From Agency to In-House: Digital Transformation for Translation

Join Smartcat for our 45-minute webinar and learn how major organizations are leveraging traditional project management methodologies + the latest AI tools to smooth the transition from agency to in-house translation management.

Q: In the AI front in house translation model, who holds responsibility for damages occurred as a consequence of a bad translation?

A: In the AI first model, where everything is in-house, responsibility lies with the team managing the translations. If you are handling the translations yourself, you are accountable for any issues that arise. This is particularly relevant when dealing with major corporations, as translation issues are not typically flagged in isolation.

If there is an issue with a translation, the responsible party is the team that either creates the content or is specifically tasked with translation. For organizations with dedicated translation teams, this accountability is clear.

Furthermore, if you are utilizing a marketplace or working with your own translators, it's essential to track who is responsible for each task. When working with freelancers, they do not receive payment until you are satisfied with their translation. For instance, in a classic model like ours, we leverage a community of freelancers to translate content into various languages.

When using AI to recommend translators, it’s important to note that many clients have already established relationships with specific freelancers based on prior work. In the initial stages, after receiving AI-generated translations, you can hand these over to a freelancer from the SmartCat community. Typically, their turnaround time is around 24 hours.

After receiving the translation, you have the opportunity to review and validate it. The project phase is not considered complete until you confirm that you are satisfied with the accuracy of the translation. Only then can you proceed to publish the content and finalize payment to the translator.

Q: How does tools like smartcat.ai perform with local languages, where the datasets are smaller?

A: We support over 250 languages and utilize a translation engine that primarily relies on established machine translation rather than generative AI tools. This approach minimizes the risk of hallucination, as we draw from machine translation systems that have been in existence for over 20 years. However, challenges can arise when working with local languages due to smaller datasets.

A practical solution involves using machine translation to achieve around 95% quality, followed by handing the output over to a professional translator for final validation. This process allows the translator to make necessary adjustments, which are then incorporated into the translation memory. Consequently, future translations of similar content can be improved based on these enhancements.

An example of this methodology can be seen in Seattle, where local government agencies are leveraging Smartcat to communicate effectively with immigrant communities. By allowing community members to validate translations in their native languages, trust and engagement are fostered between the government and these communities. This collaborative approach not only ensures accuracy but also empowers individuals to confirm that the translations reflect their linguistic needs.

Q: What is AI first in-house translation?

A: AI first in-house translation refers to a model where organizations utilize artificial intelligence to generate initial translations, which are then refined by human translators. This approach leverages established machine translation technologies to achieve high-quality results while reducing costs and turnaround times. By integrating AI into the translation process, companies can enhance efficiency, maintain control over their content, and foster collaboration with freelance translators within a streamlined platform.

Q: My company experience with Smartcat First-time AI Accuracy is that it is dictated by post-editor’s personal decisions. If the post-editor wants to have more translation work (meaning costs for us), they edit the AI translation on their own. In our case, the post-edit rate was really high meaning that the AI accuracy was really low. In the extra cost vice, we are in hands of post-editor’s own decisions.

A: There are a few options to address the situation where you feel a freelancer or post-editor is deliberately inflating costs by making unnecessary edits to AI-translated content:

1. Assign the same project to a different freelancer for comparison. Smartcat can recommend a translator to do the post-editing. Compare the results of the existing freelancer, who may be deliberately mistranslating to generate more billable hours, versus the new freelancer serving as a test case. This provides two data points to analyze.

2. Utilize Smartcat's AI validation to assess the initial machine translation quality. Smartcat estimates the translation quality upfront, often at 95-99% accuracy. You can then review where the post-editor is making changes - are they editing areas the AI identified as needing review, or making changes all over the place? If it's the latter, a conversation with the post-editor may be warranted, and you may need to replace them.

3. Run a comparative test by assigning the project to another freelancer to establish a benchmark. This provides a data-driven basis to have a conversation with the existing post-editor about the discrepancy in edits and associated costs.

By leveraging Smartcat's AI quality assessment and the ability to easily assign projects to multiple linguists for comparison, you can identify situations where a post-editor may be inflating costs through unnecessary edits. This data can inform discussions with the post-editor and guide decisions about replacing them if needed to control costs.

Q: How did you calculate the cost saving?

A: In our collaboration with clients, such as a global financial services company, we analyze their current spending on agency translations. We take their existing costs and break them down into the expenses associated with our platform, which operates on a per-word fee model for AI translation.

To calculate the total cost, we also consider the expenses they would incur when hiring freelancers for final quality assurance on the remaining 5% of the translation that requires human oversight. By adding these two figures together, clients can compare the total cost of using Smartcat against what they currently pay an agency.

We provide a cost calculator on our homepage to assist clients in performing these calculations. 

While the savings can vary—often cited as 70%, 75%, or even 78%—this serves as a baseline. Clients can then evaluate their specific circumstances to determine actual savings.

Ultimately, having concrete numbers from the outset allows for significant cost savings compared to traditional agency models.

Q: Any advice on how to make the transition smooth and "ethical" from the outsourced translators to in-house AI translations? One of the barriers that I'm encountering for this transition is the relationships with the translators who've been working with me for the last 10+ years in the same company.

A: That's an interesting question. Our marketplace of freelancers includes over half a million translators, which accounts for nearly two-thirds of all freelance translators globally. There’s a high probability that the translator currently working with your agency is also part of the Smartcat marketplace. If they’re not, they can sign up and join the community within a day.

If you have freelancers you've worked with in the past and want to continue collaborating, you can do so using Smartcat. They will likely earn more since the agency middleman won’t be taking a significant cut from their earnings.