How To Successfully Develop An AI Solution

A conversation with Abinash Tripathy, CSO and co-founder of Helpshift

Abinash Tripathy, co-founder and CSO of Helpshift created and ran a number of early and growth stage companies. Prior to founding Helpshift, he was the Managing Director (APAC) at Yahoo as a part of the acquisition of Zimbra, held several product and technology leadership roles at Openwave Systems, and started his career at Oracle Corporation in the CTO’s office.

As many software companies today are interested in developing their own AI solutions, we asked Tripathy to share his thoughts on the subject. After just launching turnkey AI capabilities for the Helpshift platform, Tripathy offers advice for overcoming challenges and addressing potential pitfalls for companies considering an in-house AI solution.

Q: When did Helpshift first lay out the roadmap for an AI solution and how close were you to meeting your original projections?

Tripathy: Helpshift was built for modern consumers who want to communicate with brands like they do with their friends: through messaging. In 2013, when WeChat introduced its bots platform to the world, it became clear to us that AI and bots on a conversational messaging platform would revolutionize how customers engage with brands.

Our AI and bots roadmap was laid out immediately, but the question of when we would introduce it to the market was a timing issue based on whether consumers and brands were ready for mass adoption. The introduction of bots on Facebook Messenger in 2015 caused a hype cycle and a rush to build out these capabilities, but its initial launch was a failure and caused a trough of disillusionment. I would say Facebook’s early mistakes cost the industry two years as most brands became skeptics.

Today, consumer messaging platforms such as WeChat, Telegram, and Facebook Messenger have an integrated bots platform customers are finding useful. Brands have also started to recognize the true business value of these capabilities to drive higher customer satisfaction and lower operating costs. We recently introduced our AI and bots platform, called SensAI, to help brands take advantage of these efficiencies in their support operations. We feel the market is now ready for mass adoption of these technologies.

Q: What parts of the development process took longer than expected and what processes did you put in place to ensure this project stayed on track?

Tripathy: From the get-go, we were of the opinion that the AI and bot capabilities needed to be native to our platform. Our roadmap began with a data engineering team. They were tasked with creating a pipeline and Big Data store where we could collect all of our operational data to be made available for use by our machine learning team.

We then hired data scientists who were tasked with specific problems to solve with specific use cases for our customers. We experimented with several approaches, from unsupervised learning to supervised and deep learning, and we learned a lot about what worked well for each use case. We then rolled out pilots for our lead customers and tested the capability thoroughly in production scenarios. In March of this year, we announced SensAI for release to our existing customers and prospects. It was a three-year journey with a lot of experimentation and learning, but now we are the first customer service platform with a native AI engine, and we are really driving business results for our customers.

Q: Why is SensAI Predict only available for Platinum customers, while the SensAI Starter Pack and SensAI Answer Bot can be purchased by Professional and Enterprise customers?

Tripathy: AI is not a silver bullet for everyone. In our experience, the larger customers that manage significant volume and have large contact operations really need the efficiencies offered by AI, and they see the ROI immediately, so SensAI Predict is offered to those large customers who purchase our platinum plan. Bots are a capability that companies of any size can use to drive efficiency in their operations and are offered to customers using our lower tiers.

Q: What other teams were involved in developing this AI solution and how did you ensure everyone stayed in sync throughout the process?

Tripathy: The new teams that were added for the AI capabilities were the data engineering and data science teams. However, our philosophy for success with AI is the features of AI need to be exposed in the product as a base capability. Also, for AI to improve over time the feedback from the people who use the capability in their workflows is really important; with that feedback, the machine learning algorithms can learn and get better with experience — just like humans learn.

This means every design and feature team at Helpshift working on the various capabilities of the platform had to work with the teams building the AI capability to integrate the functionality into the product in a very easy-to-use self-service model for our customers. Our AI is fully native to the platform, and a customer service team can start taking advantage of the capability with no expertise in AI. Our self-service model ensures our customers can see ROI immediately without hiring consultants or experts or using third-party AI platforms that are expensive and hard to manage.

We built out a data engineering team with expertise in Big Data products like Hadoop and Apache Spark. We also hired data scientists who came from the mathematics, statistics, and astrophysics fields with a strong programming and software background to work on the models that power our AI capabilities. As this was a planned multi-year project, it did not affect our existing roadmap.

Q: Helpshift’s CEO Linda Crawford said many AI solutions today are “overhyped and under-delivering.” How is Helpshift ensuring this new release changes customer perceptions of that assumption?

Tripathy: Linda is right: results talk and BS walks. Very large companies are investing millions in marketing platforms that don’t even solve the core use cases instead of just focusing on delivering results to their customers. This is a big disservice to the industry, as evidenced by the failed IBM Watson pilots. Our customers who have deployed SensAI Predict are seeing upwards of 90 percent accuracy in predictions. This is delivering on our promise and turning customers into real believers.