By John P. Desmond, AI Trends Editor
AI has changed the game for service providers. Client companies now expect the service provider will deliver on the promise of AI for them, or help them get moving in the right direction. We spoke about trends in AI and services to executives of two service providers recently: Asheesh Mehra, co-founder and CEO of AntWorks; and Prabhdeep (PD) Singh, VP of AI at UiPath.
AntWorks, founded in 2015, is an AI and intelligent automation company, with a platform that understands every data type. The company digitizes every bit of information from a wide range of industries. Mehra is co-founder and group CEO; his background includes seven years at Infosys working in business process outsourcing in Asia Pacific, Japan, and the Middle East. The company offers the ANTstein intelligent automation. It supports robotic process automation support, intuitive machine learning, and natural language modeling capabilities.
UiPath of New York City is an AI enterprise software company known for AI, machine learning, and Robotic Process Automation. The company was recently positioned in the upper right Leaders quadrant in Gartner Magic Quadrant for Robotic Process Automation Software. Prabhdeep (PD) Singh, VP of AI at UiPath, was at Microsoft for nearly 10 years before coming to UiPath a year ago. He led the product and business teams for the Microsoft Sales Intelligence AI solution.
Mehra and Singh were interviewed separately by AI Trends Editor John P. Desmond.
How has AI changed the game for service providers?
Asheesh Mehra, co-founder and CEO of AntWorks: Some customers expect AI to be a magic wand that can start delivering results overnight. However, an AI engine is not magic. It needs training at the back end before it can start performing an action. It can start learning from the representative data set that is received over a few months. Then it can start its machine learning capability to help make intelligent decisions, or start predicting or inferring dependent of the representative data set it has seen over the months it has been deployed at an enterprise.
So, is it changing expectations of customers? The answer is absolutely, yes. Some expectations are realistic, and some expectations are unrealistic. Is it impacting the end customer of enterprises? It is. In some ways it’s impacting them by making their lives, their day-to-day jobs a lot easier because it is now a helping hand, or a joint force with the human. When you put both together, you get a far superior outcome.
Prabhdeep (PD) Singh, VP of AI at UiPath: The way the older service providers would typically solve business problems would be to have a human sitting in some back room doing this stuff for you manually. But now the automation has reached a stage—and the set of technologies that are available to us have reached a stage—where you can optimize pretty much every and any business process. If you remember, the name of the game for these BPO [Business Process Outsourcing] providers was to get down the cost. That’s why you couldn’t run call centers here in the US, because the cost of employing humans was just too high.
That’s when people started going to places like India, Vietnam, and all these other places where you had English-speaking populations, but it was much cheaper to hire people. It was more of this cost optimization, cost-cutting exercise. With AI and automation coming in, there is a paradigm shift happening in the sense that you can actually increase the productivity of those humans and drive down the costs even more. We talk about this in almost every conference that I’ve gone to. If you look at the workforce productivity for the US over the last decade, it has pretty much plateaued. After we had a saturation of PCs, pretty much for all and every knowledge worker. And now in order to increase the productivity of those information workers in the workforce, you need more of AI and automation. We are seeing that, and many of our customers are getting monetary and productivity gains by automating and deploying AI in their business processes.
Is AI delivering?
Mehra of AntWorks: That’s a very loaded and very difficult question to answer. Yes, it’s delivering in certain spaces and in certain areas. I think AI is over-hyped and not delivering in certain other cases. If I had to use an example from the insurance world, I think AI is delivering on its promise for processing claims, being for your health, or your house, or your car. It is delivering there. There is room for AI to be improved and enhanced to deliver the outcome it is promising in some other industries, such as financial services.
If I was to summarize that, I’d put the bar right in the center and say depending on the use case and depending on the industry segment, AI is delivering; and for where AI is not delivering, it has not been exposed and trained enough in those spaces.
Singh of UiPath: When AI works, it’s magical. I’ve seen it work in both large companies and small startups. I’ve seen it save lives. I worked on systems that can do things like readmission prediction. It can predict if a patient is going to come back within 30 days, and the doctors can look at it and say, “Okay, let’s not discharge this patient right now.” If you have a system like that, you’re actually saving lives, because you’re not sending really sick patients home where adverse things can happen to them. You’re also saving money, because if you look at the Medicare/Medicaid guidelines, if the patient comes back within 30 days, the government is not going to reimburse you for the readmission.
The problem right now in the AI industry is what we call the last mile problem. If you look at the AI deployments, only 4% of CIOs have put something in production. Almost 90% to 95% of CIOs want to do something with AI. They know kind of where AI can be useful. Actually putting a system into production is a completely different beast. So once you have a machine learning model that works, you need to put it into production, have it interact with humans, with the existing applications. That’s where RPA [Robotic Process Automation] is useful, because RPA is the last mile vehicle for all things AI.
Are there problems for which AI is not a fit?
Mehra of AntWorks: If you take a step back and ask what is AI, the definition varies. In my view, AI is all about learning and then a machine taking intelligent decisions or providing accurate predictions on the data that it has received. Do we say, “No” to customers when we think we or the AI is not equipped? The answer is absolutely, yes. We do say, “No” to customers when we think we cannot deliver a particular piece using our machine learning or other algorithms. Because as I said, the expectation might be that AI is a magic wand.
The fundamental philosophy at AntWorks is, “Say no where you have to say it. When you say ‘yes’, get it right the first time.”
Can AI be deployed in every single use case in an enterprise? The answer is no. I don’t think AI is mature enough to go out there and solve every kind of challenge today that an enterprise experiences. We see a lot of room for the AI engines to be trained to become smarter and more intelligent to deliver to customer expectations.
Singh of UiPath: I will say, the things that are non-digitized are problems that you cannot optimize with AI. You see many AI use cases in sales and marketing, because sales and marketing is highly digital. If an industry has gone through digital transformation, that’s where AI can be very useful. But if you have antiquated processes, and you actually never digitized, then it’s a little difficult. For example, if there was a company doing everything paper-based and old school very well, the first process is to get that paper scanned and put it in digital format before you can apply anything intelligent on top of that information.
Do AI engagements take more time than the former non-AI way of solving problems?
Mehra of AntWorks: No, absolutely not. One of the whole drivers to make a business case positive is to cut the time it takes to do projects. The whole objective is to speed up the business process and to ensure that accuracy is a lot higher. So does it take more time? The answer is no. If it does take more time, it’s probably taking a lot more time because not enough training has been done for the engine and it has been deployed prematurely.
No different to when you bring a human being into an enterprise, the first four weeks of them coming in or the first three weeks or the first six weeks, is to train that individual on how to perform their job, or how to deliver that outcome. An AI engine is no different. If you expect an AI engine to start delivering the results that you’re expecting without spending enough time on training the engine, it will not deliver for you.
So the marketplace needs to understand how to make your AI or machine learning engine deliver results for you. There are no shortcuts. You need to invest the right amount of time and expose the AI engine to the right amount of representative data for it to deliver results for you.
Singh of UiPath: I would say no. If you planned your system correctly, it’s much easier to solve problems and much more effective to solve problems with AI versus the older way. For example, if you remember the old school real estate agents, there were good agents and there were bad agents. The really good sales people didn’t need any of these electronic nannies and electronic aids like CRM systems. They were just going in, doing it the old school way, pounding the pavement, being really good at selling stuff.
My point is if you have a problem which is highly dependent on human expertise, it will take time to have AI go in and improve it. But if there is a process where you are not realizing the human efficiencies to the maximum, that is where AI can make a big difference.
What are your challenges?
Mehta of AntWorks: I have a few challenges today. One of my first challenges is market share. I’m a four year old company. I’m against all the names from a competition perspective. They’ve all been around for longer, have captured a large market share and I’m playing catch up. So we took time to build the technology and the whole platform out while they were out there selling single technology tool sets. So now it’s my turn to capture market share.
My second challenge is that the understanding level of the buyer varies to a large extent. On a scale of one to 10, a very large percentage of buyers are at three and below. There is a very small percentage of buyers that are in the seven, eight, nine, and 10, from a rating perspective. That starts becoming a challenge because the expectations that they have, and what reality is are a vast distance apart. So we need to deploy larger resources to help educate the marketplace.
The third challenge is around the expectation of what AI is. There is just so much hype and white noise in this whole AI space right now. This puts pressure on you as a product company because people dream up things and it becomes a challenge. The minute you start pushing back, and say, “That’s not really what we can deliver, or a machine can deploy”, you start creating a sense of dissatisfaction in the buyer community. But you are being realistic. So, that truly is another challenge for me. Those are the three challenges for me today.
Singh of UiPath: Once you start deploying these systems in an enterprise at a very large scale, a couple of things happen. Enterprise software is a very well-understood area. The people who deploy software and applications in their enterprise, with traditional software applications, have a very well-understood product. With AI applications, it’s not just a matter of deploying software. AI models work off of data. The data must be clean, the data pipelines need to run properly, and the AI models need to work well. That whole vision is our principle. If, for example, the data shuts off at the input, the model starts behaving completely differently. For example, I had this vision readmission model, and it uses a patient feed which was giving it the economic data of the patient. And so we know that people who are low income patients, they have a higher risk of readmissions, and also if you’re not getting that feed, the model won’t be confident enough in making these predictions. It might go completely haywire. So scaling AI is a big challenge.
We have a discipline called DevOps to address the way traditional software is handled in the enterprise. What we need is a DevOps equivalent for AI. We in the industry call it MLOps, or machine learning ops. It’s basically the discipline around practices to manage, deploy, and create these models in a structured way. One of the offerings in our product, AI Fabric, is an MLOps system for the data scientists, who might not really understand enterprise software deployment cycles. We simplify it for them. We say, “You just create a model, the rest of the stuff on the deployment, DevOps, MLOps side, we will take care of.” So deploying AI applications in the real world we have seen as a challenge.
The third challenge, I would say, is the ROI quantification. The business owner needs to know the impact of putting the machine learning model into production. Is it hurting or improving the overall business? You need a system which can quantify the return on investment when you’re deploying AI. Typically, you would use BI tools for this. We are working on having an embedded analytics or BI offering, to give customers this ROI visibility. You need that quantification system in place.
What does the future hold for your company and for AI in business? Where is it going?
Mehta of AntWorks: John, we’re super excited. I’m speaking to you today from a delivery center in Bengal. I’ve just spent the whole day sitting next to my developers, talking to my product leads and the head of my product, and it’s just super exciting. It scares me when I look at possibilities that are in store for us over the next six, 12, and 24 months, and just what machines can do if trained correctly or if deployed correctly. I’m super excited about the next two years also. We’ve grown more than 350% in the last quarter and a half.
I did a Town Hall [in India] this afternoon and the last time I was here a quarter ago, I had, I think, 40 people. Today I addressed 165 people. I’m hiring people to support customer demand. So, there is a huge amount of potential and growth. Over the next 12-24 months, we are looking at dramatic growth. Automation and AI are in the top three agenda items of every Fortune 500 board today. We are addressing that challenge for those organizations.
While there is a huge opportunity, all the organizations in this space need to be responsible to not over commit and under deliver, because that will start taking away the belief in what AI machine learning and automation can do.
Singh of UiPath: I’ll be honest with you, where we stand right now as a company, as an industry, I see a whole tsunami coming our way. It’s not a bad tsunami; it’s a good one. Right now, most of these enterprises, they’ve just gone through digital transformation. They’re putting all these new-fangled systems in place. You have Salesforce for your sales system. You have Microsoft office. You have different cloud applications that you put in production. I can name you many companies, for example oil companies, they have completely digitized processes now, for even things like drilling reports. CPG [consumer packaged goods] companies who have digital processes for designing the labels that go on to their products. They also have inventory management systems and logistics management systems.
They’ve just put these systems in place. The next thing, or the next ROI that they want to get out of the digital transformation, is to try machine learning and some automation. They want to try putting AI in some of these processes, to inject some automation and AI, to see if they can make them more efficient. I would say in the post-digital transformation era, the future for RPA and AI looks really great. These will be the technologies that are at the cutting edge of this post-digital transformation age.