on the same mission: to explain to the world the value of what we do and convey it in a way that resonates with audiences outside of our bubble. I have been discussing this very same topic for years, and still cannot find a clear answer or magic recipe to make it work.

Anytime I think about who can explain the value of localization best, two names pop up in my mind: Nataly Kelly and Miguel Sepulveda. They have both written brilliant posts and books about this topic, so I would recommend anyone trying to find an answer to that question to start by looking for inspiration inside the minds of these two amazing localization professionals.

We need to start by convincing ourselves that what we do matters. That we matter.

Even if we are language and communication experts, we seem to keep failing at the most important aspect of all: explaining why what we do matters. Basically, to explain our purpose as an industry. I feel that sometimes we are lost in a spiral of workflows, technology, KPIs, processes, cutting costs, accelerating timelines (aka the HOW), and we lose sight of what matters: the WHY. If we’re able to dig deeper into the why, we will be able to better convey and connect our mission to something bigger than ourselves. We do this to connect artists with people who are craving something meaningful. We do this to help patients around the world have access to the best treatments in their language, so that they can heal. We do this so that users can access applications that make their lives easier. We need to start by rewriting our narrative and convincing ourselves that what we do matters, that we matter.

The other day, I read this article about how Klarna, after saying that they would go AI-first and lay off part of their team, are now backing down and hiring people again to do the work. And guess why? Because the AI approach led to lower quality, and customers didn’t like that. This is a great real-life example that anyone can share with their leadership to explain why fully automated workflows are not necessarily the best solution, even if they sound great on paper. Especially when those automated workflows have a direct impact on the most important asset of your company: the customers. And what has a bigger impact on people than language?

Let’s see what our Think Tank experts have to say about this.

Miguel Sepulveda

I completely understand your frustration. You’re not alone in feeling the pressure of AI-driven cost-cutting and reduced time-to-market expectations. Many in our industry face similar challenges. Echoing what Belén mentioned above, purpose is a great place to start, as it helps us connect emotionally to what we do. And a great way to continue moving forward the conversation is by showing the impact we enable, using metrics and outcomes that resonate with decision-makers, with our stakeholders. We should have different answers ready for the question “What is the impact of localization?” depending on whom we are talking to. We should be able to tweak our messaging so that it resonates with our different audiences. This is even more relevant in the age of AI.

Here’s how I approach these conversations when leadership is skeptical.

Frame the conversation thoughtfully by being proactive and not defensive.

First, I start by highlighting and clearly explaining the limitations of AI. The most common and mainstream type of limitation is hallucinations. GenAI can generate fluent but factually incorrect content, which is a critical risk, especially for emotional or creative text. Another issue comes with inconsistency and unpredictable output. AI often fails to follow established glossaries and style guides, which can damage brand consistency. At the same time, even with the same prompts, AI can produce inconsistent results, which are unreliable for quality-focused work. From a culturalization perspective, AI may generate inaccurate cultural insights, creating brand risks in global markets. If we talk purely about technology, GenAI currently poses tool integration challenges, as current systems struggle to integrate with existing workflows and tools. GenAI does not excel at learning from its mistakes, and attempts to fine-tune AI do not work yet, and can lead to even more inconsistent outputs. Last but not least, the cost efficiency argument can sometimes fail. When linguistic quality assurance (LQA) testing is factored in, AI is not always a cheaper option, and can actually result in higher costs if not planned correctly.

Secondly, I try to frame the conversation thoughtfully by being proactive and not defensive. I make it clear that I am open to AI, but grounded in reality and outcomes. It is important to explain that your priorities go beyond cost efficiency and also include reduced time-to-market, player/user experience, and overall quality. A strategy that can work is to suggest a three-tiered content model:

• Tier 1: High-impact content (narrative, DE&I, legal, marketing) should be translated by humans.

• Tier 2: Medium-impact content (LiveOps, CRM) can use AI assistance, but still requires human review.

• Tier 3: Low-impact content (support docs, patch notes) can rely more heavily on AI with light quality assurance.

Raising awareness about the consequences of falling for the AI hype trap can help manage expectations, as it has real consequences even for the biggest companies. An example that can be shared is what happened with Siri and Apple Intelligence. Apple has had to publicly apologize for its new AI last fall and now again.

If your stakeholders don’t see the value of what we do, that is definitely the metrics part. Metrics matter. These articles can help you structure your metrics strategy to show the value of what you do with data:

• Localization Metrics That Matter

• Speak the Language of Decision-Makers

• Common Localization KPIs and Pitfalls

Marina Pantcheva

Leadership and upper management often underestimate the complexity of AI implementation, largely because a bird’s-eye view obscures the tiny but critical details that can derail the entire process. For an AI implementation to be meaningful, adequately safe, and truly useful, it requires the expertise of people on the ground. These people are the members of the Localization teams. Yes, you specifically, dear question-asker.

But with this critical role comes responsibility: you must continuously learn, educate yourself, and be ready to guide, argue, and make informed techno-linguistic decisions.

To Miguel’s point above, I’d like to add two more:

When AI gets it wrong, it can get it disastrously wrong.

The risks connected to AI are by an order of magnitude higher than those associated with using human translators.

Suppose we have both human-powered and fully AI-powered localization workflows, and in both cases, 5% of the translated segments do not meet the quality bar. The key difference is that the 5% of human-translated segments will likely contain minor issues—missing commas, typos, inconsistencies. The 5% of AI-translated content is likely to contain critically wrong or even absurd outputs. The risks connected to AI are by an order of magnitude higher than those associated with using human translators.

Building your tech stack on third-party AI is like building a fortress on shifting sands.

As models get updated, retrained, or fine-tuned, their behavior can change unpredictably. AI-powered translation that works perfectly well today may break tomorrow. That's why you need a robust mechanism for continuous testing, monitoring, and validation.

Marta Nieto Cayuela

The best way to help leadership understand the limits of AI is to show them, not just tell them. Build your case with data.

Start by designing a comparative study using real content from your company. Choose a representative sample (at least 1,000 words, across different content types). Run it through various workflows: full human translation, NMT, LLMs, even copywriting if that is part of your services.

Then, evaluate the output blindly (without disclosing to the reviewers which workflow is which). Use the same assessment methodology of your choice consistently: automated scores, human review, and other methods. You can find more information in one of our articles.

Show them, don't just tell them.

Once you have the results, focus on:

1. Showing the results and trade-offs: use either your internal metrics (turnaround times, costs, quality) or industry metrics that can help you build the narrative with your stakeholders.

2. Highlighting benefits and risks: You can do a full SWOT analysis (strengths, weaknesses, opportunities, and threats) if it helps, or focus on strengths and weaknesses.

3. Stating requisites for adoption and defining the ask: what resources you need long term (e.g., engineering, AI experts, budget for AI evaluation, data specialists, tech stack, etc.), and what is the realistic outcome.

Present the results in a session tailored to leadership. The presentation should be high-level, but include examples and data to illustrate your points, add highlights and pitfalls (risks) of the different workflows, and your honest recommendation. It should be digestible and impactful. The purpose of this session should not only be to go over the results, but to steer the conversation towards expectations.

Aaron Bhugobaun

Building a case study is a strong way for businesses to get a grip on where AI actually adds value, but also a useful exercise to understand for yourself where the tech really is right now, and where it is heading. It is about testing AI on real-world tasks, looking at quality and speed across AI-only vs human-only vs human+AI. Key thing: be completely agnostic. No bias. Just run the tests properly.

Start by mapping the full workflow. Track the journey of an asset through the process, and identify where time is spent, where decisions are made, and where AI helps or gets in the way. You need to break this down clearly.

Key thing: be completely agnostic. No bias. Just run the tests properly.

Once you have run the tests, plug in the data. Compare quality outputs, time taken, cost (human hours, AI costs, etc). Then it is about analyzing what is faster, what is better, and where the friction is. Where AI works best and where humans excel.

Bridget Hylak

Selling language services is a lot like selling a box of crayons to someone who may be visually impaired - it involves conversation, education, description - and above all, trust in the seller. Clients/managers/stakeholders often have no idea what “quality” means, as they are frequently unable themselves to test or gauge that quality. Over the course of my career, I have repeatedly heard clients comment, “Oh, that translation ‘looks good,’” when evaluating text in an entirely unfamiliar alphabet, script or character system, as if their opinion of how something “looks,” though totally foreign to them, is adequate.

In this case, an argument can be made according to the wonderful examples provided by other AI Think Tank contributors (I specifically loved Marina Pantcheva's comment about building a tech stack on “shifting sand,” as so many systems and workflows seem to be in flux right now with no clear idea of where they may eventually land - what manager wants to onboard something, only to have to redo it later…?).

Selling language services is a lot like selling a box of crayons to someone who may be visually impaired.

I also encourage you to argue from the perspectives of time, training, regulatory and equity across languages.

Time

Тruly, not enough time has passed yet for the industry to project who or what will emerge as the new standard.

Training

GenAI in our workflows is new to most active linguists, still undergoing transformation as discussed, and the training needed to handle it properly is only in its infancy. Launching these technologies with little to no professional language oversight is reckless at best. Everyone is learning, and those who have made this “learning” a part/full-time job are emerging as teachers and trainers - though they, too, are still learning! (There is an additional terrible danger that these tools will and increasingly are being used by non-proficient or hobby linguists, which is equivalent to a certified public accountant using a surgical robot on a live patient. Just because the tech is good doesn’t mean the surgery will go as planned…)

Regulatory

What is allowed in terms of data handling and what is not? Is the enterprise international, as assumed? If so, the situation is even more complicated, as international privacy standards must be adhered to, and these, too, are changing (and represent an uncomfortable “grey zone”).

Equity across languages

The robustness of language resources varies greatly, specifically in the area of LLDs (languages of lesser diffusion, formerly, “low-resource languages”). While GenAI is improving and certain industry players are attempting to close these gaps, we are far from the Promised Land. This variance in how LLMs perform from language to language (especially, as mentioned, LLDs), not only reduces the quality of translations in those languages but further widens the digital divide and risks digitally marginalizing certain populations. Indeed, that “quality” will be different depending on each language combination and the methods used to get to the final product.

Libor Safar

These concerns truly reflect the current zeitgeist. There are two related but distinct issues at play: a) demonstrating the value of localization itself, and b) proving the worth of internal teams who manage it. While neither issue is new, AI has added a fresh dimension to both. I’d say that solving the first challenge goes a long way toward elegantly resolving the second as well.

The key is shifting from a defensive stance, constantly justifying our existence and purpose, to a proactive, business-oriented approach. Miguel and others have provided excellent advice for how to go about it.

What would happen if your entire department vanished overnight?

How quickly would global customers notice? When would international revenue begin to decline? What about rising support costs? Who would provide local market insights? Would risk managers lose sleep? At that point, cost-cutting would be the least of anyone's concerns.

The next real question isn't about reducing headcount through AI implementation. Instead, AI, despite its many current, and well-known limitations, enables us to accomplish what was previously impossible or cost-prohibitive. What more can we now do?

We need to focus conversations with leadership on expanding reach and impact rather than just cost reduction. The goal is to position your team as technology-forward experts who can utilize AI while providing irreplaceable human value, for the benefit of the business.

Easier said than done? Absolutely. But with solid data, information, and supporting evidence, it's not impossible.

About the Series:

As part of our "Ask the Think Tank" series, members answer reader's questions to help foster knowledge sharing and become a resource when you don't know where to turn. To submit your own question, click here.