Interviewed By Akhlaqur Rahman Sachee, Team MBR

Hishab is a telephony network-based conversational AI startup that is currently offering diversified services under the umbrellas of voice pay, customer engagement, and AI-driven call centre connectivity. Its services are being used by some renowned microfinance institutions, banks, and telecom operators in Bangladesh. The startup aims to make it easier for the impoverished, less educated, or even uneducated population in Bangladesh to receive various services over telephony networks with the help of conversational AI. Team MBR was in a conversation with the founder and CEO of Hishab, Mr. Zubair Ahmed, and had the opportunity to learn about his inspirations behind forming the startup and his future aspirations surrounding Hishab.

Akhlaqur Rahman Sachee: Founded in 2017, Hishab is a telephony network-based conversational AI startup led by a team consisting of professionals from diverse countries. Would you kindly share with us how you came up with the idea to form Hishab?

Zubair Ahmed: Before forming Hishab, I worked as an investment banker, and my job was mostly to invest in the US and European companies. Back then, I invested in one of the companies called API.AI which used to build dialogue engines that work from mobile apps or web interfaces. It was very similar to Siri or Google Assistant. The company got bought out by Google back in 2013. Later, after Google bought it, they changed the name to Dialogflow, which eventually controlled close to 90% of the conversational AI market around the world. The problem with the product that we sold to Google was that it only focused on conversational AI that works from the web or smartphones. But it was not able to do it over a telephony network. With the help of telephony network-based conversational AI, you can talk over a phone and run the conversational AI over the phone. But the aforementioned product was not able to do that. Hence, I decided to work exactly on that. My initial plan was to launch my product in Indonesia and experiment with it there. But, then, I guess the love for my country kicked in, and I decided to do that in Bangladesh first. There were a lot of challenges because the speech recognition engine for Bengali was not available, the text-to-speech engine was not available, and natural language processing was in the infant stage back then. So, we had to do a lot of R&D over the course of eight years to come up with the technology and also wait for the world to catch up to provide us with the platform that we actually needed to build that technology. Finally, about one and a half to two years ago, we were able to build the product that we planned. Now, one of the reasons for which I wanted to build a conversational AI over a telephony network was to do something that makes technology accessible to anyone regardless of digital literacy, thereby making the use of a product like Dialogflow available to everyone, which I had built before. The only way to do that was to build a conversational AI over telephony networks. So, this was the primary motivation for building Hishab.

Akhlaqur Rahman Sachee: Hishab has surely made it easier for the clients to enjoy the services of microfinance institutions, mobile financial services providers, banks, non-bank financial institutions, telecom operators, and so on. May we know how Hishab is playing its role in promoting financial inclusion?

Zubair Ahmed: The main obstacle to promoting financial inclusion, especially in emerging countries, is a lack of digital literacy and financial literacy. Both are low in the case of Bangladesh. In Bangladesh, our digital literacy stands at 6.1%. Though all the numbers by the government say that about 23% of the people are digitally literate, around 6% of the people are actually digitally literate, and the rest, 17%, are being assisted by them. For instance, when we studied the user base of some of the MFS players, such as bKash or Rocket, we found that 23% of the user base is able to send or receive money by using either apps or kiosks. Kiosks are basically small mom-and-pop shops that send or receive money on their behalf. Only 6% of the user base of bKash or Rocket are able to send or receive money on their own, but the rest, 17%, need help from those shops to send or receive money. Now, our technology has the ability to improve the scenario because it makes bKash or Rocket-like services accessible to everyone. Anyone can now just talk and perform transactions. The users do not need to know how that works. They just need to talk, and it just magically happens. This is one of the main reasons why our technology has been doing so well, especially among the microfinance institutions in Bangladesh. 95% of their clients are digitally illiterate without any financial knowledge. Even those people are able to perform financial transactions with the help of our system. So, we believe this could play a pivotal role in promoting financial inclusion. You do not need to come up with new technologies to promote financial inclusion. You can rather make the existing technologies available for everyone, such as performing transactions simply by talking.

Akhlaqur Rahman Sachee: Hishab’s services are being used by some renowned microfinance institutions, banks, and telecom operators in Bangladesh to provide customer services. Would you kindly share with us the scope of applications of telephony-based conversational AI technology other than the ones Hishab has already explored?

Zubair Ahmed: There are currently three areas of focus. The first one is support, the second one is sales, and the third one is finance. When we say support, it essentially means a customer support area where users are actually able to ask any question and get it answered, which we generally call the FAQ. They can know about the product or service information and recommendations about the products or services. On top of that, they are able to lodge complaints or give feedback about products or services. Then, our AI is able to segregate those andmadd those to the back-end customer relationship management tools, or CRMs, that the end users use to manage customer queries and satisfaction levels. Companies can utilise our generative AI to survey as well. Traditionally, we send brochures or questionnaires to the target groups or ask them the questions and put tick marks on the answer sheets so that we can analyse those. But our AI can hear and gather the voices of respondents, which enables categorization in the right manner.

When it comes to sales, our product is also able to sell products or services for you. For instance, the distribution houses send sales representatives to different routes. Usually, one sales representative covers one route in a day and works around six days a week. They cover six routes in a week, and each route generally has about 70 to 90 shops that they have to visit. Of course, it is quite difficult because most of the time, the shops do not respond in the right manner. So, sometimes they effectively cannot cover that many shops in order to generate sales. But our AI can generate sales from probably 50% of those shops because those sales are repetitive. The shops that the sales representatives visit and are able to generate sales from would buy the products either way. This is the task our AI can take over so that sales representatives can focus more on the shops from which it is difficult to generate sales. Thus, AI can increase sales for distribution houses or manufacturing companies. We are about to run a pilot on this service in collaboration with one of the largest FMCG groups in Bangladesh.

When it comes to finance, users can top up their balances from their MFS accounts just by talking over their phones. They can also make payments on their utility bills, such as electricity, water, gas, and so on. Also, they can make payments to their educational institutions, offices, and other places just by talking. On the other hand, our AI can help regarding the stock market, where less than 1% of our population is active. This is where our AI can play an important role in educating people, teaching them about the benefits, and helping them to invest, including the purchase of stocks. Our AI can do a great job of convincing investors on how and why a certain stock is beneficial for them to invest in. So, that is the innovation our product can bring to the finance industry.

Akhlaqur Rahman Sachee: Because handheld devices are susceptible to loss or theft, unauthorised individuals may take wrongful advantage of conversational AI technology to carry out fraudulent transactions. Would you kindly share with us what safety measures Hishab employs to identify such individuals and prevent fraudulent transactions?

Zubair Ahmed: We ensure safety and security in three effective ways. There could be theft, or people could wrongly use the conversational AI in order to perform fraudulent transactions. The first layer of security that we offer is biometrics. It is a combination. Your voice is your password. Secondly, you can set a certain password, which can be a phrase or a sentence as opposed to a four-digit pin number. For example, you can set ‘I am not feeling good today.’ as your password. The third security measure is security questions. We offer the option to ask the end users random two or three questions. All these indicate that the voice security system is actually much more robust than the conventional security system.

If you lose your credit card or phone, anyone can perform unauthorised transactions using those. The NFC has made it even easier. But our system offers three layers of security. If the voice matches, there is a password-protected second layer. You can also deploy the third layer, which requires answering security questions. So, it is much more difficult to breach the security system when it comes to conversational AI than on mobile apps or web pages.

Akhlaqur Rahman Sachee: Bengali is a language that has been enriched by different accents and dialects. May we know the level of accuracy of Hishab in interpreting the voice commands from clients of different accents and dialects?

Zubair Ahmed: In terms of the accuracy of our speech recognition engine, it is currently the best in the world. It performs at least 30% better than Google’s speech recognition engine. In the case of speech recognition engines over telephony networks or non-telephony networks, we perform the best in the world when it comes to Bengali, and we are continuously improving it by training our speech recognition engine with different local accents and dialects as our product is being used. Our speech recognition engine is far better than any speech recognition engine in the world, especially in Bengali, and, of course, as we move on, we currently cover almost all major dialects in Bengali. However, we still face some difficulties in recognising the accents of Chittagong, Sylhet, and Noakhali. As our system is being used every day by those users, we collect that data, and I am sure that the accuracy in those dialects will improve.

Akhlaqur Rahman Sachee: Hishab holds more than 35 approved patents and more than 20 patents under process in 23 countries. Would you kindly share with us some noteworthy cross-border business opportunities that Hishab has explored so far?

Zubair Ahmed: Hishab currently holds more than 35 patents in 23 countries around the world, and some of these countries include India, Indonesia, the Philippines, Bangladesh, and so on. More than 2.80 billion people are under our patent coverage. Our vision is that we would like to license our technology to some system integrators around the world who are the best players or the biggest players in their countries. We are currently having discussions with some system integrators, and we would like them to sell our service all around the world. In other words, it is going to be probably the first time a high-tech fifth-generation technology from Bangladesh is going worldwide. We already initiated that process by doing that in Japan. In Japan, we currently have two system integrator partners who will be aggressively selling our product throughout Japan, and thereby, we are showing the world that Bangladesh can produce world-class generative AI technology too, and we are not far behind. In fact, we are the ones who are leading the world in generative AI with more than 35 patents and more than 20 patents under process. We are a major force to reckon with right now in the field of generative AI.

Akhlaqur Rahman Sachee: Hishab is currently offering diversified services under the umbrellas of voice pay, customer engagement, and AI-driven call centre connectivity. Would you kindly share the details about the revenue drivers of Hishab?

Zubair Ahmed: We earn revenues basically from two areas. Our model is pay as you go. As you use the service, you just pay us for the period that you use it. For example, if our AI speaks for five minutes, we charge a certain fee per minute. There is no surcharge, there is no additional fee, there is no hidden fee, there is no upfront fee, there is no fixed fee, and there is no SaaS fee. You pay for what you use. The second area is sales or marketing. For example, you want our AI to call millions of people and gather data on your behalf. In those cases, we have three options. The first one is called an impression. When users just hear the messages, you pay a certain fee. When users speak about certain things for sustained periods of time over the phone and show interest, we call those leads. We charge a certain fee for generating leads. Finally, when users request downloadable links or payment links to avail certain services from certain companies, we call those hot leads. We also charge a success fee. So, basically, there are two payment models. One is pay as you go, and the other is success-based. A success-based model is only applicable for finance or sales-based solutions.

Akhlaqur Rahman Sachee: Hishab has recently received an investment of BDT 2 crore from Startup Bangladesh Limited. May we know something about Hishab's business expansion efforts in the near future?

Zubair Ahmed: As you know, it costs lots of money to actually keep yourself updated with R&D to stay on top of the game in the field of generative AI. BDT 2 Core is nowhere near enough to actually fund that kind of research. We continuously raise money from international corporations, venture capital funds, and large corporations to fund our research. Our plan is to raise about USD 30 million over the course of the next six months, and we have already received a commitment for around USD 25 million. We want to use that money to further enhance our capability in generative AI technology and build the best one for Bengali. We currently have the best generative AI for Bengali in the world. We call it the PIA Large Language Model (LLM). That is our current version, and we will continue to work on it. The next plan is to create a multi-model LLM, which means that you will not need an additional speech recognition engine or text-to-speech engine. Everything will be on the LLM. Our LLM would be able to see things and talk to you just as you are talking to a human. If you show it something or certain things, it will be able to recognise it right away and generate a conversation with you on that particular topic. So that is our next goal by the end of the next year. For that, we are continuously raising money in order to conduct those research projects.