This guide expands the workflow section of A Practical Localization Strategy for Growing Products—part of the Localize the World series, where host Linh Nguyen spoke with Stas Kharevich, Head of Localization, Alconost.
Stop asking whether human or machine translation is better
Professional translators bring contextual understanding, cultural awareness, subject knowledge, and editorial judgment. Machine translation handles volume, speed, and frequently changing content. Neither method wins every time—the question is which mix fits each content type.
The real question is not:
Should we use human or machine translation?
It is:
Which workflow gives this content the required quality at a reasonable cost and speed?
How do teams choose between human and machine translation?
Content volume matters, but it is only the beginning. Visibility, purpose, business impact, and required quality should also influence the workflow.
In the clip below, Stas Kharevich explains how content profiling helps teams choose between human translation, machine translation, and combined approaches.
On visibility tiers, Stas Kharevich's rule of thumb:
Nowadays it's popular to divide content by visibility—high, medium, or low. High-visibility content influences retention and acquisition, so you need to focus on quality. For low-visibility content with high volume, automated workflows can be absolutely okay when the output quality is designed and monitored.
Begin with content profiling
Content profiling means classifying content according to the role it plays in the customer journey and the business.
Useful criteria include:
- Who will see the content?
- How visible is it?
- What action should it encourage?
- What would happen if the translation were imperfect?
- How frequently does the content change?
- How long will it remain in use?
- How much content is there?
- Does it contain specialist, regulated, or sensitive information?
- Does it express the company’s brand voice?
- Will users rely on it to complete an important task?
Answering these questions helps the team avoid two expensive mistakes:
- Paying for an unnecessarily intensive process for low-risk content.
- Using insufficient quality controls for content that affects revenue, trust, or usability.
A three-tier content model
One practical way to choose a workflow is to divide content into three visibility tiers.
| Tier | Examples | Typical workflow |
|---|---|---|
| High visibility | Product UI, homepage, ads, onboarding, app-store copy | Human translation, transcreation, subject-matter review, LQA |
| Medium visibility | Blog posts, secondary pages, educational content, routine marketing | Machine translation + human post-editing (MTPE) |
| Low visibility | Internal docs, large knowledge bases, archived support, rare service messages | Automated translation and automated review where risk is low and output is monitored |
Tier 1: High-visibility content
Tier 1 content shapes first impressions and core product use—product UI, homepages, ads, onboarding, checkout, and app-store copy. Errors here hit acquisition, conversion, and trust. The wording may need to sound natural, persuasive, and on-brand; a technically correct translation is not always enough.
Typical workflows include professional human translation, transcreation, subject-matter or in-country review, and linguistic testing in the product. Machine-assisted translation with extensive review can work when a qualified person owns the final result. Translation memories, glossaries, and automated QA still support the process.
Tier 2: Medium-visibility content
Tier 2 is customer-facing but carries less immediate commercial risk—blog posts, secondary pages, documentation, newsletters, and much help-center material. It should be clear and accurate without the same editorial investment as a homepage or campaign.
Machine translation followed by human post-editing (MTPE) often works well here: the engine produces a first draft; an editor fixes meaning, terminology, and readability; automated checks catch formatting issues; the content owner reviews sensitive passages. A flagship report may need publication-quality editing; a routine update may only need to be accurate and readable. See also how AI fits into localization workflows.
Tier 3: Low-visibility content
Tier 3 is unlikely to shape brand perception or block a critical user action—internal docs, large knowledge bases, archived support articles, and low-traffic informational pages. For suitable content, automated translation with monitoring may be enough.
Common approaches include raw machine translation, MT with automated review, selective human sampling, or human review only for flagged or high-risk segments. Some teams use one AI system to translate and another to review output. The absence of full human review does not mean the process should be unmanaged: the team still needs to test the workflow, define acceptable quality, and monitor results.
Other factors that should influence the decision
Visibility is useful, but it should not be the only criterion.
Content volume
A few thousand words can often be handled efficiently by professional translators.
A few hundred thousand words may require a mixed workflow to remain financially and operationally sustainable.
Large volumes increase the value of:
- Machine translation
- Translation memory
- Terminology automation
- Content reuse
- Automated quality assurance
- Workflow integrations
Update frequency
A static brochure and a continuously updated software product require different processes.
Frequently changing content benefits from automated transfer between the source system and the translation platform. Machine translation may also help teams keep up with rapid publication cycles.
Content lifespan
Content that will remain visible for years may justify greater editorial investment.
A temporary status message, short-lived campaign variant, or low-traffic article may not need the same treatment.
Repetition
Translation memory can reuse previously approved translations for repeated or similar text.
This improves consistency and may reduce the cost of human translation, particularly for interfaces, documentation, and frequently updated product content.
Subject matter and risk
Legal, medical, financial, cybersecurity, and safety-related content may require specialist human review regardless of its visibility.
A low-traffic legal document may carry more risk than a widely read lifestyle blog article.
Language pair
Machine translation quality differs by language pair, domain, and content type.
A workflow that works well for one target language may produce weaker results in another. Teams should test actual content rather than relying on general claims about a particular engine—see how different translation models perform on real content.
Available reference materials
Machine and human workflows both perform better when supported by:
- A glossary
- A style guide
- Translation memory
- Clear product context
- Approved examples
- Content screenshots
- Instructions for tone and audience
A lack of context can undermine even an otherwise strong workflow.
What is machine translation post-editing?
The tier model above often points medium-visibility content to MTPE. Here is what that means in practice.
Machine translation post-editing combines automated translation with human review.
The machine produces the first version, and a professional editor improves it.
There are different levels of post-editing.
Light post-editing
The objective is to make the translation accurate and understandable.
The editor focuses on:
- Meaning errors
- Omissions
- Incorrect terminology
- Grammar problems
- Misleading wording
They may leave stylistic imperfections that do not interfere with comprehension.
Full post-editing
The objective is publication-ready quality.
The editor also improves:
- Fluency
- Tone
- Style
- Naturalness
- Brand consistency
- Sentence structure
Full post-editing can approach the quality of human translation, although the effort required depends heavily on the quality of the initial machine output.
When human translation is the safer choice
Human-led workflows remain especially valuable when the content:
- Expresses the company’s brand voice
- Relies on humor or wordplay
- Needs cultural adaptation
- Is legally or commercially sensitive
- Contains ambiguity
- Must persuade the reader
- Will be highly visible
- Requires creative rewriting
- Affects a critical product action
- Could create significant harm if misunderstood
Marketing campaigns are a common example. The translator may need to adapt the message rather than reproduce its literal wording.
This is often closer to copywriting than conventional translation.
When machine translation can be a strong choice
Machine translation may be appropriate when:
- The content volume is very large
- Publication speed matters
- Content changes frequently
- The material is repetitive
- The content has low visibility
- The primary goal is access to information
- The audience accepts functional rather than polished language
- A human review layer is available for important exceptions
- The workflow has been tested for the relevant language pair and domain
Machine translation is not only a cost-saving measure. In some cases, it makes localization possible for content that would otherwise remain untranslated.
Build a decision matrix
Teams can use a simple scoring model to select the workflow.
Rate each content type from low to high across these dimensions:
- Visibility
- Business impact
- Brand sensitivity
- Risk
- Volume
- Update frequency
- Required speed
- Required linguistic quality
| Content profile | Suggested workflow |
|---|---|
| High visibility, high risk | Human translation plus review |
| High visibility, creative | Transcreation or specialist human translation |
| Medium visibility, moderate volume | Machine translation plus full post-editing |
| Medium visibility, high volume | Machine translation plus targeted human review |
| Low visibility, low risk | Automated translation with monitoring |
| Specialist or regulated | Subject-matter translator and specialist review |
The matrix should guide the decision rather than replace professional judgment.
Test workflows using real content
A workflow should not be selected based only on a tool demonstration or a generic translation-quality score.
Run a pilot using representative content.
A useful pilot should include:
- Several content types
- Real target languages
- Approved terminology
- Human evaluation
- Review in the final context
- Measurement of editing effort
- A comparison of speed and total cost
- Documentation of recurring errors
The final context is particularly important.
A translation may look correct in a spreadsheet but fail inside the product because it is too long, appears beside the wrong visual element, or does not match the user’s action.
Measure total workflow performance
Cost per word should not be the only metric. A cheaper first step may create additional costs through heavy post-editing, internal review, rework, engineering intervention, customer complaints, inconsistent terminology, or delayed publication. See the localization cost guide for what actually drives total spend.
Useful metrics include:
- Turnaround time
- Human editing time
- Error rate
- Rework rate
- Reviewer satisfaction
- Terminology compliance
- User-reported issues
- Cost per completed content type
- Percentage of content requiring escalation
- Performance in the product or market
The goal is to identify the workflow that performs best overall.
The best workflow is usually mixed
Most growing companies do not need to choose one translation method for the entire organization.
They need a portfolio of workflows.
High-impact content may receive professional human translation. Medium-priority content may use machine translation with post-editing. Low-risk content may use automation with selective monitoring.
This approach directs human expertise toward the content where judgment matters most while allowing technology to handle volume and repetition.
The most effective question is not whether machines can replace human translators.
Where does human judgment create the most value, and where can technology help the team move faster?
For a broader framework covering content planning, localization maturity, outsourcing, and workflow design, see A Practical Localization Strategy for Growing Products. Related guides cover why localization starts earlier than many teams expect and what to keep in-house vs. outsource.
Stas Kharevich has spent more than 16 years at Alconost and leads localization for product and software teams—SaaS, games, and marketing content among them. The video clip in this guide is from his conversation with Linh Nguyen on the Localize the World series. Read more from Stas on how Alconost runs localization.
