The rise of a technology has Bill Gates issuing warnings of an apocalypse. Elon Musk, too. Even Stephen Hawking.
What’s worrying these technologists? It’s Artificial Intelligence or AI, an idea whose time has come—it is incubating in science labs and being deployed by start-ups and industrial units alike.
Why are Gates and Co. worried? Specifically, it’s over machine learning, an early form of AI that has in recent years become mainstream, causing both delight and nervousness among AI experts and technology companies.
AI involves building computers capable of taking smart decisions by themselves, the way humans do. Machine learning and various other sub-fields such as deep learning are the means to achieve AI.
Google announced this week that it is rethinking all its products to base them on AI; it has created a new unit called Google.ai to facilitate this shift.
The revival of interest in machine learning has been driven by a confluence of factors, such as the massive increase in computing power, emergence of neural networks (connected transistors that replicate the structure of neurons in the human brain) and the easy availability of vast amounts of data, thanks to the Internet.
Compared to AI leaders in the Silicon Valley and China, India is a laggard but even here, nearly 300 start-ups are using some form of AI, according to Tracxn, a start-up tracker. Among dedicated AI-only Indian start-ups, 23 per cent are working on providing solutions to multiple industries, 15 per cent are in e-commerce, 12 per cent in healthcare, 11 per cent in education, 10 per cent in financial services, and the rest in fields such as retail and logistics, according to a 2017 report by Kalaari Capital, a venture capital firm.
Internet companies tap machine learning techniques for a range of uses—to recommend products for you, for instance, or to predict where cabs should be placed so that when you open your cab-hailing app, there’s one a couple of minutes’ drive away.
Healthcare start-ups use AI to help hospitals make speedy and accurate blood reports and medical diagnoses, saving lives. Others get fashion brands and retailers to buy the right quantities of stock.
Suddenly, AI is everywhere.
Lounge looks at 10 exciting start-ups that are using AI and machine learning techniques to transform healthcare, education, auto and retail. Their success is far from assured. They struggle to find quality talent; they face stiff competition from other AI start-ups; their solutions haven’t yet stood the test of time (most are less than five years old); and they are vulnerable to a variety of other factors that typically bring down start-ups, AI or not.
Given the number of AI start-ups, the list is far from comprehensive. Yet the work that these 10 start-ups (and others like them) do is exciting and sometimes cutting-edge, holding out the promise of a significantly improved product or service.
How does this work? Traditional code writing entailed programming computers, essentially telling computers what to do. With machine learning, computers are taught to do things themselves. Pedro Domingos, a professor in computer science and engineering at the University of Washington, puts it fittingly in his book, The Master Algorithm: How The Quest For The Ultimate Learning Machine Will Remake Our World: “The Industrial Revolution automated manual work and the Information Revolution did the same for mental work, but machine learning automates automation itself.”
This is what has people worried—and thrilled. The likes of Gates and Musk fear that machines that learn to think for themselves may become smarter than humans and start taking control of things—the stuff of science fiction. But others such as Domingos believe it will lead to a near-utopian state where people, freed of chores and menial work (now automated), will pursue things that truly excite them.
No matter which side of the fence you are on, there’s little doubt that AI represents potentially the most disruptive leap in technology in decades.
Apart from disrupting incumbents across businesses, AI may worsen a major socio-economic headache for India: employment. The threat from automation to engineering jobs in Indian IT companies is well documented. But IT is by no means the only business where jobs are at risk. Automation may be a threat to 69% of the jobs in India, according to an October 2016 World Bank report. Even Domingos, who expresses a never-failing belief in the potential of AI to do good for humankind, acknowledges that the technology is likely to make many jobs redundant.
Case in point: One of the 10 start-ups we feature replaced its 40-strong content team with a few computer scientists when it adopted machine learning techniques.
That represents only a mild threat from AI, especially in a country that is already failing to meet the growing demand for jobs. Investors and experts in AI privately worry about this threat—but they aren’t worried enough yet to look for solutions.
That must say something about where their priorities lie—and about the disruption that Artificial Intelligence will cause.
Netradyne, Bengaluru and San Diego, US
Netradyne co-founder Avneesh Agrawal. Photo: Hemant Mishra/Mint.
Investment: Raised $16 million (around Rs102.8 crore) from Reliance Industries Ltd in June 2016.
The team: Founded by Avneesh Agrawal, a former Qualcomm Inc. president for India and South Asia, and David Julian, a former principal engineer at the US-based chip-maker.
Intelligence factor: Netradyne’s Driveri, a powerful camera that analyses driving patterns and can help determine the cause of an accident. The soap-bar-sized device is attached to a vehicle’s rear-view mirror and rests on the inside of the windscreen, pointing towards the road.
In July 2015, Agrawal, along with his long-time friend and colleague Julian, set up Netradyne, wanting to tackle that intractable problem of the modern world—road accidents.
Netradyne’s main product is Driveri, which packs in four high-definition cameras generating 360-degree footage of the vehicle’s path (transmitted on a real-time LTE network), has a global positioning system, gyroscope sensors and accelerometer, and a Nvidia processor, the same one that is used in the iPhone 5. The unit captures visuals of the car’s surroundings, analyses driving patterns and stores the data on a cloud platform.
It uses machine learning and deep-learning systems to analyse the entire scene in front of the car: traffic lights, stop signs, objects in its course, distance to every other vehicle, relative speeds and direction.
The data generated enables the platform to determine whether the driver is overspeeding or driving rashly, adhering to traffic rules, is potentially drowsy or drunk, or taking multiple halts along the route. In the event of an accident, it also sends real-time alerts to the fleet operator.
According to Agrawal, similar technology is used to power autonomous car projects across the world, such as those being developed by Google (Alphabet Inc.), Uber Inc., Robert Bosch GmbH and large car-makers like Toyota, Volkswagen Group and Daimler AG.
After testing the product for over 750,000 miles (around 1.2 million kilometres), Netradyne launched Driveri in the US market in March and plans to launch it in India by June-July. It recently signed its first US client, Load One, a mid-sized cargo company that operates a fleet of 350 trucks. The solution is also useful for retail customers like parents of first-time teenage drivers.
Agrawal says Netradyne’s main focus currently is to ensure driver safety. The product is being marketed to commercial fleet operators to enable them to monitor drivers and enhance the safety of cargo.
“Using the solution, fleet managers can train drivers on safe driving practices, besides maintaining a close check on valuable cargo…. They can also create a scorecard for drivers and incentivize them on safe driving,” says Agrawal.
Agrawal, whose company employs around 60 people, expects Driveri customers to see a significant reduction in the number of traffic violations and accidents; in turn, safer driving would result in greater fuel efficiency and lower maintenance costs.
The other big market is insurance.
“What we are essentially doing is quantifying risk,” Agrawal says. His plan is to bring Driveri to every car in the world with insurance partners who agree to use its data for investigating accidents and deciding on a more accurate premium amount.
Embibe founder Aditi Avasthi
Investment: $9 million from venture capital investors such as Kalaari Capital and Lightbox.
The team: Founded by Aditi Avasthi, a former Tata Consultancy Services (TCS) and Barclays executive.
Intelligence factor: In a country where the overall quality of education consistently lags behind those in developed nations, Embibe’s core AI product can be a game changer.
In the book (and movie) Moneyball, a baseball coach and an economics graduate from Harvard use an analytics engine to ditch decades of baseball practice and identify a world-class team that ends up breaking numerous Major League Baseball records. Major teams end up adopting the model.
Embibe is hoping to create a similar impact in the world of education.
Its learning platform is being used by thousands of students and the start-up is in talks to ally with educational training centres. Embibe, which runs a website and a mobile app, collects data from students, charging only for advanced analysis and personalized learning recommendations. Students can actually improve test scores by fixing basic mistakes using its AI platform.
For instance, Embibe’s First Look Accuracy metric trains students to maximize scores in a given test by first answering the questions they can attempt confidently and then moving on to the tougher ones.
Another metric, “‘Time Spent On Questions Not Answered’, indicates no confidence, an inability to know what you know yourself. You’re unsure of your ability. This value is actually 41 minutes for Indian kids for the IIT exams out of 180 minutes—in an exam where 1 mark is almost equivalent to 10,000 ranks. So, think about it—if you’re able to bring this value down by 50%, the scores automatically trend up,” explains Avasthi.
The company says it has identified 27 parameters that can, for instance, predict scores in an IIT entrance exam with 93% accuracy.
To get the bulk of data that its core cloud-based platform crunches, the start-up bought 100Marks, another ed-tech start-up, in 2015.
Embibe uses advanced-level analytics besides its core AI engine, including self-learning algorithms, machine learning and cognitive computing akin to IBM’s Watson. It started out as an analytics company in 2012 and began using AI and machine learning tools in 2014 in an effort to create personalized solutions for individual students.
Embibe started with a 40-member team that was primarily responsible for content creation; now it’s down to just three. The cost of creating content is down “from the high hundreds to Rs20 per unit of content through automation,” says Avasthi, who has hired three machine learning and AI experts from top companies, including IBM.
Tricog Health Services Pvt Ltd, Bengaluru
Tricog Health Services co-founder Charit Bhograj
Investment: $2 million from Inventus Capital Partners, Blume Ventures and angel investors.
The team: Charit Bhograj, a cardiologist; Zainul Charbiwala, who worked at IBM India and Qualcomm; Udayan Dasgupta, who worked at Texas Instruments; and Abhinav Gujjar, who worked at Thomson Reuters and Microsoft.
Intelligence factor: What can take up to 6 hours before treatment starts, Tricog accomplishes in a few minutes.
Nearly 7.5 million people die every year of heart disease. Around 1.5-3 million of them are in India; half of them can be saved by early diagnosis, according to Bhograj.
“Yet almost 70% of the general physicians do not have an ECG (electrocardiography) machine,” he adds.
ECG—a heart health test—is conducted through a machine which records the heart’s electrical movement. These machines are not widely available in India, nor are there enough cardiologists to interpret ECG data. And it can take up to 6 hours before a patient is diagnosed and sent for treatment, says Bhograj.
So Tricog set out to help doctors make instant diagnoses of heart attacks and ensure treatment is not delayed.
While Tricog sources ECG machines from General Electric Healthcare, it has built its own sensory device and fits it on the machines. It gives these devices to general physicians, clinics and nursing homes on a subscription basis. Through the Internet, this device sends the ECG or recorded heart movement to a set of algorithms, which then generates a report. Before the report is sent, a specialist doctor verifies it.
In 2016, Tricog processed the ECG reports of 200,000 patients; about 11,000 of them were diagnosed with a heart attack. The company claims it has never reported false results.
Yet, given the stakes, Bhograj reasons that “no matter how accurate the algorithm gets, there will always be a specialist (cardiologist) to verify the report churned out by it.”
Founded in 2011, the company spent close to four years putting the technology together and launched the product in February 2015. As of April, Tricog had 600 clients (general physicians, clinics and nursing homes) across 178 cities and towns, and 20 specialist doctors, including clinical cardiologists, who study the reports. In the next five years, Tricog aims to be present in 100,000 locations—it claims that though such a scale would ordinarily require around 700 doctors, it would be able to deliver comparable service with only 50 doctors.
Arya.ai’s co-founder Vinay Kumar Sankarapu. Photo: Aniruddha Chowdhury/Mint.
Investment: An undisclosed amount from YourNest Angel Fund and VentureNursery.
The team: Vinay Kumar Sankarapu, chief executive officer, and Deekshith Marla, chief technology officer.
Intelligence factor: Many start-ups are working with AI to solve problems in banking and insurance. Arya.ai tells them how to.
Last year, world champion Lee Sedol was pitted against AlphaGo, a computer program made by Google, in a game of Go, an abstract board game. All AlphaGo needed to be told was the desired result. It figured out the rest for itself. Sankarapu, seated in his office in Mumbai’s Andheri area, says the day isn’t far when a business firm will be able to apply AI for similar results. His company, Arya.ai, is taking steps towards this; it is helping build AI that “becomes more intelligent the more it is used. The next generation of AI will be built by AI,” he says.
Many start-ups in India are working with deep learning, which aims to make human effort minimal. Arya.ai works as an “enabler”. For instance, if a consultancy firm is building AI for its investment banking client, Arya.ai will provide the firm with the tools to build it, creating the “neural network”—a vast computing system that mimics the human brain; it will create a cloud system which will allow the AI to evolve, learn from its earlier tasks and apply them to the next one.
“Basically, we help build a complicated system really fast, help you automate a lot of stuff. An algorithm may take months to build. We could actually help you build a complete system in weeks,” he says.
Arya.ai also has direct clients, such as banks or insurance companies, where cases of fraud claims and bad loans can be identified with precision. The change is already visible: the phone call you get from your bank when an irregularity in your transaction pattern is registered is courtesy its AI-enabled system. This is the result of the data fed into it, based on geographical location and your records of the past 30 days. But as hackers get sharper—making low-key transactions over a long period of time after carefully observing the customer’s banking habits—deep learning will take it to the next level, with greater accuracy.
Currently, Arya.ai is engaged with one of the world’s biggest consultancy firms, New York and London-based banks, global insurance companies and leading manufacturing firms. Their work is mostly at the business-to-business level but the effects, says Sankarapu, will also be felt by the man on the street. “Things will become super-fast, personalized and cost-efficient for the end consumer,” says Sankarapu, who grew up in Andhra Pradesh’s Kadapa district and now shuttles between Mumbai and San Francisco.
In 2015, Arya.ai was chosen as one of the 21 most innovative start-ups by Paris&Co, a French innovation agency.
Arya.ai started out as a way of making the lives of researchers easier in 2013, at a time when Sankarapu was studying at the Indian Institute of Technology, Bombay. If there were 200,000 books on a particular subject, it was not humanly possible to read them all; Sankarapu was trying to build AI that would absorb as much information as these books would hold in less time than the human brain ever could. “I soon understood that the bigger problem is not building a solution but enabling it. Everyone wants to use deep learning, but they don’t know how to use it,” he says.
Sankarapu is a Marvel fan but he’s never had any romantic notions about AI. He says that like any new, exciting technology that comes every decade or so, AI has been hyped beyond its capabilities. To him, it is a formula, one which is more exciting than fantasy. “When you start understanding how to build it, you start understanding what we are made of. The human brain is nothing but a simple mathematical computer.”
Locus.sh (Mara Labs Inc.), Bengaluru and US
Locus.sh’s founders Geet Garg (left) and Nishith Rastogi. Photo: Hemant Mishra/Mint
Investment: At least $2.75 million from Exfinity Venture Partners, Blume Ventures, BeeNext and others.
The team: Nishith Rastogi, who has worked with Amazon and eBay; and Geet Garg, who was with Amazon. The two had co-founded PinChat, a location-based conversational platform, in August 2014.
Intelligence factor: Locus has developed route-planning algorithms so companies can chart the best possible route to deliver an order and allow a salesperson to cover the maximum number of points in the shortest time possible.
Locus aims to automate all the human decisions involved in sending a package. With clients such as Hindustan Unilever, Quikr, Urban Ladder and Lenskart, it has a lot on its hands.
The company, which says it has more than 25 clients, has developed a route-planning engine, its core business, apart from a 3D packing engine that provides configurations for loading cargo into containers. Locus also offers companies a weekly schedule of the most efficient routes and outlets for their sales teams.
Rastogi says each schedule planned by their routing engine takes 5-10 minutes, whereas a skilled human being would take 1-4 hours to process the same data. The company says it helps its clients reduce their logistics costs by at least 25%.
Locus offers solutions for both intracity and intercity operations. It estimates the domestic freight industry is around $100 billion; a bulk of this is intercity logistics.
Planning the best possible route is not as easy as it sounds. “In many cases, we will have a landmark and street name with a wrong pin code. We have to convert this information into latitude and longitude. For this, our systems need to understand and interpret the English language and build natural language processing systems,” says Rastogi.
“Let us say you have to deliver a package before hitting the gym at 6pm. But if you reach the gym 5 minutes late, that is also fine. How do I make the system understand that it is better to reach the gym by 6.05pm instead of not delivering the package at all? In the real world, if a truck is full, the driver can keep one package next to his seat. It is important in the real world to understand what that soft threshold is. It is important for the system to understand this world fuzziness,” explains Rastogi.
Companies provide Locus the origin of the package, destination and expected time of delivery. Locus sources all the possible routes from Google Maps or sifts through data on past deliveries. This data is then processed, factoring in weather information, traffic, historical data on time taken to cover the distance, etc. to arrive at a few best routes. If the client is new, logistics solutions take about three months.
Are the companies willing to wait? “We haven’t lost a single client yet,” says Rastogi.
Mad Street Den, Chennai
Mad Street Den’s co-founder Ashwini Asokan. Photo: Hemant Mishra/Mint
Investment: $1.5 million from Exfinity Ventures and growX Ventures, in addition to an undisclosed sum from Sequoia Capital.
The team: The husband-wife duo of Ashwini Asokan and Anand Chandrasekaran. Asokan, who previously worked for Intel Corp., and her husband Anand Chandrasekaran, a neuroscientist, who is the firm’s chief technology officer, both worked in the US before deciding to return to India in 2013.
Intelligence factor: One of the several AI start-ups trying to improve the retail experience in India.
Asokan, among a handful of people in India who speak openly about the skewed gender representation in the start-up business, runs Mad Street Den, a start-up that is trying to make the business of retail more efficient.
The start-up’s main product, Vue.ai, offers visual search technology, product recommendations, and personalized home pages based on the tastes of individual shoppers, among other features.
The AI-based technology offers several benefits to customers: uploading pictures and finding matching products, browsing in a way that is specifically tailored to their tastes, discovering products they may otherwise not have been aware of, and reducing the amount of time they spend shopping or having a more efficient shopping experience. As the customer experience improves, it results in better conversion rates for retailers. Vue.ai also helps retailers reduce cataloguing errors and product returns and make merchandising more customer-friendly. The company also hopes to turn omni-channel retail, so far a pipe dream, into reality.
Mad Street Den currently works with firms such as furniture e-commerce site HipVan in Singapore and fashion e-tailers such as TataCliq, Craftsvilla and Voonik in India.
The Chennai-based company is one of the tens of start-ups selling efficiency-improvement tech to retailers. Others, such as Stylumia, Staqu and SnapShopr, offer similar services.
Mad Street Den, however, is expanding into gaming and other businesses this year.
“Scale markets globally, scale our AI systems and establish ourselves as the first computer vision business that’s making money on scale,” says Asokan.
She sounds the part of a change-the-world entrepreneur.
“To build generalizable AI is one of the founding principles of why we started this. The second part of the founding principle is making AI accessible to millions of people across the globe. Not just to build it, but actually to build it on the scale that applies to people across the world. The first one represents Anand’s vision, the second one represents why I am here,” says Asokan.
Niki.ai (Techbins Solutions Pvt. Ltd), Bengaluru
Niki.ai’s founders (from left) Sachin Jaiswal, Nitin Babel, Keshav Prawasi and Shishir Modi. Photo: Hemant Mishra/Mint
Investment: Niki.ai raised undisclosed seed investments from Ronnie Screwvala’s Unilazer Ventures, Tata Sons chairman emeritus Ratan Tata and Haresh Chawla of True North between October 2015 and December 2016. Niki.ai declined to disclose the quantum of funds raised.
The team: Founded by four Indian Institute of Technology, Kharagpur, alumni. Before starting the venture, Sachin Jaiswal, chief executive officer at Niki.ai, had co-founded InnovAccer Inc., a healthcare analytics company in the US. Technology head Keshav Prawasi was a software engineer at Amazon; Nitin Babel, who looks after support and marketing, worked for Ipsos, a global market and opinion research firm; and business head Shishir Modi came from a stint at a multinational energy distribution company.
Intelligence factor: Niki is used for over 20 services, including hotel, cab and movie ticket bookings, ordering food and paying bills. If you’re rude, it will reprimand you and can even block you.
Chatbots—like Apple’s Siri and Google’s Allo—are essentially computer programs that make human-like conversation with users.
Banks, retailers and others use chatbots to answer basic customer queries and help users navigate their websites.
Niki, one of the early chatbots developed in India, is designed to serve as a virtual shopkeeper that assists in the purchase of products and services. It can help you book a cab, order food from the nearest restaurant, book tickets for movies and events, and pay general utility bills. Flight bookings, payments for insurance and courier services will be introduced soon.
Niki was conceived in early 2015 and rolled out to customers in October that year.
It works on a series of sophisticated algorithms that understand human language by breaking it down into structured queries that the machine can understand. Another set of codes then generates a response relevant to the query, and this goes on till the task is accomplished.
The core of the platform rests on language-recognition systems. “We started with off-the-shelf solutions like Stanford CoreNLP and went on to develop our own custom models,” says Babel.
These were developed by studying the interactions on Niki and constantly tweaking algorithms to accurately determine what the user is looking for.
Currently, Niki has over 300,000 users; 50-60% of them are in the age group of 24-35, 20-30% in the 18-24 age group, and about 10% are over 35.
“We have been collecting the conversational commerce data for Indian customers for the last two years. Close to 50 million interactions have happened on Niki and have been used to improve the accuracy from 50-60% to 90%,” says Babel.
He says the company earns revenue by charging a commission on each service or product purchased on Niki.
Earlier this year, the company launched a software development kit that allows other sites and apps to integrate Niki with their platforms. It is available on HDFC Bank’s OnChat, the Oxigen Wallet app and the operating system of Intex smartphones (as the LFTY feature). The company will launch 20 more partnerships soon.
Zenatix Solutions Pvt. Ltd, Gurugram
Zenatix’s co-founders (seated in front, from left) Amarjeet Singh and Vishal Bansal. Photo: Pradeep Gaur/Mint
Investment: Rs11.5 crore from pi Ventures, Blume Ventures, Rahul Khanna, managing partner, Trifecta Capital, Rajan Anandan, vice-president and managing director of Google, South-East Asia and India, and Snapdeal co-founders Kunal Bahl and Rohit Bansal.
The team: Vishal Bansal, who worked as portfolio manager at ING Insurance; Rahul Bhalla, who worked in the legal outsourcing industry; and Amarjeet Singh, who worked as an assistant professor at the Indraprastha Institute of Information Technology, Delhi.
Intelligence factor: Zenatix helps businesses save energy and cut down their electricity bills.
Zenatix, a team of 35, works with large chains of quick-service restaurants (QSRs), bank ATMs and retail stores which have anywhere from 200 to more than 1,000 stores/outlets.
A Zenatix client that has around 1,000 stores, each 2,500 sq. ft in area, used to spend Rs80-100 crore annually on electricity. Zenatix helps it save 10-30% of this amount.
All these outlets have electrical assets such as ATM machines, air conditioners, and ovens and exhaust fans in the case of QSR kitchens.
When it comes to ACs in a workplace, someone would generally switch them on at the start of the day and switch them off at the close of day. “The person is not equipped with the intelligence that, based on the occupancy of the space, temperature outside and even time of day, the usage of AC can be optimized. Or consider, if an ATM breaks down, it could take a few hours to days (depending on where it is located) to get it fixed,” says Bansal.
“If we know that a grocery store or an ATM sees a minimum footfall during certain hours of a day, one set of ACs may be switched off to maintain the same ambient temperature inside,” he adds.
Through its WattMan device, Zenatix captures data points such as the time of day, electricity consumed, and outside temperature, every 30 seconds. Based on this, it can reboot an ATM when it stops working, predict when an AC in a retail store will break down or needs servicing, and automate its switching on and off.
Zenatix plans to double WattMan’s presence to 1,000 sites by June, as its existing clientele of over 20 companies hands over more stores/outlets to Zenatix. As of April, it had installed WattMan at 500 sites.
Zenatix wants to build accurate computing and predictive models. “Imagine if a door to cold storage is left open, and as soon as it reflects a change of temperature, the door closes automatically. That’s the kind of advanced temperature profiling we want to build… We want that when an air conditioner is not working properly, the correlation between temperature and electricity can tell us whether it is the gas, filter or compressor causing the problem,” says Bansal.
Active.ai (Active Intelligence Pte Ltd), Bengaluru and Singapore
Active.ai’s co-founder Ravi Shankar. Photo: Hemant Mishra/Mint
Investment: $3.5 million (around Rs22.5 crore) from Kalaari Capital and IDG Ventures India.
The team: Ravi Shankar, chief executive officer and co-founder, has worked with banks such as HDFC Bank, ABN AMRO and YES Bank; Shankar Narayanan, the chief operating officer, founded the start-ups Fastacash and Cazh and co-founded Tagit; Parikshit Paspulati, chief technology officer, founded Finoculus and co-founded Tagit. Narayanan and Paspulati are Singapore-based, where the company is headquartered; more than half the workforce is based in Bengaluru.
Intelligence factor: Let’s say, you need to know your account balance and don’t really have the time to walk into an ATM or even look into the bank’s app. Send a message to the bank through Facebook Messenger or WhatsApp and you will get the reply. Chances are high that your bank is using Active.ai platform, an intelligent interface that allows banks and consumers to connect over chat.
Shankar, Narayanan and Paspulati have all worked with banks and fintech start-ups. Perhaps naturally then, Active.ai wants to improve customer engagement for banks and insurance companies.
“Say you type, I lost my wallet. What will the bank understand? There is no transaction for ‘lost my wallet’. The system will understand that it means all the cards are gone. The bank will say I will block your card,” says Shankar. “Now, let’s say you are at an airport. The bank will say, ‘Looks like you are at an airport and travelling. Can we help you with a (travel) insurance policy?’ The ability to hold this conversation is what we bring to banks.”
According to Shankar, Active.ai will help banks almost halve their customer engagement costs.
Active.ai essentially offers a platform that has chatbots through which banks can interact with customers.
The more customers interact, the more the Artificial Intelligence behind Active.ai’s platform will learn about them and the better it will be able to offer personalized advice and service, increasing customer engagement and loyalty.
Usage for consumers can range from opening an account and getting balance information, to transferring money to other accounts or paying bills. Banks, on the other hand, can push relevant products and services, drawing on the data available on customer history.
The bot can be integrated via various channels, including WeChat, LINE, Kakao Talk, Facebook Messenger, Amazon Echo and Siri. Largely, the company sources data from banks and other partner financial institutions, apart from Envestnet | Yodlee, a data aggregation and analytics platform.
Active.ai has developed algorithms that can convert unstructured messages, i.e. input by consumers, into structured instructions for banks using natural language processing and machine learning.
“Say you are typing on Facebook. This is unstructured messaging. We convert that into entities and intent. For instance, you type, ‘I want to know what is happening in my account.’ Here, entity is the account and intent is ‘I want to know’. Our algorithm captures this and it then figures out the context. That is the third component. So the system will think this guy has an account and wants to know what is happening there, I should get his balance,” explains Shankar.
In India, the company counts Axis Bank as a client; deals with several others are in the pipeline. It says it has already signed up with banks in Malaysia, Singapore and Thailand. Discussions with banks in North America are under way, says Shankar.
“Apart from reducing the costs of customer engagement, banks can cross-sell or upsell. It helps them retain customers with higher engagement and client service levels. It also helps banks reach out to new consumers, especially the youth, via chat,” he adds.
Having started with the banking sector, Active.ai now plans to work with insurance companies to boost customer engagement.
SigTuple’s founding team (from left) Tathagato Rai Dastidar, Rohit Kumar Pandey and Apurv Anand. Photo: Hemant Mishra/Mint
Investment: $6.4 million, from investors such as Accel Partners, Axilor Ventures, IDG Ventures, pi Ventures, Endiya Partners, VH Capital and Flipkart founders Sachin and Binny Bansal.
The team: Started by former American Express executives Rohit Kumar Pandey, Tathagato Rai Dastidar and Apurv Anand.
Intelligence factor: SigTuple is helping hospitals and healthcare centres improve the speed and accuracy of blood reports.
In medicine and healthcare, accuracy and time are important factors. A delay of even half an hour can result in loss of life; and a wrong diagnosis could prove tragic. SigTuple is looking to address both these issues through its AI-powered Manthana platform, which can, among other things, analyse blood samples and generate medical reports in less than 10 minutes—and with complete accuracy, according to Pandey, who is also the chief executive officer.
“From the moment someone reaches the labs and provides a blood sample, the total time taken for the generation of a report and review by a pathologist is less than 10 minutes,” says Pandey.
Manthana is being used by a number of mid- to large-sized hospitals, diagnostics labs, and eyecare and healthcare chains. Among their prominent clients is Anand Diagnostic Labs, one of Bengaluru’s biggest chains, which uses one of SigTuple’s digitization tools to analyse blood samples.
Now the team is working towards making Manthana available for a complete diagnosis.
At present, Manthana can analyse blood, semen and urine samples—Pandey says the platform is capable of generating complete blood reports, which typically analyse 21 different parameters, including red and white blood cell count and platelets. For example, SigTuple has used the AI-powered platform to analyse a blood sample to catch a malarial parasite, says Pandey.
Accuracy is another area that the team hopes to improve on—currently, SigTuple’s accuracy rates differ for different solutions. For instance, accuracy rates for urine samples stand at around 85-88%.
For SigTuple, evolving from traditional analytics to machine learning was a crucial pivot, since it wanted to create tools that could analyse medical cases, just like a doctor would.
“Earlier, the algorithms that were being used, as soon as they saw new data, the results used to get skewed. With the advancement of AI, as more and more new data keeps coming in, the machine continues to learn—like humans. And it starts generating solutions, which was not possible with traditional algorithms,” says Pandey.
The limitations of traditional algorithms were addressed by machine learning, where a machine is trained to mimic a human brain—somewhat like IBM’s Watson.
For example, a hospital chain which wants to open a lab in a tier-II or tier-III city usually has to spend hundreds of crores of rupees on hardware, equipment and medical experts.
With SigTuple’s platforms such as Manthana, hospitals can scale up in smaller cities and towns. Since it also eliminates the need for labs to ship blood samples to different locations, it helps diagnostic chains save on logistical costs and, more crucially, time.
“There is no need for a manual review by a pathologist by putting a slide under a microscope. Second, remote diagnosis is made possible, because the pathologist need not be sitting next to the microscope or blood slide. No shipping is required—the pathologist can be anywhere,” says Pandey.
SigTuple gets most of its data from the labs and hospitals that it works with. Over the next year, a top priority for the company is to gather more data through more such partnerships to improve the accuracy of Manthana.
(Yuvraj Malik, Anirban Sen, Arushi Chopra, Sankhayan Ghosh, Sayan Chakraborty and Sadhana Chathurvedula contributed to this story.)
Also Read: T