How to win with AI: Insights from a CEO

The rising pressure to use AI

AI is no longer a future concept. It’s on the table in every leadership meeting I’m part of, and it’s becoming a top-down priority across industries. More CEOs, including myself, are asking their teams to integrate AI into their workflows. The expectation is clear: every team should be using AI to improve efficiency and performance.

The pressure to operate with an AI-first mentality is real, but many teams are struggling to understand what that actually means in practice. The goals (faster output, lower costs, better quality) are valid. But when teams jump in without the right strategy, the results can miss the mark. I’ve seen companies spend heavily, lose time chasing fixes, or end up with translations that need so much cleanup, it would have been faster to use conventional human or machine translation.

So let’s reset the conversation. Being AI-first doesn’t mean using every tool you can find or building bespoke AI applications. It means knowing when and how AI actually adds value. In this article, I’ll share how we think about AI implementation at Smartling, where it’s delivered meaningful results, and what other teams can do to make smart, lasting progress.

The hidden dangers of DIY AI

A groundbreaking MIT study found that despite $30–40 billion in enterprise investment into GenAI, about 95% of projects fail to deliver a return. I’ve seen many teams try to build AI into their workflows on their own, and at first, it seems doable: someone selects a model, sets up a few prompts, and things start to move. “Vibe coding,” or perhaps now “vibe translation,” often gets you 80% of the way there. But as all software engineers know, it’s the last 20% that’s the problem. At that point, the real work begins: training, testing, monitoring output, managing costs. And that’s where it usually breaks down.

The DIY approach might feel efficient in the short term, but because engineering resources are finite, these efforts often drain time and talent. In my experience, engineering teams deliver far greater ROI when applied to customer solutions rather than internal software development. The talent cost simply isn’t worth the return. And, without the right guardrails, the output of DIY AI can be inconsistent or flat-out wrong. As a result, companies often spend more time cleaning up results than getting value from their investment. 

Handling AI in house is also rarely as affordable as executives expect. Large language model (LLM) translation, for example, appears affordable on the surface, but after accounting for the loss of translation memory, high error rates, hallucinations, and expensive human correction, the economics flip. In fact, it’s often more expensive than trained machine translation (MT).

The biggest risk is thinking AI will run itself. It won’t. AI translation needs structure, oversight, and expertise. Trying to do it all internally may seem like the faster route, but more often than not, it slows everything down.

Why the right translation partner is key

LLMs can do impressive things, but each new GPT release has shown smaller and smaller quality gains; even the latest models still score 50% lower than trained MT on Bilingual Evaluation Understudy (BLEU) and Translation Edit Rate (TER) quality benchmarks. Progress is slowing, not accelerating, as LLMs begin to exhaust available training data.

That’s why companies need more than just access to the technology; a solid framework is critical to a successful AI translation program. The providers pulling ahead in this space aren’t dabbling, they’re designing purpose-built solutions that exponentially increase the value and benefit of AI. 

Working with the right AI translation partners gives you a shortcut to the expertise that’s difficult to build internally: integrations, automated workflows, linguistic asset management, quality assurance, prompt engineering, and more. A partner gives you the space to focus on results, not translation model setup.

What it looks like to be AI-first

Being AI-first doesn’t mean using AI everywhere. It means using it where it matters. At Smartling, we ask one question before committing resources: Will this deliver real outcomes and ROI for customers? If yes, we move forward. If not, we don’t. That discipline keeps teams focused on results, not buzzwords.

We draw on this discipline to guide our AI decisions. That’s why we’ve paired trained LLMs, sophisticated prompt engineering, and MT inside an enterprise translation management framework. With this framework, our customers start to see real benefits: faster turnaround times, more consistent messaging, and improved brand voice across markets. The hybrid approach—using each technology for what it does best—is where the value is.

Where to start with AI in localization

Start by picking one real problem. Maybe it’s slow turnaround times. Maybe it’s rising costs. Maybe it’s the need to scale. Either way, begin with a specific pain point that AI can solve for you. Avoid trying to fix everything at once.

Start with content that feels safer for experimentation; user generated content, internal documentation, and help center articles are good examples. These give you room to test without putting high-stakes materials at risk.

At the same time, make sure people are still part of the process. A human-in-the-loop model works well here, where AI handles the first draft and then skilled linguists step in to ensure the output is accurate, aligned with your brand, and ready to go.

Use AI, but do it with purpose

Implementing an AI-first strategy isn’t about checking a box or chasing trends. It’s about making smarter decisions that improve how teams work day to day. The teams that take time to learn, test, and apply AI with care—with the help of a strong partner—will get better results. They will also position themselves as more strategic and more valuable in the long run.

You don’t need to overhaul everything at once. But waiting isn’t the answer either. Ultimately, being AI-first isn’t about doing it all. It’s about doing what matters, and doing it well.