Solving the Problem of Translation with AI in a Supporting Role

Over time, the insurance landscape has evolved to accommodate a diverse array of consumers, each with unique needs and preferences.

Today, 67.8 million people living in the United States speak a language other than English at home; and over 25 million are considered to have limited English proficiency (LEP)—defined as individuals who do not speak English as their primary language and who have a limited ability to read, speak, write or understand English. Given that 76% of consumers prefer to buy products from businesses that serve them in their language of choice, and 40% won’t even consider buying from those that don’t, it has become increasingly clear that ensuring the satisfaction of non-English speakers is imperative for conducting business.

Insurance products are inherently complex, often characterized by confusing terms, conditions and coverage options. From health insurance policies with varying deductibles and copayments to life insurance plans with nuanced clauses, the complexities can be overwhelming, especially for those whose primary language differs with that used in the insurance documents they receive. Translating insurance communications and documents into different languages to support LEP populations can alleviate the struggle to understand these materials and make better choices. The challenge, of course, is that translation requires high accuracy to avoid misunderstandings or misinterpretations that could lead to costly errors or legal disputes.

Translation is often outsourced to third-party providers, which can be incredibly expensive and time-consuming. Whether outsourced or managed by internal teams,  the process is generally the same. Materials are translated one at a time and the process is managed using a patchwork of disconnected tools such as email and Word documents. This can produce inconsistent results, particularly when similar policy language, clauses, product descriptions, etc are repeated and duplicated across materials. Not only are you paying for teams to replicate their work, but you are creating the risk that those elements are not implemented consistently. The time, cost and risk associated with the traditional methods of translation create a significant barrier to organizations being able and willing to translate more of their materials and therefore, better serve a diverse customer base.

Breaking down barriers with AI

Fortunately, technological advances have opened avenues for innovative translation solutions tailored to the insurance industry. These can be used to accelerate key processes of translation and translation accuracy checks dramatically, with or without human-led processes wrapped around them. AI-powered translation is succeeding by incorporating additional context from the content into providing a higher quality translation. While multiple sophisticated AI platforms are now available to translate content, such as DeepL and ChatGPT insurers need to carefully consider how to harness their power to achieve desired outcomes of faster work processes, at a lower cost, while ensuring accuracy.

The use of AI platforms for translation, while seemingly simple, has its nuances. For example, insurance terms and concepts need to be given special consideration and often organizations have very specific ways in which these are translated. Also, certain elements of customer communications and insurance documents should not be translated, yet need to be maintained, such as variable data. There are techniques and tools that can be used to guide AI properly to get the desired output or response. They include:

• Prompt engineering, which is the act of crafting requests or inputs for the AI engine can help to elicit desired outputs. This is important, but also has its limitations as when prompts become too detailed, and long, they may extend beyond the AI platform’s ability to leverage them at once
• Glossaries of standardized translations support corporate and industry-specific terminology in translations for greater accuracy and consistency
• Ensuring what variable data or entities and formatting are used during translation so nothing gets lost

Without internal expertise and solutions for leveraging these tools and techniques to get better results from AI translation solutions, insurers will likely find that the results of AI-based translation are not as effective as they hoped. Insurers need to recognize the need to have the right supporting structures and expertise surrounding this paradigm shift in order for it to be effective.

What to consider in an AI-powered platform

One-way insurers can achieve both the desired results and efficiencies is by leveraging solutions that take on the burden of steering the AI.  Insurers should look for solutions that provide prompts rather than putting that requirement onto their own teams. Also critical is the ability to incorporate their own glossaries into the translation process, and the ability of the system to recognize variables or entities that need to be preserved.

Equally critical to realizing efficiency gains is ensuring the translation process is integrated into the document and customer communications management (CCM) systems that house the content that needs to be translated. By disconnecting these two systems, insurers will be faced with the daunting task of content implementation post translation -which is not just a simple copy-and-paste job. Insurers will quickly see speed and savings gains eroded by having to manually implement content changes many times over for new languages. Ideally, insurers will choose a translation solution that also supports the workflow associated with the translation process right from the systems used to manage communications. This will ensure tracking of tasks, understanding the status and providing an audit trail for changes and reporting.

Humans and AI working together

One objection many teams wanting to innovate will hear is a concern regarding ensuring accuracy. We recommend all organizations begin their process using AI-powered translation and AI-powered translation accuracy checks in conjunction with humans-in-the-loop processes. This provides peace of mind and leaves control over the final product in your team’s hands while they get comfortable with the new paradigm and technology. Doing so enables organizations to still achieve efficiency gains, completing tasks up to 20X faster than human-led processes. AI accuracy checks have also found errors in human-translated texts that were approved and in production, and while the rate is small, the belief is that overall accuracy can be improved using these tools. AI can be leveraged to assess content by doing entity verification, missing content checks, semantic accuracy analysis, and ensuring the correct application of glossaries to validate and measure accuracy.

By harnessing the power of AI for translation processes, insurers can improve accuracy, speed, and cost efficiency, ultimately enhancing customer satisfaction and driving business growth. Recognizing the pivotal role of language in facilitating understanding and accommodating the needs of diverse populations enables insurers to empower consumers to make informed decisions about their coverage and ensure seamless interactions throughout the insurance lifecycle, fostering a path toward a more equitable and responsive insurance landscape for all.

Article originally published in Insurance Innovation Reporter.