There was once a time when an employee had to fill an application form or requisition form for everything from applying for leave to requisitioning for stationery. Well, it still continues to this day, but things are changing. Outmoded forms and the practice of sending email are being replaced by chatbots in the enterprise. AI bots are playing an important role in the transformation of business processes. With cutting-edge technology like artificial intelligence and Natural Language Processing (NLP), services-driven organisations are building conversational interfaces into their applications.
Take the HR function for instance. Employees applying for leave need to check their leave cards for balance leaves and then fill an application form — or send a written request by e-mail to the HR department. There is a period of waiting before the employee is informed about the status of the application. But this process is being transformed with chatbot interfaces. There won’t be a person from HR answering questions, but a bot. In the conversational enterprise an employee would just have a free-flowing conversation with a bot and type questions like: “How many sick leaves do I have left?” or “What is the maternity policy of the company?” The bot would instantly respond by showing the number of leaves left, or by displaying a link to a PDF file that outlines the leave policies of the company. What’s more, it would helpfully show links to similar questions or open the PDF document and highlight the sections pertaining to the employee’s query. And if the bot is unable to answer, it would ask the user permission to escalate the question to an authority.
Here’s another use case from the healthcare industry. In a typical hospital, a doctor meets 100 to 150 patients a day. Obviously, it is not possible for the doctor to remember all cases, so it is all recorded in files. Hospitals have EMR (Electronic Medical Record) and EHR (Electronic Health Record), but these are very cumbersome. A typical screen has 150 fields.
A doctor could use conversational AI and ask what medicine she prescribed to a particular patient with a specific ailment. And the bot would respond in an instant.
In another scenario, a lot of feedback comes in from hundreds of thousands of patients. Doctors and administrative staff are unable to read all of that and understand it. Who will categorise it and pull out insights? This can be done by an NLP engine, which extracts insights. An insights bot could try to figure out what the problem is (based on context), rather than on numbers.
Some of these use cases are seen at the Manipal Hospital.
The Enabling Platform
Here are a few examples of companies building this technology and how it is being used in different industries.
Light Information Systems provides a platform called NLPBOTS that helps enterprises create customisable and scalable NLP AI solutions. It enables contextual NLP conversations. The Pune-based startup has been in existence for five and a half years and has received $2.2 million in funding from NB Ventures. Its solutions are live in more than 27 companies ranging from automobiles, pharmaceuticals, and modern-day unicorns. Light Information Systems is doing various use cases for companies, such as employee engagement, customer engagement, automating hiring processes, marketing and sales processes. Some of its clients are Accenture, Oyo Rooms, Lenovo, Piramal, Mahindra, Tata Communications, Edelweiss, PNB Metlife, and Mondelez.
“We are doing this for some large automobile and pharma companies,” said Animesh Samuel, Co-Founder and Chief Evangelist, Light Information Systems. “It began as an HR Assist to help employees with their HR problems. Today that has also transformed into Employee Assist: tech support, meeting room, scheduling. All those things can be built into our bot.”
Light Information Systems has also built conversational bots for other business functions such as service support (Customer Assist), and sales & marketing (Marketing Assist). Tata Communications is using its marketing solution.
“We did a solution for Tata Communications called Marketing Insights. It started out as a CXO level solution, If a CXO was going to be on a panel with somebody or was going to meet with somebody, they needed to get a lot of information about who they were meeting, and the strategic focus of the company. They also needed to learn about the strengths, weaknesses, etc. Knowing something unique about the panellist would also help. To create all that, they usually have a team of people to get this information from various sources on the Internet, who put the discussion points together. But that would be a time-consuming process. So we stepped in and automated that process. Based on the name of the company or person you input, our bot will decide which resource to go to dynamically, and fetch all this information,” informed Samuel.
Bangalore based Senseforth.AI is another startup making conversational interfaces for more than 20 clients that include HDFC Bank, HDFC Life, ICICI Bank, ICICI Lombard, Airtel, Nestle, Manipal Hospitals, SBI Card, Godrej, Club Mahindra, and many others. Last year this startup raised $2mn series seed funding in the US. Its conversational chatbot mimics human cognitive abilities in reading, listening, comprehending, interpreting and conversing. This ability combined with several “actionization” modules creates intelligent bots for a business so that it can respond to customers in an automated, real-time, and contextual way.
“For a long time, the only way we could interact with computers was through graphical user interfaces,” said Shridhar Marri, CEO, Senseforth.Ai. “Interfaces have not evolved much over the years. Human cognitive abilities were never in the realm of computing. For a long time, computers could never understand what we said or wrote. But that’s changing. We can now make the computer understand our instructions.”
Senseforth created a complex conversational AI for HDFC Bank. The bot can fulfil tasks like transferring money.
“In a suite of 300 – 350 transactions, at least half of them can be fulfilled without a human, said Marri.
Senseforth bots comprehend natural language input that could be in various formats: text, SMS, a document, or an email. It has an NLP engine which can read human text. It enables employees to interact with an organisation without the need to talk to a human.
“It’s a natural language conversation. Once it understands the intent of the user it needs to take an action based on that. An action could be a response, a workflow, or an integration into a backend system. It could kick off multiple workflows, open up different systems, fetch data, rewrite into it — a whole suite of actions; we have an ‘actionisation’ layer that takes care of that,” said Marri.
Bots are also helping organisations to interact with customers. These days it is quite common to see bots on banking websites answering queries about loan applications. These bots can even check one’s credit history and determine if the applicant is likely to default on the loan repayment. Bots can also collect user documentation for eKYC.
The startups are not the only ones pursuing conversational interfaces and bots. Big tech companies like IBM have for long been doing research in this domain. IBM claims its Watson Assistant can build conversational interfaces into any application, device, or channel. The solution helps call centre agents to quickly source information by typing queries in natural language.
“Most chatbots try to mimic human interactions, which can frustrate users when a misunderstanding arises,” said Vikas Arora, IBM Cloud and Cognitive Software Leader, IBM India/South Asia. “Watson Assistant is more. It knows when to search for an answer from a knowledge base, when to ask for clarity, and when to direct you to a human. Watson Assistant can run on any cloud – allowing businesses to bring AI to their data and apps wherever they are.”
Watson Assistant comes pre-trained with industry-relevant content. It can make sense of your historical chat or call logs, and it has a visual dialog editor. This is enabled by IBM Watson Discovery, a natural language processing based content search feature of Watson, that can extract semantics, sentiments, entities, and concepts.
Challenges
It took a long time for the technology to evolve and address challenges. Five years ago the conversational chatbots were not even heard of and developers were then working to perfect searches based on keywords — on search engines like Google.
“In those days searches were a keyword or n-gram based classification. It also had a lot to do with machine learning. If you typed in a few words, it would depend on how humans reacted to that — for its ranking. Most of the learnings were based on the relevance factor,” said Samuel. “While that may work for a search engine like Google, it is not acceptable in the enterprise world. We thought we had an opportunity if we could teach machines to understand natural language; there could be a lot of use cases.”
Five years later his efforts have paid off and their technology has evolved to a platform for solutionizing any use case when it comes to dealing with unstructured text.
“NLPBOT can read tons of documents, and that’s what enterprise systems lack today; they lack cognitive ability. We are solving that problem using NLP and AI. NLP is a set of algorithms or rules for converting text into a mathematical equation. On that, we have the AI models that keep evolving and getting more intelligent as it learns. It is evolving at the client’s end with their data,” said Samuel.
But the challenge now is giving the bot some context, specific to the domain. For instance, in the insurance sector, the term ‘dwelling’ means a place where people live. But in the hotel industry, it would have a different meaning. So the bot would need to self-learn the lexicon and concepts within a particular industry over a period of time. But how long would that take?
“There is a learning period for the AI model. AI will never be 100% accurate. There will always be a learning curve,” said Samuel. “Take a case where we automated communication between employees and the enterprise. On day one, when we went in, we were just 63% in efficiency. This is the percentage of conversations that the bot can close by itself. But 6 – 8 months later it is more than 88%.”
Experts say this learning cycle can be shortened if you feed more data to the machine learning model. The more conversations that AI bot has, the better will be its understanding of conversations and the accuracy of its response.
Related story
‘Our Conversation AI bot can shorten response times by 50 per cent and boost productivity’
