78% Cheaper And Five Times Faster: How We Optimized The Software Localization Process
Here at 1Ci, we build solutions for business automation that include developer platforms and applications for end users. Today our main products are translated into 13 different languages.
The growing volume of texts, number of software products to localize and number of employees and contractors conducting localization (there are 74 of them in the team now) forced us to re-think our localization toolbox as well as our approach to business processes. Today we will talk about these changes and the results they have brought.
Automation for tactical and strategic tasks
Before we started optimization, we had our own proprietary tool for translation automation. It worked just fine with relatively small chunks of text, but when we needed to translate big pieces like ERP system configurations, its response time could exponentially grow to dozens of minutes.
This was not the only problem with using this tool. As it was custom-made, our translators, content employees, and contractors needed to study it. This was not our competitive advantage as it is always more comfortable for a specialist to work on a project when he or she uses only well-known instruments, and won’t need to learn new tools which are used by only one company. Those who agreed to do so, usually, raised their financial demands significantly. We had to switch to something else.
After a thorough analysis, we chose a Smartcat system. It is a powerful translation solution and a marketplace where companies can find and recruit localization specialists. It is good for us for several reasons:
- The system features useful functions we might need during localization, such as a cross-product glossary.
- It is ready-made for automation and optimization. For example, we can enable an auto-complete of a similar phrases function which helps in speeding up the translation process.
- There is zero cost for adding a new translator to the system, as we can always pick the ones who’ve worked with this tool from Smartcat’s marketplace.
- Employees’ productivity increased due to the widespread usage of the translation tool Smartcat, instead of the custom solution.
However, it was not enough to switch from an in-house translation tool to succeed. Some of our products were not initially designed with future localization in mind. This meant that very often it was hard even to locate all information needed for the localization in different places in applications. That is, we’ve got a useful third-party tool for translation but had to upload more useful data into it.
In order to do so, we have developed a utility which searches for the data in the development repository for a specific notion (product, etc.) we need to translate. It significantly reduces the overheads the localization team faces every day. Also, it clearly shows the importance of localization-driven software development. Now our development teams are making their code ready for automated localization even if it is not part of the project from the beginning. Our tool can also check if most of the requirements for such multi-language support are satisfied.
Fighting the need for unnecessary work
Before we started using Smartcat, we quickly understood how big the amount of repetitive content was. Very often, it turned out that our employees were working on the localization of texts that had already been translated. The amount of unique content was up to five times less than the number of pieces processed by the localization team.
Features like a cross-product glossary and the automatic translation of pieces of text that had been previously translated allowed us to speed up the overall localization process and reduce costs by up to five times.
The new data we were able to gather has also affected our business processes. For example, when working on the localization for a new product, a piece of software or feature, we used the Translator-Editor-Proofreader scheme. Initially, the Translator was always a target native language speaker who understood the source language (e.g., Italian interpreter with a knowledge of English).
This approach turned out to be ineffective, as very often such translators could not fully understand the text written not in their native language. Our topic (business process automation) is hard to understand, and many translators failed to deliver localized content that fully corresponded to the source. We struggled with distorted translation.
To fix this problem, we’ve changed our approach, and now the first specialist to touch the text is the translator who is a source language native speaker who knows the target translation language. Statistics show that this has allowed us to reduce the number of errors and distortions significantly.
Another major improvement that has allowed us to reduce localization costs significantly was the change of approach to the translation of content for different types of users. At 1Ci, we work on 1С:Enterprise platform that can be used for building business applications. In the past, we tried to localize the content both for end users and apps developers including sections such as Help and FAQ. The amount of content to translate to several languages was huge.
Now we do it differently. We still think that the end user should be able to access the interface in his mother tongue, however software developers should know English too. Thus, there is no need to localize the system with all the Help database for, say, Italian developers, if we can do it once in English. This new approach has shown its effectiveness — we significantly reduced the localization team’s load, and developers using our platform are still happy.
Save on the hiring process in a smart way
One more thing we were able to fix is the hiring process. When you have a product to translate into multiple languages there is a problem of translation fraud. Say, if you need Polish translation but no-one in your team speaks Polish, some malicious translator may want to try to trick you and submit Google translated text and get paid for it. Our stats show that every sixth application for translation jobs is fraudulent.
To fight such fraudsters in the first stages of the hiring process and save time and money, we’ve developed a custom test which allows us to identify cheaters even without the help of the target language native speaker. This test is a text to translate that contains some indicators that clearly let us know whether the person translating it really did the job or not.
For example, it contains the names of our products which if misspelled, show the lack of effort put into completing the task. For example, our product names have a specific format 1C:Product name — without space. Automatic translation software also adds a space after the colon. Also, we may include in the test misspelled names of third-party products like Gogle or CQL Server — if the translator fails to identify such mistakes, it is highly likely that we can’t work with him.
Only if this initial test is passed will we then show the result to our native speakers or partners in the target country who will provide us with the final feedback. And, of course, we’ve implemented an automated check on whether the text has been automatically generated. On average, we receive 70–80 applications for every translation job — 10 of each are usually copied and pasted Google translate output.
The final decision on every translator is made by 1C partners in target countries, as they will use the content produced by this employee to sell our products.
All these improvements have allowed us to achieve significant results:
- Employees’ productivity has increased due to the widespread usage of the translation tool Smartcat, instead of the custom solution.
- Our tool for automatic data sourcing and tips generation allowed us to speed up the localization process to be five times faster.
- The cost of the typical translation project has decreased by up to 78%.
- We’ve managed to shrink the number of fraudulent and low-quality job applications that make it to the final stage of the review by 100%.