Reflections from each session at XTM Live City Tour Dublin 2026: what was said and what it means for enterprise localization
There's a nice symmetry to bringing the City Tour to Dublin. This is, in many ways, where our industry grew up, the place global software companies came in the '90s to localize for the world, and where many of the people in the room first learned the trade. One of our speakers reminded us he arrived here in 1992; another runs the company from just outside the city. Coming home felt right.
As in London, we kept the room deliberately small. Localization leaders, customers, and partners, close enough to have the conversations the moment demands. We opened with a quick live temperature check on AI: where teams are today, the biggest objection they hit when trying to scale, and their biggest blind spot. The honest answers ("holding it together," "too messy to train / it'll break our brand voice / it moves too fast," and "the human element and quality control") set the tone for everything that followed.
Here's what came out of the day, session by session, with the highlights worth taking forward.
Alison Toon, Principal Analyst, CSA Research
Alison opened with preliminary data from CSA Research's annual global market survey. The headline isn't comfortable, but it's clarifying: the industry grew steadily until the pandemic, then flattened, and the flattening has far more to do with automation and machine translation than with COVID. Adjusted for inflation, she expects a real-terms contraction of around 17% relative to where the market would otherwise sit.
The more useful story is the divergence beneath that average. The industry has split into a K. On the bottom leg, providers still selling translation as a transaction are struggling. On the top leg, those who've moved into solution selling (solving business outcomes, not shipping words) are growing. In the preliminary data, the top third reported a median revenue increase, while the other two-thirds declined. For enterprises choosing partners, the lesson is direct: pick the businesses on the top leg, because they're the ones who'll still be here.
She also reframed the scale of the opportunity. Of the content generated globally every day, roughly a third could valuably be translated, a volume so large that human and machine translation combined still only touch a sliver of it. The bottleneck is no longer capacity. It's governance.
That's the crux of CSA's "transformational imperative" argument. With everyone now one "translate" button away from publishing, the real crisis is one of sovereignty, ownership, and control: where the data lives, who manages it, which laws apply, and who's accountable for what goes out. The TMS isn't disappearing; it's evolving into a post-TMS ecosystem that acts as the governance, access, and control layer for the whole enterprise, powered by the language intelligence (TM, terminology, style) organizations have built up over years.
Risk becomes the gateway to automation. Low-risk, low-impact content can be fully automated; anything touching brand, regulation, or people needs human oversight, with audit trails and traceability as first-class requirements. And the people doing this work need to talk to executives in their language (efficiency, risk avoidance, retention), not cost-per-word.
I'm fed up with hearing 'human in the loop.' It's human at the core: human-centered, with AI as the tool. The market is reorganizing, not disappearing, and governance is becoming the single most critical consideration."
— Alison Toon, Principal Analyst, CSA Research
WHAT IT MEANS
If you're building an internal business case, lead with the K. The question your leadership should be asking isn't "how do we cut localization cost," it's "are we on the top leg or the bottom leg?" Position the function around outcomes and governance, and the budget conversation changes shape. The volume data also gives you air cover: the work isn't shrinking, it's relocating, toward managing risk across far more content than any team could previously touch.
Sara Basile, Director of Platform Product Management, XTM International
Sara picked up exactly where the market data left off. Content volume is exploding while the operating model for getting it translated (at quality, on time, on budget) has barely moved in 10 to 15 years. Stitching products together with integrations isn't the fix. A platform with intelligence built in is. And critically, it's one platform: the same foundation serving the product team that needs strings in hours and the regulatory team that needs every word audited.
She walked through the three-layer model that frames the XTM Intelligence roadmap:
XTM Go is the visible new surface: a headless triage capability. Content arrives, gets analysed by risk, intent, and structure, and is routed to the best strategy: you set the quality bar, budget, and risk tolerance, and the system does the rest. Because it's headless, it comes to the user inside the tools they already work in. Sara showed an example of a Microsoft Teams user being alerted to a quality or due-date risk without ever logging into a portal, with the platform going to the user, not the other way around. The LLM Gateway sits behind all of it as a control centre for every engine and key in the organization, giving the cost transparency and audit trail that feel so hard to get today.
The trajectory she laid out: configured today, adaptive tomorrow, autonomous over time. Each layer moves along that curve: brand memory that starts injected and learns to self-curate, routing that starts rule-based and grows to weigh budget and outcomes, quality evaluation that moves from flagging to fixing.
Right now, when you need to localize a piece of content into 5, 15, or 47 languages, deciding the best route is still a human call. We believe it should be the platform's, smart enough to understand the content, weigh the risk, and choose the right strategy on its own.
— Sara Basile, Director of Platform Product Management, XTM International
WHAT IT MEANS
This is the clearest statement yet of where the localization leader's role is heading: not extinction, elevation. The work moves from operating the machinery to defining the policy the machinery follows. The open, composable direction (bring-your-own-LLM, MCP-first, headless) is also the right defensive posture: it lets you extend governed translation to the long tail of content currently bypassing you, and it stops other teams quietly building point solutions that fragment your control.
Lorcan Malone, CEO, XTM International, with Antonio Tejada, Vice President of Translation and Localization, LanguageLine
The closing fireside paired Lorcan with Antonio, a 30-year veteran who started his career in Dublin and now leads enterprise transformation at LanguageLine. It was the practitioner's reality check on every theme of the day.
Antonio's framing was the one worth keeping. The industry has transformed before (the move to Ireland, then translation memories, then NMT) but those were industry transformations, things the rest of the world neither saw nor cared about. This time is different: we're transforming on the same page as everyone else, because LLMs have democratized access.
Everyone is suddenly a translator. That's both the opportunity and the threat, the risk being that the world transforms us rather than us driving the transformation. On careers, his read was unsentimental: the "linguist" as we knew it is becoming a different role, drawing people from AI and NLP backgrounds, with communication as the durable skill.
The most memorable moment was an anecdote. LanguageLine spent a year with one of the world's largest banks designing new AI-driven localization workflows, only for a senior executive to call and say the bank's AI policy was staying entirely in-house. A year of work, seemingly gone. Except the same executive's next sentence was the point: what we want from you is the human component: validation and verification. It turned out to be roughly four times the value of the technology deal. The lesson, echoing Alison: accountability for published content always sits with the publisher, and the enduring human value is in training, validation, and managing compliance, not in dictating anyone's AI policy.
Lorcan tied it back to XTM's posture. Whatever a customer's AI stance (happy to use ours, strictly in-house only, or a mix of third parties), the platform's job is to orchestrate it. The open-platform philosophy holds.
We're probably not going to be the ones dictating any organization's AI policy. But this time it's an issue for everybody: the accountant feels it, the lawyer feels it. It's a shared problem now. We'll converge on where the human value sits, and that's what keeps me optimistic.
— Antonio Tejada, Vice President of Translation and Localization, LanguageLine
WHAT IT MEANS
For buyers and LSPs alike, the strategic move is to stop competing on "we have AI" and start competing on the human layer that AI makes more valuable, not less: validation, governance, model training, and the institutional knowledge no model owns. For platform selection, the criterion is openness. Can it govern and orchestrate whatever AI the enterprise mandates, including the customer's own in-house models?
→ The market is reorganizing, not disappearing. The K is real. Providers and internal teams that sell outcomes and governance are growing; those still selling words by the unit are not. Decide which leg you're on, and build the business case from there.
→ Governance is the new non-negotiable. Speed is no longer the differentiator; anyone can paste a document into an LLM. Data sovereignty, audit trails, traceability, and clear accountability are. The way to tame shadow AI isn't to ban it; it's to make the governed path the easiest path.
→ The human stays, but moves to the core. Fewer people post-editing "click here." More validating high-risk content, shaping brand and risk policy, and training the models themselves. "Human at the core," not "human in the loop," and the data says it's the durable role.
The afternoon shifted into two back-to-back round tables. One on the AI reality check beyond the hype, one on the tech stack, connectivity, and integration. The nuggets that stuck: outcome-based pricing is where everyone wants to go, but nobody can yet agree what "outcome" means; the real cost of AI is the total cost of ownership (infrastructure, coordination, manual review), not the token bill; sometimes the sharpest question isn't which workflow, but whether a workflow is needed at all; and poor AI output is almost always a data problem; terminology, style guides, and clean source content still decide the result. On connectivity, the room got about 30 seconds into "connectors" before the word "painful" came up, and then MCP, which more than one table called the future of integration.
We also marked the launch of our new industry book, produced with Multilingual and LanguageLine, ahead of LocWorld. Thank you to our partners Argos Multilingual, Intento, and ContentQuo, and to every customer, prospect, and team member who joined us in Dublin.
That's a wrap on the City Tour for now, and many of us will see you at LocWorld this week.