The trap is applying one tier to everything. Full review on 40,000 UI strings is waste. No review on your pricing page is a lawsuit.

Why does MTPE often fail to save money?

Because post-editing can cost more than translating from scratch when the MT output is bad. A linguist fixing scrambled machine output spends time unlearning before rewriting.

The fix is measurement. Track post-editing effort per segment using edit distance or time-on-segment, not just the invoice. When a content type consistently needs heavy editing, that content type should move up a tier, not stay in the cheap lane with a permanent cost overrun.

Rules that hold in practice:

• If a linguist edits more than 60% of a segment, MT added nothing. Move the content type up.
• If a linguist edits under 15%, the content can probably drop a tier.
• Watch short strings. UI labels under five words have the worst MT quality ratio because there is no context.

How do you build the three tiers?

Map every content type to a tier once, in a table, and put that table in your localization brief.

Two-thirds of a typical SaaS company’s word volume lives in tier 3. That is where the savings are.

How do you make tier 3 safe?

MT-only is acceptable when the failure mode is cosmetic. It is not acceptable when a wrong translation changes what a user does.

Guardrails for tier 3:

1. Lock the glossary. Product names, feature names, and UI verbs must be do-not-translate or fixed translations. Unlocked glossary is what produces a settings menu in three different languages.
2. Exclude interpolation placeholders from translation. Broken variables in translated strings are a common production bug.
3. Spot check 5% of every batch in each language, chosen randomly. Catch systematic errors, not one-offs.
4. Pilot with one language for four releases before widening. A tier-3 policy that works in French may fail in Japanese.

How much does this actually save?

Worked example: 100,000 words a quarter.

• All tier 2 at $0.15 a word: $15,000.
• Rerouted as 20k tier 1 at $0.22, 30k tier 2 at $0.15, 50k tier 3 at $0.06: $4,400 + $4,500 + $3,000 = $11,900.

That is a 21% reduction, not 50%. The bigger savings come from the second effect: fewer review cycles and fewer post-launch fixes in languages you can barely audit. Teams that measure total cost including rework typically report 30-45% reductions.

Be skeptical of anyone quoting 70% savings. Those figures almost always come from comparing MT to full human translation while ignoring the review labor that keeps MT publishable.

What breaks when you push too cheap

Support content in tier 3. Users reading a badly translated help article file the same ticket twice and lose trust in the product. Help content deserves tier 2.

Screenshot and video captions. Translating captions without updating the screenshots produces instructions that reference UI that no longer matches. Localization is not only text.

Right-to-left languages. Arabic and Hebrew need layout work, not just string translation. A tier-3 string that overflows a button is a design bug MT cannot fix.

How do you keep quality visible?

Report two numbers quarterly: cost per word and user-reported localization issues per language. If cost drops and issues stay flat, the routing is working. If cost drops and one language’s complaint count climbs, that tier assignment was wrong.

That pairing is the whole governance model. Cost alone is a metric that always looks good right before quality collapses.

Key Takeaways

• Route content into three quality tiers instead of reviewing everything identically.
• Tier 3 MT-only is safe when the glossary is locked and placeholders are excluded.
• Measure post-editing effort per segment; above 60% edits means MT added nothing.
• Realistic savings are 30-45% blended, not the 70% vendors quote.
• Track localization complaints per language alongside cost per word. Cost alone hides quality collapse.

Translation workflow options are covered on the Welocalize page at iAIFeed.

That framing matters. If you want to translate a blog post in ten minutes, this is the wrong product. If you need 40 locales updated every release with consistent terminology, it is the right shape.

What does Welocalize actually do?

Four services matter for software companies:

• Machine translation post-editing (MTPE) — MT output reviewed and corrected by a linguist. The standard production path.
• Terminology and style guide management — the glossary that keeps “workspace” translated the same way in 14 languages.
• Continuous localization — API and repository integrations so strings translate as they change.
• Multimodal work — subtitling, dubbing, and audio localization for marketing video.

The differentiator versus DIY MT is the terminology layer and the review step, not the translation engine itself, which is often a commercial model anyone can license.

How does a localization program get started?

Expect four to eight weeks before the first production delivery. The sequence below is what makes that timeline real.

1. Inventory your content. Count words by content type — UI strings, help center, marketing, legal. Legal text is a separate, slower track.
2. Pick two launch languages, not eight. Usually your biggest non-English market and one where you have support coverage. A language you launch without support becomes a complaint channel.
3. Build the glossary first. 100-200 terms, with do-not-translate lists for product names. This single asset determines long-term quality more than any engine choice.
4. Connect the content system. A CMS or repo integration beats file exchange. If you are on Contentful or a Git-based docs setup, the connector saves hours per week.
5. Set the quality threshold by content type. Marketing copy gets full human review. UI strings can run MT-only with spot checks. Help center sits between.
6. Review the first 200 strings yourself even if you do not speak the language — check that product names stayed untranslated and that formatting survived.

How good is MT with post-editing?

Good enough for UI and support content in major languages. German, French, Spanish, Japanese, and Korean all produce clean output with a competent post-editor.

Weaker in three situations:

• Marketing headlines and puns. Wordplay does not survive. These need transcreation, which is a copywriting service priced separately.
• Long compound nouns in German and Dutch. Odd but usually comprehensible. Native reviewers will rewrite them.
• Low-resource languages. Thai, Vietnamese, and most African languages still need near-full human translation for quality marketing content.

What does it cost?

Welocalize does not publish rates, and that is a genuine friction point. Typical enterprise language service pricing per word runs from about $0.06 for MT-only through $0.12-$0.20 for post-edited translation, with transcreation well above that. Minimum commitments usually apply.

For planning: 50,000 words a quarter across four languages lands in the tens of thousands per quarter. Compare that against a distributed freelancer bench before signing — for two languages and stable volume, freelancers plus a terminology sheet can be cheaper.

Where a vendor like this beats DIY

Volume and consistency. Once you are past roughly five locales and shipping monthly, the coordination cost of freelancers — chasing invoices, inconsistent glossaries, timezone lag — eats the rate difference.

Compliance is the other one. Regulated industries need documented review trails and named translators. A vendor provides that by contract; a freelancer does not.

Where DIY wins: one or two languages, low volume, and content that changes often enough that a formal review cycle slows you down.

What should you ask before signing

• Which engine do you use, and can we swap it mid-contract?
• Is the glossary ours to export, in a standard format?
• What is the turnaround for a 5,000-word urgent batch?
• Who are the named linguists for our languages, and can we request changes?

That last question is a quality signal. Vendors with a stable reviewer bench answer immediately; those running on a rotating freelance pool get vague.

Key Takeaways

• Welocalize is an enterprise language services provider built around MT plus human post-editing and terminology management.
• Expect four to eight weeks to a first production delivery; start with two languages, not eight.
• The glossary is the highest-leverage asset — build it before connecting content systems.
• Typical per-word pricing runs about $0.06 for MT-only to $0.12-$0.20 for post-edited work.
• Best fit is five-plus locales updated monthly; for one or two languages, freelancers are cheaper.

Localization tooling comparisons are on the Welocalize page at iAIFeed.