Machine translation has been one of the fastest developing areas of AI in the last five years and has quickly become a target investment for CX services players. As the technology evolves from statistical analysis to the use of neural networks, CX services companies start to benefit at scale. Webhelp has been active in the space for several years, initially through partnerships with machine translation tech vendors.

In 2018, Webhelp started to study the potential benefit of machine translation for CX, and in 2019 it launched Polyglot, an in-house translation capability combining AI and human translators with an exclusive focus on CX processes and service lines.

Polyglot’s machine-human platform

Under Polyglot, Webhelp builds client-specific service to translate text from one language to another in real-time and then uses humans to verify and improve the quality. Polyglot has ~110 language pairs and is powered by a combination of third-party technologies from DeepL, Google Translate, SYSTRAN and proprietary algorithm. The platform has features for monitoring and reporting, comes with built-in taxonomies, adaptation to the individual program, and has APIs and connectors to major client systems such as Salesforce and case management systems for integrations with chat, email, webforms, instant messaging, and social media.

Webhelp also brings human language experts into the loop. Machine translation still cannot fully account for the context and nuances of language, especially at the beginning of a new program. Humans come in to assist with complicated terms, industry-specific vocabulary, and, most importantly, conversation content. The AI then learns from the human conversations and corrections and improve accuracy.

The typical profile for a bilingual language expert is B2 levels in English and native proficiency another language. The requirements for the talent are accuracy, speed, and flexibility. The typical manual text review lasts 30 to 90 seconds, where the expert prioritizes grammar and vocabulary. These universal language experts are process agnostic and do not require client or program-specific knowledge. This flexibility allows for shared services verification across clients and offers borderless support, for example, from multilingual hubs such as Webhelp's Barcelona site.

Webhelp has deployed Polyglot on 26 language pairs, including major European languages, Chinese, Japanese, Arabic, and Brazilian Portuguese. For 17 languages, it employs human language experts. 

Direct benefits of machine translation

The traditional multilingual support format uses dedicated language resources serving one or several markets with restrictions on availability for long-tail and difficult to source languages. The model often requires exclusively onshore delivery, higher costs, and advanced WFM to account for different hours of operations, peak volume management, and absenteeism and attrition. With machine translation, all these challenges are alleviated, if not fully removed.

From its Polyglot implementations, Webhelp also experienced improved business resilience during lockdowns, a greater concurrency, faster speed to proficiency, and higher ESAT, as agents can operate during regular business hours and with more flexible work models such as work-at-home.

For CX clients, the lower cost and ability to serve minority languages increase the outsourcing model's attractiveness. Webhelp is actively employing Polyglot as a differentiator in RFPs and to hunt for new opportunities. It already has couple of dozen clients in the travel, fashion, consumer goods, high tech,  and media sectors. A highlight is the zero client churn for all Polyglot client engagements.

Not just traditional customer service

The Webhelp Polyglot-supported programs are not limited to just customer care. The company has implementations for technical support in high tech and consumer goods and for B2B sales. The size of the programs varies significantly.

Example implementations include:

• A multinational fashion brand for which Webhelp introduced Polyglot to simplify the multilingual operating model and generate 40% cost savings. Webhelp provides email customer care and support over WhatsApp and Facebook using automated translation in 15 languages, including French, Spanish, German, and Dutch. It reduced the delivery sites from six to one and now has ~250 agents in India and language experts in Greece and Estonia. The program handles ~2k cases per day, achieving a 35-point higher NPS than the native language teams
• A consumer electronics OEM where Webhelp engages with customers on Facebook and Amazon reviews in 14 languages using 3.5 FTEs from a location in the Netherlands and language experts in Spain and Turkey while maintaining 5 points higher NPS than the native resources
• Another consumer electronics brand needed to lower TCO on chat and WhatsApp customer service. Webhelp has a 14-member team in India covering French, Dutch, Italian, and Nordic languages, with CSAT 9% higher than the native language team, delivering 95% translation quality, improving average first response time by 76%, and a 45% decrease in annual costs.

Translation technology vs. BPS

The main differentiations between pure technology vendors and BPS players are the partnerships with different translation engines, expertise in CX use cases, and end-to-end sourcing. In the case of Webhelp, it owns the process, the CX setup, and resources. It offers Polyglot, closing the loop for model training, talent augmentation, quality improvement, security, and ongoing feedback capture. Webhelp also eliminates the complexity of separate technology licensing by rolling in Polyglot in the overall BPS contract.

For example, at the end of 2021, Webhelp integrated Power BI into Polyglot with dynamic dashboards for operational data such as tickets handled by each agent and the AHT of both agents and language experts. It also monitors rejected translations and shows the quality of translations over time.

A factor not to overlook, of course, is the high cost of the leading machine translation engines.

Machine translation technology leaders continue to improve with their neural language engines incorporating human inputs to a lesser and lesser extent. Still, this solution is only applicable for text communication, and human voice support at the quality required by CX programs is still some distance away.

The limitation to text only strengthens the positioning of CX outsourcing companies as customers migrate to multichannel interactions and vendors increasingly support voice and text with universal agents.

Opportunities ahead for Polyglot

Machine translation is becoming a core enterprise need within the larger CX digital transformation. Webhelp readily adopts these needs within its solutions ecosystem, where Polyglot sits within the BOOST family of voice transformation, RPA, and email automation.

The company is working on version 3.3 of Polyglot in the autumn of 2022. On the roadmap is adding language resources for Asian languages. It will also integrate with new case management systems to target mid-market opportunities and a new machine translation engine.

Webhelp also plans to expand into machine content translation for customer-facing FAQs, self-service articles, and VOC analysis to assist with client needs around compliance and management of minority languages. Automated CX content translation is likely among the next priority areas for clients and vendors. Brands are looking for international growth with custom sales, marketing, and support content while customer expectations for personalization quickly catch up across markets.