Localization pricing models used to revolve around buyers counting words and vendors quoting rates. But this shared logic is now breaking down. AI has changed what the work costs and where the effort sits. As a result, a word count no longer maps cleanly to value.
Many teams sense the problem yet have no clear replacement. They know per-word pricing feels dated; however, they are struggling to picture a fairer commercial model.
Why the old model is breaking
Per-word pricing made sense when translation was mostly human labor. Each word took time, skill, and judgment. Counting words was therefore a fair proxy for effort. That link has weakened sharply. Machine translation and large language models now handle the first pass on much of the content.
Studies show that per-word pricing is under review as translation moves toward humans supporting machines rather than the reverse. When a model drafts and a person edits, word count stops measuring the full picture of the work.
Procurement adds to the equation. Most teams still benchmark vendors on a single per-word figure. Yet that number now hides more than it shows. Two quotes at the same rate can involve very different levels of AI use, human review, and risk. Consequently, the metric that once simplified buying now obscures it.
“What’s changed most is the level of the conversation. A few years ago, I was talking to procurement about rates. Now I’m talking to the same companies about risk, governance, and who carries the liability when something goes wrong in a market they can’t read. That’s not a purchasing conversation anymore. It’s a much more senior one, and it tells you where the real value has moved.”
Caroline O’Connell, Chief Revenue Officer at Vistatec
What has changed operationally
The work itself looks different now. Translation used to be a clean transaction. A vendor produced a set number of words at an agreed rate. Today localization covers far more than text.
A single project may involve voice, video, subtitles, regional testing, cultural adaptation, and platform integration. Each layer shapes whether the content performs in its target market. A per-word rate captures none of that complexity.
Content mix also matters more than raw volume. A marketing headline and a regulated legal notice carry very different risks. AI handles some content types well and others poorly. So effort now tracks risk and the level of human correction, not word count alone.
Edit distance shows this clearly. It measures how much a person changed the machine output. A clean AI draft needs little correction, while a sensitive one needs heavy rework. Word count treats both the same, which is exactly the problem.
What buyers still need to pay for
Cheaper output does not mean better output. When extra translation costs almost nothing, teams tend to produce more of it. Quality can then erode without anyone noticing. Volume is easy to generate, but trust is definitely not.
The real value has also moved. Clients no longer pay mainly for words. Instead, they pay for judgment and accountability. Someone must decide what translates to good quality in each market and own the result.
That accountability still costs money. A specialist who signs off on high-stakes content carries real responsibility. This does not shrink to zero just because a model wrote the first draft. Buyers therefore still pay for several things AI cannot supply on its own:
- Expertise in high-risk content such as legal, brand, and regulated material
- Accountability when an error carries commercial or reputational cost
- Governance over AI use, data, and quality across every language
- Measurable outcomes tied to real performance in each market
Governance has become a real cost line of its own. Regulators now watch how companies use AI on customer-facing content. The EU AI Act, for instance, raises the bar for transparency and oversight. So someone must manage AI quality, data, and compliance across markets. That work can only be handled by real people, not AI models.
What better localization pricing models could look like
No single structure has replaced per-word pricing yet. Several credible alternatives are taking shape, though.
Most tie cost to risk, outcomes, or ongoing capacity rather than raw volume. Buyers and providers are currently testing a few main options:
- Risk-tiered pricing. High-stakes content carries a premium for human accountability, while low-stakes content runs on lighter automated workflows.
- Program or subscription fees. A recurring fee covers a defined scope and capacity, which gives finance teams predictable costs.
- Hybrid pricing. A base platform fee sits alongside metered usage, so predictability and growth stay in balance.
- Outcome-based pricing. Charges connect to measurable impact, such as conversion or market performance.
Each of these options carries real trade-offs. Subscriptions, for example, can push providers toward high volume at thin margins. That pressure risks the same quiet drop in quality. Outcome-based deals, meanwhile, need shared definitions of success and clean, trusted data. Buyers should therefore weigh predictability, risk, and accountability together, not price alone. You can see how these choices play out in practice across AI adaptive workflows.
Tie the commercial model to the outcome
Changing the pricing model means changing how you define value. A word count measures output. However, it says nothing about whether the localized content performs. That gap is the real question for senior leaders.
So start with the outcome, then work back to the commercial. First, ask what each content type must achieve in market. Next, decide where human accountability is essential. Then choose a model that pays for those things directly.
The market is already splitting along these lines. Some research describes a K-shaped shift, where commodity translation compresses while outcome-focused content solutions grow.
Providers that still compete on words alone sit on the wrong side of that divide. The winners instead price the parts of the work that AI cannot replace.
A value conversation, not a rate conversation
The pricing debate is really a value debate. Word counts will keep losing meaning as AI absorbs more routine work. Consequently, the models that last will price expertise, accountability, and results.
If you are reassessing how you buy or sell localization, test these options against your own content and risk. Vistatec can help you design a commercial approach that fits today’s localization workflows. Start a conversation with our experts today.