Redesigning Enterprise Work Around AI Employees With Ema

[00:00:05] What changes when AI stops assisting one employee and begins owning complete business workflows? Well, my guest today is founder and CEO of a company called Ema, spelled E-M-A. And he's joining me to discuss why companies need to rethink work rather than just attach agents to processes that were designed for a completely different era. And Ema stands for Enterprise Machine Assistant.

[00:00:35] And my guest describes an AI workforce in which agents coordinate across systems, data, teams and approvals. And we examine a deployment that is supporting around 240,000 employees across 65 countries where Ema reportedly reduced average response times from five days to under five seconds.

[00:00:59] And we'll also bust a few myths along the way and why imperfect data shouldn't become an excuse for a delay and how organisations can raise an agent's autonomy as that trust continues to develop. And yes, we will also dig deep on why human oversight must always remain available for sensitive or high-profile decisions.

[00:01:24] So today's conversation will ask every leader to redesign the work, not simply automate the paperwork. And if that's off a light bulb moment, you're going to love this one. So enough from me. Let me introduce you to my guest now. So thank you for joining me on the show today. Can you tell everyone listening a little about who you are and what you do? I am Surujit Chatterjee. I'm CEO and founder of Ema Enterprise Machine Assistant, Ema.ai.

[00:01:53] Well, thank you for sitting there with me. I did a little research on you before you joined me on the call today. And I was reading how you've argued that we're asking the wrong question when we debate whether AI will replace jobs. And a quick scroll down our news feeds on platforms like LinkedIn, that's all everyone's talking about. But you argue we should be asking how organizations redesign work around autonomous AI employees.

[00:02:21] So tell me more about what that shift actually looks like in practice and why it's such a different way of thinking. Because the workplace was designed for a very different era, really. And it does feel like a new mindset is needed. But tell me a bit more about how you see this. Absolutely, Neil. Look, the easiest way to think about this is how we have seen technology change work or affect work even in the past.

[00:02:46] For example, when computers came in for the first time, you didn't just say, OK, wherever I put a checkmark on a paper, I'll put a checkmark on the computer. And it's still the same number of people or same set of process. And it's like instead of a paper, I have a screen. That's not what we did. We actually reoriented, changed the entire process. Right. And human had, of course, had a role.

[00:03:14] In fact, the advent of computing and software created more jobs than the overall net jobs were lost. In fact, Internet created more jobs. All those jobs are well established. I believe the same needs to happen, except that the change now is 10x more. Now you have intelligence on tap. How do you use it? That's up to us. Right.

[00:03:41] So even though it is intelligent, the new technology, generative AI, it's not going to go and fix your company on its own. You still need to have humans manage it, commander it, tune it, verify what it's doing, put the right card rails, train it, all of that.

[00:04:04] I think the most successful companies will do this in a way where they're not just replacing existing processes with some one-off AI agents here and there, but rethinking end-to-end business processes.

[00:04:19] Rethinking how we can reimagine it using an intelligent kind of entity on our side and how humans will collaborate with this new intelligent entity. That's what we talk about. We talk about AI employees. That's what we have built, our AI workforce.

[00:04:42] So AI workforce is basically a number of agents working together hand-in-hand, coordinating a number of AI agents and automating entire business process, not just one-off assistant, personal assistant, which is like a co-pilot or chatbots of the past. So it's a new era, and only a few companies today are thinking this way.

[00:05:05] It can be quite scary to think, oh, I need to change the entire business process for every function I have. But I believe every company will need to go through this. We call it an agentic business transformation. That's what is coming. And I think there will be many people listening in organizations that have plenty of AI pilots, but not so many meaningful business outcomes.

[00:05:31] I'm curious, from what you've seen at Ema, what is it that is separating the companies that successfully redesign their workflows around AI from those that simply bolt AI onto existing processes and, hey, this is the way we've always done things? Yeah. I'll give you an example. One of our largest customers, Wipro Systems, they are a global professional services company, around 240,000 employees worldwide in 65 countries.

[00:06:00] So very complex operations and particularly employee experience is very hard to get consistency across because every employee in a different country, they have different benefits, different laws and regulations. Also, it's a very mobile kind of employee base because they are in services business. They often fly into the client's location and so forth. So very complex.

[00:06:27] What they have done is really put EMO at the center of the entire employee experience. They call it WeNow, internally, WiproNow, and it's powered by EMO. So around 70 odd different use cases from hire to retire. The employee base, don't go anywhere else.

[00:06:49] Just come to EMO for anything from onboarding, expense report or travel or even booking a cab, payroll, checking the pay stub to changing things in the payroll. So taking any actions, asking for any clarification, filing for vacation, whatever you think about. How we do it?

[00:07:13] Basically, in our employee experience layer, our AI employees are sitting on top of 100 plus different applications they have in-house. Some of them are standard applications like ServiceNow, Workday, but many are in custom built applications. We are sitting on top of terabytes of data, policy data, other types of data. Basically, we built this entire context graph of the enterprise.

[00:07:41] And the employees get a very seamless, simple experience. Also an experience that's assistive and that actually helps get things done. So fundamentally, the role of HR operations have changed. And they talk about it in various case studies. Instead of HR operations actually answering employees' questions, they are mostly managing AI employees now.

[00:08:07] They're giving feedback to AI agents or AI employees. They're figuring out, okay, how the process can be even further improved to streamline experience. And the experience and the impact is very, very real. They used to take average of five days to respond to any employee requests. And that has cut down to like less than five seconds or so. Employee happiness has increased, employee CSAT, by almost 20 percentage points, right?

[00:08:37] The number of people needed, even though that was not originally intended, has been, has cut dramatically from like 1,000 people to like 550 people. They have been able to release these people to other areas of in the business and to do more interesting things and so forth. So the idea is the role, individual role itself will change from everybody will be a manager.

[00:09:06] You are managing now not just humans. You're managing AI employees. Everybody has to think how I work with AI, right? And you have to learn how to give feedback to AI correctly, how to verify where to set the right card rails. When should human be in the loop? It is work, significant amount of work and some upskilling, but very doable.

[00:09:32] And again, when I was doing a little research on you, your recent work on healthcare particularly stood out because I think it was, you said that technology doesn't transform organizations, but operating models do. Is this a lesson that's relevant outside of healthcare too? And what could maybe leaders listening in every industry take away from that too? Absolutely.

[00:09:56] I mean, healthcare, we do a lot of work in prior authorization, automation, also a lot of work in patient access, having patient like easily scheduled with doctors, like first level of triaging what ailment they have and directing them to the right specialist and so on with large providers. But yes, the whole point is, look, even with internet, the operating model changed a lot.

[00:10:24] For example, even if you look at the retail industry today, you can buy online, collect in store or go to store, find something, but it's not there in the store right now. You can order from store and it will be shipped online to your house. These are all process changes. You couldn't have done it like 10 years back or 12 years back. You couldn't do this because those separate businesses and so on.

[00:10:52] And they have understood people buying patterns and how they operate technology. Very similarly, I think every enterprise now need to think through their operating model, like how their customers will interact with them. Customers may prefer actually interacting with an AI agent to get something done rather than waiting on a call for three hours on a customer support.

[00:11:26] Customers may prefer to think through their operating model, like how their customers will interact with an AI agent to get something done. Customers may prefer to think through their business, like how they operate and how they operate. The big thing here is trust and the guardrails. Will I be able to trust that the AI agent or the AI employee will do this right thing?

[00:11:55] Do I have a way to bypass it or work around it? If I really feel I cannot trust, just like if I really feel like I'm a human, I can talk to someone else in customer support. There's no difference in my mind. We sometimes will be like, okay, can I talk to your manager? Same thing. Can I talk to your real human manager? Maybe because I don't think I'm getting anywhere.

[00:12:20] As long as those work arounds are available, I think people will accept AI a lot more. And you can see it even in like very different areas, like autonomous cars. A lot of trepidation. Will people be scared to get into one? It is kind of an eerie feeling. There is no driver. You are getting into this car. God knows what will happen.

[00:12:45] But it has been most largely normalized, at least in the Bay Area, San Francisco, and other places where Waymo has been running for a little bit. Nobody even looks at them anymore. It's not even a novelty anymore. People just use them as it has been there for 100 years. Yeah, I went to San Francisco last year and had my first ride in there. And it was quite funny how quickly I accepted it. I got there from day one just taking photos of seeing them in the street and being amazed.

[00:13:15] And then having a little ride in one. And then 20 minutes later, it was just normal, like it had been there 100 years. And I quickly accepted it. So you're so true what you're saying there. And when I was doing that research, another misconception I read that you challenged is that companies must have perfect data before they can successfully even think about deploying AI. And you argue that the real requirement is discipline rather than perfection.

[00:13:41] So can you expand on that and what practical steps leaders listening should be taking first? Yeah. I think this is probably the biggest myth today. Oh, I need to go and clean up my data. And I think this myth has been perpetuated because of previous era of AI where you actually did need to do that. And you did need to clean up data. And even now, if you use like certain types of products you use where you may need to do that.

[00:14:11] We are not in that camp. You don't need to clean any data. You can leave your data where it is, inside the application, wherever it is. The whole idea is you're hiring AI employees. You are hiring AI workforce. Just like when I go and hire humans to do people to do something, I don't say, don't hire. Let me clean my data. In fact, the whole idea of hiring is they'll figure out if there's something wrong in the data.

[00:14:41] And they will raise it to me or to their manager. And they will learn on the job and adapt to it. And that's how at least we have built our AI employees. They will look at the data. They will actually surface if there is any contradiction in the data. They will figure out if there are gaps in the data. For example, there are certain workflows the human wants to execute. Oh, we don't have that information anywhere. They'll point it out.

[00:15:11] In fact, some of our customers have commented one of the largest insurance companies in the country today, top four. And we are doing a lot of claims processing work for them with our AI employees. And they commented, and this is a very established, long-running business, right? So it's, oh, we didn't even know that so many of our processes were not really documented.

[00:15:36] So as part of this exercise, we are also figuring out, okay, all the things that were not documented, we are documenting them. Or your AI is helping figure out and codify what's not documented. Thing is, AI will evolve, and that's how we have built it, and recursively self-improve inside your enterprise.

[00:16:03] So the way our system works, it creates a context graph. It understands your data, your systems, every execution that happens, every workflow run, it learns from that. But with all of that information, it's actually creating a map of your organization. And it's not a static map. It knows it's dynamic and it's evolving. As the organization is changing, processes and systems and products are changing.

[00:16:30] Now, big thing is, it is not a human's mind. It's an AI's mind. And you can ask AI to, like, print it out for you or write it out for you. And it's not lost when specific humans, people, or your employees say quit on you or they're away for something. And you're like, okay, well, this specific thing, Rob knew it. Where is Rob? He's on vacation. What should he do?

[00:16:56] This is the story of every enterprise, even largest enterprises. I'm surprised and shocked sometimes. Oh, I thought you are a well-running business. Yeah, mostly. But underneath, there are a lot of skeletons in the cupboard. And I was also reading how you've written that AI employees should be capable of owning entire workflows, rather than just simply assisting individual workers. And this caught my attention too. So, I'm curious, where have you seen this happen successfully?

[00:17:26] And what kinds of business processes are best suited to that kind of model today? Yeah. So, this is our mantra. We think the real application of AI is when it moves from individual assistants to enterprise-wide automation. And it is, you can think of any process, right? For example, onboarding a new employee.

[00:17:52] It is not an individual assistant kind of thing where, oh, it will help you write an email better or help you summarize some data from somewhere. It's actually coordinating across maybe tens of different systems, provisioning the new employee, setting up learning and development courses, setting up meetings in the first few days, whatever it is, right?

[00:18:20] You can think of finance, same thing, like accounts receivable, accounts payable processes, like procure-to-pay, code-to-cash kind of processes. You can think of sales, like processes, customer support. These are complex processes spanning many, many systems internally, many humans, many teams. And there is orchestration and coordination needed across. And that's the true power of AI workforce.

[00:18:50] That's what we are building. It's really a true AI workforce where multiple AI agents work with each other, coordinate with each other, just like humans would do, to do a multi-step process successfully. And it may be that in some steps of the process, they need to get a human approval or human has to check their work and so forth. You can configure it all.

[00:19:16] And when you do that, this is what I was giving the example of Wipro. We have many, many case studies, Hitachi and some of the largest enterprises, Fortune 2000, that we work with. You get massive ROI. That's the point. Today, a lot of times people are saying, oh, there is no ROI in AI. Of course, if you just use it for, if you just give it to all employees and say token max, you will not get any ROI. Yes, people are learning.

[00:19:45] Maybe a longer term, you'll get some ROI. People are doing a lot of experiments and learning and using AI for their individual personal use. But you are not realizing real ROI. The real ROI is automating actual business processes and reimagining tasks and roles of human teams, not just individuals. Entire teams need to do different things now.

[00:20:12] And as organizations will inevitably continue to introduce AI employees alongside their human employees, how do you think leaders should be thinking about governance, accountability and decision making? And where should AI almost just operate autonomously? And where should human judgment always remain part of the process? Again, massive talking point right now. And I think every organization is trying to get that balance right. But what do you see here? Yeah. Yeah.

[00:20:42] I think that line is not kind of specific to the task. It's more dependent on time, like how much AI should do, how much humans should do. Let me explain. In our head, it's very similar to how we work with humans. When you hire a new person, let's say, we hired a very smart person from top university, has done really well before.

[00:21:10] But you don't on the first day, you don't say, hey, prepare this and go to the CEO and talk to them. Unless, of course, that's that or whatever, right? You give them information. You give them time to ramp up. You train them, right? You also look at their work and verify. And you build some trust. Okay. What you are doing is aligned with how we do in this company. And then you expose that person more and more.

[00:21:39] And then at some point, you take a back seat and say, okay, you can run with it, right? It's very similar with AI, except the cycle can be much faster. You will need to give all the information and access and data, et cetera. You need to help ramp up AI, like show examples of faster execution, let's say customer support or employee support. How has human employees responded to that?

[00:22:07] It's useful for AI to look at that and learn and also learn which were the mistakes that are made and so forth. And then you want to run a lot of testing, like red timming and blue timming and looking at what works, what does not work, when does it fail. When you have sufficient level of confidence, you want to open it up to a small set of employees, customers, or basically users, and slowly expand that set, right?

[00:22:35] At the same time, you want to put clear guardrails on which are sensitive matters. For example, we do a lot of work in HR area, employee experience. Employees may come and ask for anything and everything, sometimes sensitive issues, like a discrimination issue or harassment issue or whatever it is.

[00:22:52] These are, and you can put clear guardrails that, okay, do not try to answer this, immediately inform humans or direct to a QBRI, HRBP, a human to get involved, right? And over time, as you get more and more trust, you can give more autonomy to AI. So this is really, it's a dial of autonomy. It's not zero or one. It's okay.

[00:23:19] Early days, you know, I'm going to check most of your work. So you have 10% autonomy or 5%. Then I'm going to dial it up slowly as I get confidence that AI is working well. And then I still have some clear red lines, like this is where AI needs to stop and not get involved. And human has to be involved. So this is a part of a transformation process.

[00:23:48] As I said, this is not a zero or one thing. You need to work with experts who will help you guide through this transformation process. And I think one of the themes that continues to run throughout your work is that trust will drive adoption more than speed. So whether we're talking about healthcare, financial services, or any other enterprise, how do organizations better build confidence in autonomous AI without slowing innovation to a crawl?

[00:24:18] Again, a bit of a balancing act. Yeah. I think trust gets built by repeated results, right? Consistency and performance, right? If you have repeatedly consistent performance, you can look at self-driving car is a good example. It took them a long time to get to the level of trust.

[00:24:39] And honestly, they have to be a lot better than humans to get humans trust, which is interesting, right? Self-driving cars today are way safer than human cars. But even if they make one mistake, it's on the news, right? And there's pictures and there's a front page news. While so many driving accidents are happening on a daily basis, we rarely see making front page news.

[00:25:06] There's reason for it because people understand that AI, you know, it's kind of a collective intelligence, right? Once AI reaches a stage, every AI will reach that stage very quickly, right? Or if AI is not working, we conclude like most AI will not work. I will give you an example where the human trust is so important to establish.

[00:25:32] Like in Wipro's case, their chief experience officer actually called me and said, Hey, very interesting, 30% of the cases where Emma is still answering well, and we think it is correctly answering, we have verified. But our employees are still going and asking again to a human. And our HR team actually going back to Emma, getting the answer, copy-pasting and sending. But they're very happy to get that.

[00:26:02] We don't understand what's happening. So we did some survey or some focus group, talk to their people. And turns out, people are not really sure. Are you sure this will work? Let me just ask and verify and get a second opinion. Over time, that number is decreasing. And they're saying, okay, it's consistent and it's consistently good. So I can trust. So trust has to be earned. Big thing. Just like for humans.

[00:26:31] And trust can be lost very quickly. So one mistake can lose trust, right? So the bar for AI, unfortunately or fortunately, is much higher. In the same task, you have to be much better than humans for humans to trust you. And if we were to dare to look ahead into our virtual crystal ball here and maybe revisit this conversation in three to five years. And I appreciate when I say that out loud.

[00:27:01] It's like another lifetime in today's money. But if we look into the future, what do you think the best AI native organizations will look like? What will they be doing differently from companies that simply added AI tools to the same old ways of working? How do you see this evolving? Yeah. I mean, I can give an example of what we do in our company. We are completely a native new era organization. A few things, right?

[00:27:27] First, we always think, do we need a person here? Or do we need like how much AI can we use? And what will be the role of the human? So we are always thinking AI employees and human employees together. How they come together in the organization chart. And how does that look like? So every process, every system, every function. That's the first thing.

[00:27:52] Second, we are thinking, do we need the complexity of another application? Today, enterprises have thousands of hundreds, if not hundreds, thousands applications. SaaS applications, software as a service. And that creates silos and isolation of data. And really, humans are acting as glue between these applications. We think the new enterprise will be very different.

[00:28:21] The data and the AI layer. There will be an agentic layer of intelligence and actions. That's how we have built AI employees and accessing the entire data set for the application. Of course, with the right control and permission and so forth. So future companies will be a lot more simpler that way. Their enterprise stack, their software stack will be very simple.

[00:28:47] They will be a lot leaner and a lot faster, a lot more efficient. I think the third thing is today, enterprises spend 70-80% of their time just keeping the lights on. Like there are people spending 70-80% of their time to keep the lights on rather than thinking through, pushing the boundary of technology, innovation, service, how do we improve the service and so on. That will fundamentally change.

[00:29:16] And you will see as human energy and human bandwidth gets freed up more, there'll be a lot more creativity and innovation in every area. And you can think some areas like I always say, like airline travel. It's 50 years and it's pretty much the speed of travel has remained the same. Why should that be?

[00:29:42] And you can find in many other industries, it seems like we have reached a certain area and we are just coasting, we're just holding. We're not thinking what could be the next level of innovation in that product, that service and that industry. And we'll see that happen. I am very optimistic about the new era of creativity and production that we'll get into. It's really a lot of people are talking about this.

[00:30:11] The next industrial revolution, but 10x more. The next, it's an intelligence revolution. And intelligence is when intelligence becomes so kind of cheap, so to say, right? And it's untapped. It's easy to access. And how will you incorporate this machine intelligence across your organization and really think about a completely new way of running the business?

[00:30:39] Well, we covered a lot in a short amount of time today. So for anyone listening that wants to learn more about why companies are moving towards AI native systems and how that is representing a fundamental shift in organizational design, not just another productivity trend. Anyone wanted to find out more about that or the work that you're doing or connect with you or your team? Where can they find out more information about Emma? Just go to Ema.ai, E-M-A dot A-I. Very simple.

[00:31:10] Excellent. Well, I've loved chatting with you today, especially how we shouldn't be focusing on whether AI replaces jobs, but rather how organizations can redesign work around autonomous AI employees and capable of owning entire workflows instead of simply assisting individual workers. And so many big talking points around that. I'll add links to everything for Emma so we can check those out. But more than anything, thank you for starting this today. Really appreciate your time. Thank you so much, Neil.

[00:31:39] Really enjoyed the conversation. So many big takeaways there. And I think the big question I walk away with is, are you using AI to improve isolated tasks or just redesigning how work moves through the business? My guest today suggests larger gains appear when agents coordinate complete processes across systems while the people set goals, review performance, define boundaries and handle sensitive cases.

[00:32:09] And I think that autonomy dial is a useful idea. And I think that autonomy dial is a useful idea to carry forward. Begin with limited authority, test the work, learn from the mistake and increase freedom only as consistent results earn that trust. And this approach recognizes AI's potential and the speed at which confidence can disappear after a bad outcome.

[00:32:33] So remember, you can learn more about Ema and its AI workforce at EMA.ai. But over to you, which business process would you redesign from the ground up first if AI agents could coordinate that work from beginning to end? Lots to think about.