Every company’s situation is different in terms of the scope of work involved, languages targeted, etc. This is intended as an exercise to help you understand roughly what you’ll need to plan for.
Experimental Localization
You’ll generally have a lot less in app content than you do on your website, help center, etc. You can think of app localization as a form of marketing or market testing. Also, in some markets like Japan, not being localized is a direct barrier to adoption and usage. You can quietly turn on new interface languages and set up a before|after AB test to evaluate how key metrics for adoption churn, upgrades, etc changed as a result.
How much will this cost?
A typical app will have somewhere between 5,000 to 20,000 words of in app content, towards the higher end if the app is more complex or has embedded help content. Translation services are generally priced by the word. Pricing is a function of how much review is involved, and which languages are targeted (some languages are easier to staff than others).
The gold standard in terms of quality is a four step process that includes the following steps:
- Translation : First pass translation (often seeded with a machine/AI translation
- Editing : A reviewer checks this translation and edits it as needed
- Language Lead Review : A language lead who is also an expert user of the product signs off on this work
- In Context QA : The language lead may also test the app interactively to look for UI bugs, layout issues, etc.
I tell customers to expect this to work out to 25-40 cents per word when all of these steps are factored in (the actual cost will be lower, especially if a blended human|AI process is used but its best to over-estimate a bit for budgeting purposes). That works out to between $2,000 to $8,000 per interface language. That typically works out to a next extra 20 to 80 paid seats for a typical SaaS product for top line breakeven. You read that correctly. It is one of the best growth levers, and relative to the revenue it enables, dirt cheap. Some of the companies I have worked for generated well over half of their revenue outside the US.
Can’t we just get AI to translate everything for free?
We are big fans of hybrid machine + human translation approaches. That said, it is generally not a good idea to rely on unsupervised machine translation especially for high visibility content. Machine translation has improved a lot. The risk isn’t that it will translate “Sign Up Now!” as “Your Mom Has Sex With Goats!”. The issue is that native speakers will pick up that the translation is machine generated (translations are often decent but stilted or robotic sounding, and also contain grammatical errors that look sloppy to native speakers). Not what you want at the top of a signup funnel.
Another issue is that in app content often consists of short phrases or single words. These are difficult for machine translation engines to deal with so they will often translate a noun phrase as a verb and vice versa (human translators get confused too, but they know to ask for clarification in situations like this). This is particularly an issue if English is your source language because English is gender neutral and has simple pluralization rules among other things. AI also has a difficult time translating into low resource languages, because there is limited training data and it is often polluted by AI/MT generated text (garbage in → garbage out).
TL;DR you can definitely use machine translation to pre-populate translations and for UI testing, but you’ll want to have humans in the loop to tweak or rewrite them. As AI translation improves, reviewers will have to make fewer changes, so unit costs will come down. Most translation management systems have AI|machine translation as a built in option for workflows. Even better, they intelligently route to different providers depending on the language targeted because some AI|MT platforms do better than others with specific languages.
Caveat: this is all a moving target right now with the rapid evolution of LLMs.
What language(s) should you start with?
I recommend picking a couple of languages. One language, like Spanish, that is widely spoken in many regions (over 30 countries and large parts of the US). Plus another language spoken in a region where you are seeing organic demand. At Notion, for example, we had huge user communities in Korea and Japan, even when our app was English only. We figured that adding these languages could only help. Korea and Japan are two of the company’s top regions today.
For further reading see
Which Languages Should We Target?
360 Degree Localization
The next step up from experimental localization is to localize all customer touch points. This typically includes:
- Company website and marketing content
- Help center
- Lifecycle emails and communications
- Sales supporting assets (brochureware, etc)
- Video catalog
The amount of material involved varies by company, but this will typically work out to somewhere between 100,000 to 1,000,000 words of source material. You’ll want to use different translation workflows for each content type because some will require more review and attention than others.
For high visibility | high impact content such as your marketing site, sign up funnel, etc, you’ll be looking at around 20 cents per word, or more you hire a transcreation agency. For your long tail help center content, you can use a less intensive process called AIPE (AI + post editing), typically at a cost of 10-cents per word or less. For rough estimation purposes I tell clients to expect an average of 20 cents per word, for a range of $20,000 to $200,000 per language depending on the volume of material to be translated. That works out to 200 to 2,000 net extra paid seats for a typical SaaS product. In the grand scheme of things, and compared to other international operating costs like opening sales offices, dirt cheap. It’s also a pre-requisite for success in important markets like France, Japan, Korea and much of Latin America.
💵 The best way to minimize translation costs is to reduce the amount of material that needs to be translated in the first place. Company help centers are often bloated and as we know most people don’t RTFM. As you embark on localizing your product, pay special attention to your help content and do the following:
- Migrate away from a web based help center to in app help. This is often a better user experience. Also the constraints of displaying help in app encourage content authors to keep it short and to the point.
- If you need to explain a feature in a detailed document, that might be a sign that the app itself is confusing users and can be functionally improved, thereby reducing the need for documentation.
- Be ruthless in editing and curating your knowledge base. You can probably cut the amount of content in half without negatively affecting users. If you have a team whose job is to create content, they will create content (that’s their job!).
- If you have an active customer base evangelizing your products, leverage them to provide in country support, tutorials and consulting. We did this with great success at Notion. Scaling customer support in other countries is hard and this is a good way to provide better options, especially in the early days of international expansion.
Extrapolating from here we can work out roughly what the long term budget for a localization program will be (I am just tracking direct translation costs here).
A typical company that aspires to operate globally will eventually end up in 10 to 20 languages. That is enough to reach the majority of markets that matter to them. That works out to between $200,000 to $4,000,000 depending on the volume of content and number of languages targeted. Most companies will scale this up over several years, and will focus on low hanging fruit first (e.g. regions where they already have traction).
💶 Another way to reduce translation costs is to use unsupervised AI|machine translation for low visibility content, while routing high visibility content to human translators and reviewers. One of the things you can do with good process automation is to rank assets based on the amount of traffic they are getting and set a threshold to trigger human translation and review. Chances are good you have a long tail of infrequently accessed content that you can machine translate, and then if traffic levels pick up, have humans clean up the AI translations as needed.
Analytics : Ranking Strings By Visibility
Technology Spend
As we discussed in Selecting A Translation Management System the cost of translation technology will generally account for 10-15% of the overall program budget. The unit costs decrease with content volume. For a typical mid-size localization program, you will probably end up spending $25,000 to $100,000 per year on tech, and several times as much on translation, review and program overhead.
Measuring ROI
Measuring the direct ROI from localization is often tricky because many other factors besides language can affect adoption and monetization in another region, including:
- Cultural differences and local business practices
- Lack of market awareness in a region
- Currency and payment method support
- Differing price sensitivity
One thing you can do to get some sense of how language support alone is impacting your business is to quietly enable interface languages and measure how key usage metrics such as churn and paid activations change over time. This uplift will vary by region. For example, in regions where the rate of English proficiency is relatively low, you should see a stronger signal than in regions like the Nordics where most users are fluent in English.
Localization alone is just one piece of operating internationally. It unblocks users, but it doesn’t necessarily guarantee that your product will sell itself. If your product experiences organic growth in international markets without marketing support, that is a very strong demand signal. That said, for most products the impact of localization is more subtle and compounds over time.
Special Cases
There are some special cases to consider that go beyond the scope above. Two common examples are user generated content and dynamic content.
Social media sites like Facebook allow users to translate their feed. This is typically done using machine|AI translation on an as-is basis. It’s just not practical to use human translation for this.
Another case is dynamic content such as store listings, hotel inventory, etc. This can involve thousands or tens of thousands of assets that need to be translated. There are a number of strategies that can be used here.
If you fall into either of these situations, feel free to reach out to brian@loctechpartners.com