Haptik leverages Microsoft Azure Machine translation models supporting more than 100 languages; increases local queries by 2.5X and reduces human interventions by 80%

Growing need to connect with customers in local languages

When the COVID-19 pandemic struck India, businesses moved online and started exploring digital messaging channels to connect with customers. India is a country of many regional languages and it was important to provide support in more than one language for a more natural and seamless customer experience. 

Haptik noted that a significant section of people relate to a brand better if the communication is in Hindi or a regional language. Another massive challenge was users using Romanized Hindi or Hinglish for communication. However, the translation capabilities available in the market were focused on translating long-form content like books and articles riddled with high inaccuracies and a lack of comprehension of unstructured conversations. 

Haptik supports Jio Mobility’s Intelligent Virtual Assistants (IVAs) across three platforms – Jio website, MyJio App and a WhatsApp Business account to address support queries from users. The company boasts a strong core NLU (Natural Language Understanding) engine. However, IVAs were available only in English and enabling vernacular support for users required custom language translation capabilities. 

They needed to build an extension on their NLU engine to translate from a local language to English, and then translate back to the local language for the output. Haptik partnered with Microsoft to leverage an industry-leading AI translation model for Hindi.

Microsoft and Haptik create an effective translation model for Hindi

Haptik began using Microsoft Azure Translation to enable end-to-end conversations in Hindi, Hinglish and English. Generic translation models cannot handle domain-specific terminologies, grammatical errors and unstructured sentences which are important for conversational agents to perform well. Haptik and Azure teams came together to make the translation models perform at high accuracy for short sentences and domain-specific conversations that goes way beyond generic translation models 

The engine manages Hindi in Devanagari as well as a roman script (Hinglish) allowing users to switch between languages and continue to receive relevant  responses. This makes Haptik one of the few platforms that support Hinglish as a language.    

The translation solution jointly developed by Haptik and Microsoft is part of a complex stack of AI models that let users chat with the IVA at Jio Mobility in regional languages. As a result, the organization is now reaching out to multiple customer segments that it had not been able to serve before.  

Jio Mobility has one of the largest Telecom subscriber bases in India and receives millions of messages every day on the Haptik bot. Of these, 5-10% of messages i.e. over 2 million of them are in Hindi/Hinglish and addressed by leveraging Microsoft Azure. With the high accuracy translation of the Haptik-Azure model, the number of daily messages in local languages has gone up by 2.5 times.

Earlier, when the message volume increased, the IVAs would break more frequently, i.e., there were more instances of the IVA being unable to answer a question and human intervention was required. In this case, the accuracy also improved as bot breaks decreased. This reduced human intervention by 80%. 

The scale of Jio, and the combined work of Haptik and Azure provided an unprecedented multilingual customer experience. For instance, Jio Mobility’s customer care now features an excellent Hindi conversational service, powered by the Haptik-Microsoft engine.

Enabling better conversations

Haptik currently supports 130 languages in three tiers, with its core tier managing English, Hindi and Hinglish. The company is eager to further develop its NLU stack in collaboration with Microsoft, including exploring integrated models. It is looking forward to jointly solve adjacent, but critical, challenges like better translation from other languages, transliteration in the roman alphabet and speech-to-text conversion and vice versa. Speaking on the solutions and the alliance with Microsoft Azure, Praful Krishna, CPO of Jio Haptik, said,“Haptik currently supports 130 languages. This is driven by our focus on eliminating all language barriers. It does not matter what language your customers speak; they should always have the option to interact with your brand in that language. Most other translation tools today in the market are built for specific purposes - translating a book, an article, or just a small phrase. However, Conversational AI needs translation technology that is natural and a lot more powerful. It should be able to understand the meaning behind the sentences, recognize spelling mistakes and process slang, colloquialism, and other forms of improper grammar to respond accurately. This is where other tools lacked, and where our collaboration with Microsoft– and the use of Azure Cognitive Services comes in. We are eager to further develop our Natural Language stack in collaboration with Microsoft and explore integrated models to solve similar challenges.”