Progress, in just about any endeavor, is often driven by pioneers who see pathways where others only see dead ends. In the field of AI, one such trailblazer is Chetan Dube, whose quest to make machine intelligence approximate human intelligence led him to found IPsoft (now Amelia), the world’s largest privately held AI software company.
In Part I of this 2-part episode, we speak with Chetan & learn why traditionally risk-averse industries went all in on conversational AI and the differentiating edge it provides them; which metric (other than ROI) best captures the impact of automation; and the biggest challenges organizations are experiencing in deploying automation and conversational AI.
Guy Nadivi: Welcome, everyone. My name is Guy Nadivi, and I’m the host of Intelligent Automation Radio. Our guest on today’s episode is Chetan Dube, President, CEO and Founder of Amelia, an IPsoft company. Amelia is a leader in enterprise AI software, with a long history of innovation in automaton and conversational AI. The Amelia website states that their technology creates, “fulfilling human experiences through groundbreaking AI solutions.” Amelia’s lengthy list of happy global clients attests to just how successful their technology has been.
As a pioneer in digital labor and hyper-automation, Chetan has been on our radar for quite some time, and we’re very fortunate he’s carved a slot out from his extremely busy schedule, to join us today. Chetan, welcome to Intelligent Automation Radio.
Chetan Dube: Thank you very much Nadivi, for having me. It’s a privilege to be here.
Guy Nadivi: Chetan, please tell me a bit about your career path. I understand you were a mathematics professor at NYU. Why did you switch over to the private sector and the field of AI?
Chetan Dube: Thank you very much for the kind introduction. But, I was in Courant Institute of Math Sciences, I was an Assistant Professor. At that time, a large part of my research was centered around deterministic finite state machines, that were trying to see if we could clone system engineers’ brains. I was one of the ignorant few that walked into, this is about 1998, about the same time, it was the fall of 1998, I walked into my advisor at that time, Professor Dennis Shasha, his office. And suggested that, “Professor, we seem to be using these deterministic state machines to be able to clone system engineers’ brains. If you just extended it, we should start to get to customer intelligence and general intelligence, and we should start to approach at least bounded CRM intelligence, in a couple of summers.” I suggested that.
I still remember the expression on Professor Shasha’s face that said that, “Oh fool. Don’t you know that even the father of artificial intelligence, John McCarthy, gave up on the problem, saying that it turned out to be a lot harder than anticipated?” But, being a profound ignorant of the challenges that lay ahead, we set sail for this aspiration to trying to see if we could make machine intelligence start to become close to human intelligence.
And so, it has not been a couple of summers. In fact, we’re just about in the 21st summer. And I think when you meet Amelia, you can be the better gauge of if there is one sincere machine intelligence digital embodiment that is starting to approach close to human intelligence.
Guy Nadivi: So, that work led to Amelia, a conversational AI product, and I’m curious Chetan, what are some of the low hanging fruit best suited for conversational AI applications within an organization?
Chetan Dube: That’s terrific insight. You’re looking at, the biggest uptake we are starting to see is, I’m surprised it’s in banking and insurance. I felt that the banking and insurance would be perhaps the most de-risked sectors, and perhaps the last ones on board, to be able to take advanced technologies for some significant step gains in their returns. I was talking to one of the CEOs in fact, of a very large, one of the top three large banks here, trying to understand how BFSI (Banking, Financial Services and Insurance) has taken lead in conversational AI applications. And his assertion was that, it’s exactly the risk profiling, because, the risk of non-adoption is far greater than the risk of early adoption. His assertion, there are two kinds of companies. The companies that do have AI will be the ones that would actually have created a competitive advantage. And the companies that wouldn’t have would have in the next three to five years, will face some existential pressures.
So, the low hanging fruits that are identified, the sectors that we see adoption in is BFSI very much leading because they have done the risk profiling, and understand that this is perhaps the best de-risking measure, is to be able to adopt, to be a digital front runner. And then, we find particularly both pandemic healthcare taking off in a big way, and telco maintaining a big, huge footprint of call centers, as being a big proponent of digital technologies.
In these spaces, in these verticals, the areas that are typical to what you suggested, low hanging fruits, in banking for instance, if you look at a high friction, low margin asset, credit card replacement, nobody is making tremendous amount of margins on that. Insurance, first notification of loss, a high friction, low margin asset. Retail, the returns in the season coming up right in front. Retail has got returns and processing of returns. Again, a high friction, low margin asset. These ones are rapidly being moved towards being rendered by digital AI applications.
You also find certain things like origination, which is now a higher margin kind of an activity also. It’s starting to get moved towards loan origination, insurance origination. One of the digital banks that has done about 45,600 calls per month, having an effective origination where, their digital agents are 17% more effective than the human agents, in being able to originate these mortgages, which are higher margin assets for this bank. So those are some of the lower hanging fruits that we are starting to see.
I want to, if you’d permit me, I’d like to elaborate. This is again, credit goes to the CEO of the bank, one of the top three banks I was having a discussion with. His assertion was that, is the core getting commoditized? Is the core of banking, or the core of insurance companies, is that getting commoditized? Do you really get a different experience when you get a mortgage from JPMC in this case, versus a Wells Fargo, versus Citibank? Are you really getting a different credit card when you get it from JPMC versus you get it from Bank Of America? And is the interest rate not really coming about the same when it comes to one bank versus another? Would the core getting commoditized? Isn’t the differentiation really moving to the edge, and the customer gains and the customer capture already moving to the edge?
So, where is that edge? The edge is in the quality of customer care, and the quality of interactions with the customer and the responsiveness. That’s where digital technologies, which used to be just for costing cutting, are now starting with the maturity of digital technologies that have gotten to human, and in some cases exceeding human levels of competence, are starting to provide that edge of differentiation to banking and insurance companies. Sorry for the elaborate nature of the answer to your question, but these are all the assets that we are seeing progressively getting digitized.
Guy Nadivi: I think it’s fascinating. Now, you listed some general use cases, but I wonder if you can share with us examples of some of the more interesting use cases your clients have used Amelia for.
Chetan Dube: So one of the largest, actually top five banks in Europe, average trades done by this bank for its securities are of 10 million Euros, average securities trade. That is for its premier customers, the Total and Range, that they service. They are actually moving the entire securities platform for trading. They are employing digital solutions. I shared with you the digital bank, but I’ll also give you an example of the spectrum of not only banking insurance and retail healthcare, but also telco. I’ll give you an example of a telco where, Telefonica in Peru for example, has over 7 million calls coming in every month.
Now, here’s the interesting thing, Nadivi, that you might find very interesting is that, they are having the NPS characteristic of humans rendering that call responses, versus the NPS characteristic of Amelia, a digital agent, rendering the service. For a call volume of 7.1 million calls last month and the month of July, Amelia outperformed humans by a differential of 11 points on the NPS scale. So you can well appreciate the fact that, the numbers and the quantitative analysis is telling an unequivocal story, that digital experience from a customer perspective, as reflected by the Net Promoter Score, can actually be superior. Even when you are talking about conversations at scale, of the scale of capacity of like 7.1 million customer care calls coming in to an organization at the top of telco chain as Telefonica.
I could give you examples of insurance companies, and I could give you examples of healthcare, but I think it illustrates the point that the shift from human-rendered services to digital-rendered services is not only happening, but it’s happening gainfully, when the right technology is being employed.
Guy Nadivi: You mentioned NPS, which is a great segue to my next question. Other than ROI, is there a single metric that best captures the impact of automation and conversational AI on business and IT operations?
Chetan Dube: Guy, that’s fantastic. It’s brilliant. That’s why your podcasts are unique because, you hit at the center of why do this. Why do intelligent automation? Why do conversational intelligence? The average containment, what is ROI reflected by? What is the return on investment if you do a digital? For the most part, often a digital transformation has been nothing but the story of taking your assets, which are currently on prem assets, and moving it to the cloud. Now, that’s the first step of a digital transformation, just to be able to move the assets to the cloud. Now the question becomes, who is going to service those assets now? Who’s going to provide the customer service? Who’s going to provide the ERP? Who’s going to provide the supply chain? Is that going to be human employees? More cost-effective human employees coming from a labor arbitrage-centric company, country? Or, is it going to be digital employees?
That’s the typical S-curve. The first S-curve has almost reached maturity where, the assets have been moved to cloud for most of the companies, or the assets are progressively being moved to the cloud. Now comes the second big S-curve, which is exponentially more steep in its returns, but that requires one to have engaged in the catalyst of switching over from human-rendered services to digitally-rendered services. Now, what is the single metric that captures this impact of automation? Containment.
So, you will have, God knows, with us, we’ve talked about the fact that there are 2,100 vendors, but when we started, there was just us and Watson. Now, there are 2,100 vendors. And we have done a decent enough job of expressing to the market that this conversational AI technology can do everything. So how does a discerning buyer, how does a discerning chief information officer, a CEO, decide which technology can deliver the return on investment? Containment. What is the amount of calls that were coming in, that could be satisfactorily solved by this digital technology, which were otherwise handled by humans? That’s really the effect. The efficacy is captured by that single metric.
And if you look at like… If you tried to deconstruct that and you say, “All right, what is the total amount of calls coming in?” I happen to be a CIO or CEO, and I look at my call center and I say, “What is the total amount of calls coming in?” Now, I’ve got plenty of vendors which are coming in and telling me that, “Look, they can do my…” I’m an insurance company. What is my policy number? I want to file a claim or with a banker saying, “What is my account balance? Or please transfer this much amount of money,” essentially, all the tasks that are exactly the same as the hamburger menu. No?
Now, the CIO of NTT would tell you that, happy case scenarios, perhaps something you already know because you’re running these centers, happy case scenarios are only 14% of your total call volume. Nobody is picking up the phone and calling. Your contact center, 86% of the people that are calling your contact center are calling about a problem. That’s where most, if not all, chatbot technology is very effective at the happy case scenarios, but ineffective at the real problem case scenarios. So the containment is limited by the 14% that they are able to have. An average containment in the industry is between 6.2% and 7.8%. You can do password resets and active directory unlocks for specialized domain cases, but what are the real calls coming in? Guy, if you actually feel that the calls… I’ll give you examples from this morning for the largest insurance company here, called. “Why is it that my policy was canceled when I just paid $68.13 last month?”
Now, you try to bucket that in an IVR kind of way, or a chatbot kind of way. What chatbots do is that they classify. And there is certainly a profound chasm between classification and comprehension. Classifications will deep neural network match you with the language models and machine models to policy cancel. Now, you’ve experienced this, Guy, where you called somebody and said, “Why is my policy canceled, and I just paid $68?” And it said, “Oh, very well sir, Mr. Nadivi, let me go ahead and cancel your policy.” At which point you lose your mind, and you will try to press different buttons to be able to say “operator” or “representative”.
To illustrate this point, and just another example from this morning’s chats would be, and this is actually Telefonica, so I should just be transparent about that. “I generally pay 153 Soles. Why is my last month’s bill 168 Soles?” In Peru. You try to now bucket that, or classify that into one of the buckets and say, “Here’s the canned response,” and you will have an irate customer on the other side. So the essence of, if you asked me for a single metric, I would say the single metric is satisfactory containment. Containment of, what proportion of the calls that came in was satisfactorily handled by the digital agent? And as you can see, hopefully in the illustration I was giving, that most of the chatbots you will find to be very effective at the happy case scenarios, but very ineffective unfortunately, and sorry for my direct candor here, on the real complex problem cases.
But that’s where you need, because 86% of the pie is dependent on that, and that’s where you need a technology that can actually have a containment that starts to push that from single digit percentages, to above 50%, 60% and starts to approach human levels of containment, which are 60%, 70% plus. And then of course, there’s escalation where, even a human could not handle that. Again, apologies for the elaborate nature of the response, but I hope the math is of interest to you and your viewers.
Guy Nadivi: Chetan, no need to apologize for that or the candor. In fact, I’m going to ask you to keep the candor going with this next question. Which is, when you talk with Amelia customers, what are they telling you are some of the biggest challenges they’re experiencing in deploying automation and conversational AI within their organization?
Chetan Dube: Time to value. Guy, that’s the biggest thing that has been… Time to value first, and second, when it is mature, the extent of containment and the extent of returns that they are getting. If you look at returns, there are formulas that you can actually maintain, that can talk about the resolution rates times the volume that was being fielded, the coverage times, the satisfactory Net Promoter Score that you were able to achieve. So, the biggest challenges that people face is that, my time to value was stretched out, and the technology was moving forward. So, people feel that we could do simple, rudimentary cases, but when I come to customer facing cases, it’s the time to value is stretched. And I think that’s where AI has to build AI. Technology has to build technology. The onus should not be on humans to be building this technology because, AI has to have the learning capabilities to be able to absorb, process mine, and AI has to have the learning capability to be told natural language, what it should be doing to modify that, and for it to itself build an incarnate, an artificial intelligence system that can provide good value off the gate.
In fact, a good yardstick that we maintain is the time to value for any deployment for our customers must be within 30 days. We have been successful in that. A large part of the credit goes to a bilateral commitment from the customers’ side, to making sure that they realize value in that timeframe.
Guy Nadivi: The vision statement on Amelia’s website states that, “Our work does not replace the people who make companies successful. We streamline IT operations, automate processes end-to-end, and enable true conversational experiences with cognitive AI.” This statement to me, clearly reflects the fact that introducing automation into an organization can trigger resistance from some employees due to fear of job loss or radical job change. We refer to that resistance here on the podcast as robophobia, and it can in fact pose a serious cultural obstacle for enterprises deploying automation on their digital transformation journey. Chetan, what tactics have you seen prove most effective in overcoming robophobia?
Chetan Dube: Well, thank you. I’ve learned a new word, and with your permission, I’ll use it. Robophobia.
Guy Nadivi: Please do.
Chetan Dube: Thank you. Overcoming robophobia, Guy, there are two schools of thought. These robots or these AI agents are a great thing, the optimists. And the other school of thought, the pessimists are saying, the singularity that this is obviously going to be the Hawking Club, this is the final invention. I ask a third question. Do we have a choice?
I have had the pleasure in these two decades, of meeting the chief executives on three continents, who are running very large banking, insurance, and retail, and healthcare, and telco organizations, who are very much concerned about… They’re very much into this welfare of their people and their societies. And I have yet to find a single executive who was told, and who saw McKinsey saying that, “You could have a 35% margin enhancement by adopting digital modes of delivery, versus a 40% margin compression if you did not have adequate digital adoption.” With a spread of that kind, I’ve yet to find a single chief executive that said, “Maybe we should just not go this route of digital.” Do we have a choice, Guy?
I think this is… The technology has got such an overwhelming advantage in its returns. In the next few years, it’s popular school. By 2025, we’re going to have companies that have adopted AI for gains, that will start to dominate the Global 2000, and the companies that haven’t adopted AI that will start to face extinction. 40% of the Global 2000 are not Global 2000 since the year 2000 alone. So just look at the mathematics of it and say, what is the catalyst, if it’s not digital? It is the digital catalyst, and that is causing these things.
So, I would say to anyone that is… And by the way, this is not a question that is taken very lightly. I would say, I was privileged to be invited to the House of Commons in England, and they asked the same thing about, “What do we do about our society?” I was invited to the Palais de Luxembourg by the ex-prime minister of France, and again, the same question. France, of course you can see, these are countries that have a very socialistic mindset also. They were also very concerned about the fact that, what do we need to do about this? The people have to realize that, technology is going to be the biggest asset. What on a given day, on a given day, a mind like yours, let’s talk about a mind like yours. How much of your brain do you use for creative thinking on any given day? Less than 25%? Is that not a colossal waste of a brain like yours and your listeners, that 75% of the time they’re driving a car, they’re crossing the street, they’re handling their invoices, they’re doing the mundane chores?
Are you not shackled by the ordinary chores that haunt your existence? Who is going to come along and liberate you from saying, “Monsieur Nadivi, I got all this for you. You don’t have to worry about all these mundane chores. I will take care of them. You can focus on high creative thinking, on elevating your work to a higher plane.” Technology is the ultimate liberator for mankind. It’ll allow us to be able to broaden our horizons, and it always has. It’s the robophobia — I would say, “What happened to the horse and carriage driver?” I would ask those people. I don’t see them really unemployed. They’re moving on to driving cars, flying planes, tomorrow flying space rockets. Isn’t it the inevitable course of history? I will ask the same people, what happened to… How many of us were farming at the turn of 1800s? 90% of us were farming. No? Because subsistence was the only way to get food on our table.
How many of us are farming in North America? Now I will ask the same people. 2%. So what happened to the 88%? I don’t see them unemployed. They are actually running podcasts that can influence worldwide people’s opinions, and give them a better future, and more efficient organizations. Don’t turn against progress. Embrace progress. Embrace it to be able to lift you and your quality of life, by taking away all mundane chores from you so that you can have elective… And look, if you really do want to do mundane chores, I’ve not found many people who say, “For 30 years, all I wanted to do was to tighten the same nut and bolt in the same exact spot in my factory.” But if you do want to do that, you should have the elective freedom to do that, as technology provides you the minimum viable income that is required for you to be able to exercise the elective and creative freedom, to be able to engage in those things. Again, I find myself quite passionate about these things, and the elaborate nature of my answer, my apologies.
Guy Nadivi: Again, no apologies needed, but continuing on the theme of labor, Chetan, can you talk a little bit about the labor shortage happening right now in the US, in industries such as manufacturing and services? How can AI help companies struggling to fill job openings?
Chetan Dube: This is obviously… US labor shortage is an interesting one where, the only thing that catapults the country to the next realm or keeps it in the front of being the number one economy in the world, is the labor force. It’s about the 350 million Americans that are… Take the total productivity of the GDP of the country divided by the 350 million, and you have a per head productivity assessment. And that basically gives you an assessment of what the country is producing.
The labor shortage is a very key point. When a labor shortage happens, you run the risk of having your GDP, which has rebounded with 6.5% in the last quarter, you run the risk of it slowing down. So, what options do you have, Guy? You could say, “I want to be able to go to the wage arbitrage countries and find some people to be able to supplement.” Well, wage arbitrage countries, particularly for manufacturing, you would be sending a lot of the jobs also off shore, but besides that you don’t have the adequate force there. But, you would also be often in cases found compromising the quality of outcomes that you are able to achieve.
AI on the other hand can help you not only retain that footprint, fill that manufacturing or services opportunity, but also provide you the quality that is needed for you to be able to turn that from being an import to an export. You can actually now, since you have a digitally efficient, scalable, and cost effective way of manufacturing, of providing these services, you can export to the rest of the world. Look at the digital opportunity for digital customer agents. McKinsey estimates this to be a $2.9 trillion opportunity by 2025. $2.9 trillion. Who is going to take a big bite out of the pie? Which country is positioned? Which company is positioned to actually say, “No, no, I will provide these services.” And when you start providing those qualified, quality digital services, you have that much headroom to be able to grow into providing the services, not only to your country, but also to export those services, digital services worldwide. That’s why there seems to be this Darwinistic race amongst countries as well, that is happening, is you find different countries on different places at the scale of algorithmic and AI evolution.
President, CEO, and Founder of Amelia, an IPsoft company
Chetan Dube has served as the President and CEO of Amelia, an IPsoft Company, since its founding in 1998. During his tenure, he has led the company to create a radical shift in the way IT is automated and managed, and introduced the industry to market-leading Conversational AI technology. Previously, Chetan served as an Assistant Professor at New York University, where his research was focused on deterministic finite-state computing engines. Chetan is a widely recognized and sought-after speaker on autonomics, Conversational AI, cognitive computing and AI’s impact on the Future of Work. He also serves on the board of numerous IT-related institutions.
Chetan can be reached at:
Conversations (Chetan’s personal site): https://chetandube.ai/