[Product Roadmap] How Haptik evolved its offerings across the conversational AI spectrum

A product roadmap clarifies the why, what, and how behind what a tech startup is building. This week, we take a closer look at Mumbai-based conversational AI startup Haptik, which builds chatbots that companies can deploy on websites, apps, and other applications.

25th Mar 2020
  • +0
Share on
close
  • +0
Share on
close
Share on
close

Haptik was founded in 2013 with the belief that conversational AI would be the cornerstone of user engagement.


The growing penetration of smartphones made this a reality, and the Mumbai-based startup is now a B2B solution for enterprises. It uses a mix of artificial intelligence, machine learning, and natural language processing to build chatbots that companies can deploy on websites, apps, and other platforms.


Haptik

Swapan Rajdev (L) and Aakrit Vaish, the Founders of Haptik.



Claiming to be one of the world's largest conversational AI startups, Haptik has reportedly reached over 100 million devices and processed over three billion conversations.


The conversational AI tech is no longer limited to chatbots, but includes all-encompassing, AI-enabled intelligent virtual assistants (IVAs) that support both text and speech.


Haptik's virtual assistants specifically cater to two key functions, sales and customer support, and are complemented by its capabilities in data, analytics, and prediction. These assistants are trained to handle several support queries, help drive sales, and enhance customer experience. They are geared to achieve higher business RoI at all stages, and augment the roles of human teams.


Speaking to YourStory, Swapan Rajdev, CTO of Haptik, said,


“To tap into the full potential of our global approach, we realised that there was a need for IVAs that can carry out conversations in vernacular languages, especially in a country like India.”


Voice provides accessibility to a number of non-English speaking users as well.


“Ours is one of the only conversational AI platforms built on real consumer data as opposed to proxy data. This has helped us gather accurate and relevant data sets needed to train our models over the course of six years,” Swapan said.


By leveraging robust Natural Language Understanding (NLU), machine learning and dialogue management, the virtual assistants are now able to better understand a customer’s needs, provide additional information, and make personalised recommendations.


The core idea is that anyone should be able to use Haptik's platform to build their own IVA within minutes.



The first-ever prototype

The first Haptik product was a B2C-specific app. It was a chat-based personal assistant for both iOS and Android, to help users better manage their schedules and tasks. The app acted as a means to showcase the benefits of conversational AI to enterprises, at a time when it was something that they were either not prepared for or didn’t fully understand the need for.


Swapan said, "It helped us highlight how conversational AI could help people get things done seamlessly, and enable companies to cut down costs and simultaneously boost revenue."


To find the perfect response, Haptik processes every message through six different algorithms, including language and intent detection, domain classification, sequence to sequence, and named entity recognition among others.


“Our team of technical experts kept adding training data to strengthen the algorithms over time,” he said.


Haptik ensured that the base model had the capabilities to understand the nuances of the language, after which domain-specific data and labels could be added with an IVA builder. The IVA builder is a graphical user interface (GUI) tool, in which the user can enter data that trains the NLU in a very non-technical and easy-to-understand manner.


He said, "The extent of customisation is our most unique value proposition; it’s carried out 100 percent by our in-house teams. What gives us an edge over our competition is the fact that we are with our clients through the whole process, from building the conversational interface to deployment and managing it across text and voice channels."

How was it built?

When the team first started building the app, Aakrit Vaish, the CEO, was responsible for building the product. Swapan was the engineer building the iOS integration, backend, and agent chat technology.


Initially, they built a beta version on iOS to get some early feedback about the idea and the product.


Later, Aakrit and Swapan built the main Haptik app and hired their first employee, who helped the duo build the Android version.


“Between the three of us, we burnt the midnight oil to get the first version of the app ready across various platforms. After the launch, the focus was on taking feedback from users, which helped make constant improvements.”


The beta version of the app and the backend work took around four months, followed by another three months to build the main production app.


At that time, the team worked towards deploying periodic updates and performance enhancement. A new version of the app was released every two weeks and the backend function would have a release planned once every two to three days.


Every three to four months, a major rehaul of the app would be done from a design perspective; this would take around a month to be rolled out. Swapan kept a close eye on the analytics data and regularly monitored reviews on Play Store and App Store.


“We would often speak to our users for feedback, collect all this data, and use it to improve user experience. This also helped us with ideas for new features,” the CTO said.

Learnings from building the product

For conversational AI, the team realised that it was good at handling “level 1 queries”, which constitute about 80 percent of the questions but require relatively simpler solutions. The other 20 percent require greater expertise, and in some cases more empathy.


Whenever an assistant could not interpret a question, the proprietary routing algorithm routed the chat to a human agent to take the conversation forward. This was done while ensuring that IVAs were constantly learning. The IVA attempted to analyse the approach taken by the human agent to decode the query and how s/he went about resolving it.


This process made the IVA smarter with every conversation, giving it the ability to independently arrive at more satisfactory solutions as compared to the previous time.


"Haptik was scaled via automation on Twitter, where any user who was facing issues and tweeting would receive an instant reply with a solution. This helped us greatly enhance user experience. We also built entertainment bots to drive engagement during special occasions,” Swapan said.


Some of the growth hacks ended up becoming really successful features for Haptik. One of the growth hacks was the task of reminding people to drink water during summers. The users loved the reminder so much that it became a full-fledged feature. The team then went on to add several other reminders.


Haptik also built a testing tool for IVAs, which is now a product offering in itself. This tool helped automate the testing of the IVAs.


“We also found a unique way to handle data. Since building IVAs is all about data, we found a way to manage data between environments so we don’t have to keep re-entering the data when we move from development to testing to production.”


The entire technology for Haptik was developed in-house, which Swapan calls its “core strength”.


(Edited by Teja Lele Desai)

Want to make your startup journey smooth? YS Education brings a comprehensive Funding Course, where you also get a chance to pitch your business plan to top investors. Click here to know more.

  • +0
Share on
close
  • +0
Share on
close
Share on
close