Dave Cowing

In a world that’s flush with 1000s of companies chasing AI use cases, I think that’s a very interesting approach. And to me, as you said, an approach that resonates, feels more natural to me than sort of these point insertions and use cases all over the place.

Amaresh Tripathy

Absolutely. There are different challenges around that, but some of the technology stack is the same, regardless of what LLM I go to use, like OpenAI or Llama 2 or what options will you have? How will you fine tune it? Will you do kind of retrieval augmentation of the models? I mean, they all are technology questions we’re trying to answer. And we have a platform to do that, but more importantly, what is the decision flow, what is the decision that a person makes throughout the day, right? What are the activities? And most knowledge work activities are about, hey, I want to sequence the work that I want to do in the right way. I want to actually do the work, which is a lot of it is finding information or summarizing information, or I need to kind of fill out the paperwork and update systems. I mean, that’s kind of a lot of the work that all of us do. And so how do you actually kind of make that easier so that a nurse can spend more time with the patients? An actuary can actually start thinking about underwriting rather than reading documents. A salesperson can focus on building relationships with his super part versus thinking about updating SFDC or who to contact. That’s essentially kind of what we’re trying to enable.

Dave Cowing

I want to come back to the personas and a copilot idea, but you took us down a path I’d like to dig into just a little bit more. You mentioned Open AI and Llama 2. I’d love to understand your perspective, especially in kind of light of what happened last week with this apparent implosion and resurrection of Open AI over the course of three days. What’s your perspective what happened, how you guys use them, the place for the various LLMs and technologies, as well as where you fit in the stack and how you guys are navigating that?

Amaresh Tripathy

So, my perspective of what happened, and then I’ll get to how we’re using it. In a broader scheme of things, it was nuts. There’s one big takeaway. And obviously, we can talk about corporate governance, we can talk about leadership, we can talk about all kinds of things, but I think there are better people to do that. Big takeaway for me personally, is if you go and see the root of what actually was happening, it’s a debate between how fast we should go with commercializing some of the technologies, things that we understand kind of, but not completely. Which tells me GPT 5 is going to blow our minds, even more than GPT 4 does. You can argue what is good, what is bad, but basically, the next release is going to be even more shocking. That’s the big takeaway for us. I think it is going to widen the gap between the technical promise and the organization reality.

If you really think about tools like AuxoAI, we actually are in that gap. We are trying to understand the technical realities and the technical potential of things that are coming in, whether it’s through Open AI, whether it’s through a lot of other open-source great tools like Llama 2 or Falcon. There are some amazing large language models, but also the entire stack that is developing. You think around prompting, there’s a whole term that has just come up literally a year ago, that didn’t exist before. There are tools like Lang Chain, which is an orchestration tool, those things didn’t exist a year ago. And all those kinds of tools, and how do you actually make sense of it and put it in some sort of a coherent way and a platform. Eventually, I think it will become more and more standardized, but right now it is not. To take the technology promise and make it an enterprise reality, you need to stitch it all together, which what we do, which is to build that platform. But then how do you take platform take that platform and make it real in context of individuals and people who are making their jobs better. That’s the whole copilot and the whole AI companion story. Those are the two things we do. We have a platform and then we deliver copilots for individuals.

Within that platform, we are fairly certain everything will change over a period of time. So how can you create it modular enough that, if it is not Open AI or suddenly Falcon is a lot better, how do I switch it very, very easily, because your cost curves also change based on volume. There are a lot of lot of things that you need to start thinking about from an economic value that an enterprise cares about and translate that into technology. So that translation layer, we essentially try to make it easier for enterprises through the platform and then, obviously, drive value through the capabilities that the platform can provide.

Having said that Open AI is an amazing company and it’s probably the best models right now. I mean, it’s mostly GPT 4, it has some very, very good models, though there’s some tradeoffs worth thinking about. Having said that, six months from now will it remain the same? We almost certainly know it won’t remain the same. Our job is to make sure our enterprise clients are focused on the value to be delivered rather than the technical minutiae, which we can insulate them from.

Dave Cowing

You made an important point that maybe the folks not as close to it don’t get, which is in all of the models today, they have usage based commercial terms. So managing the cost to actually execute something when it ties directly to what your users are doing is an interesting challenge.

Amaresh Tripathy

Yeah. The API-based ones. If you use open-source ones, you have to put in the upfront investment of GPUs, which are very hard to get, and they are expensive right now. And so there are trade-offs that you need to just balance and work through it, depending on use case.

Dave Cowing

If we turn back to the persona as you talked about, the nurse, the underwriter, the salesperson. You’ve got this near-term vision. I’m curious to see kind of how far this extrapolates out in the future or if you have kind of a bigger future vision of how AI works with people, particularly in the enterprise in the future.

Amaresh Tripathy

From what we’re seeing in the work and the kind of the client conversations that we’re having, two dominant themes are emerging. For high leverage workers, you will need companions because there are fewer in numbers, and they make a huge difference. A pricing analyst can make a massive difference on the top and bottom line of any organization, or a salesperson can, or a nurse can in a hospital or an actuary in an insurance company. I mean, those are very, very high leverage roles. And those are hard roles, very complex roles. By and large, those roles don’t have enough tooling available to assist them apart from BI dashboards and things like that. In some ways, we are serving our highest leverage workers with not enough tools, and we try to automate out the most commoditized people. The idea is, can you change the balance and you provide the right tooling in the form of AI companions so that they can actually amplify the work even better. That’s bucket number one and I think we’ll see that. We will also see a whole new flavor of hyper automation. Which will be a productivity play. I think both those things will work simultaneously. It’s evolving in that direction slowly. There is a third thing that is actually happening more and more and is actually led by I think the Innovation Officer, the CIO, and CTOs. Which is I think, eventually, these language models and degenerative eye tools become a brain of the organization, they become the operating system on which the organization works. Because if you really think about it, the methodologies, the knowledge, all the knowledge graph there is of how the work actually gets done in an organization, it’s in people’s brain right now. That’s why a one-month customer care agent versus a five-month customer care agent, you can see a huge difference. The reason is that there’s a lot of implicit knowledge that is there. I think where organizations will eventually end up, obviously my point of view, is there will be an organizational brain that will get developed where a lot of the implicit knowledge will get explicit, as people use these tools everywhere, and they will be essentially an organization brain. That is going to drive a different kind of productivity, because we just haven’t had the intelligence of the center and at the edges till today. So the future of work, I think, it’ll take some time, it’s not tomorrow, but a lot of the piece parts that we are all starting to put together, whether it’s like in copilot, Microsoft is launching their own thing connected to Office 365, or how Salesforce will do their thing. There will be these tooling available everywhere. But essentially, what all of this tooling will do is create a digital knowledge graph of the organization. Once you start leveraging that you change everything in an organization, I think it’ll change org charts, it will change a bunch of things. We are probably a 10-year journey towards that. But these are the foundations, and then copilots and hyper automation, all of these will play a big role.

Dave Cowing

The notion of the brain or the enterprise OS is really interesting. Not just how you bring somebody up to speed, but how you then can leverage the collective intelligence, the collective near real-time intelligence of an organization. What are the trends in my client base and how does that apply across segments, across roles? How can I then use that to talk to my client in a sales role about the next thing? It’s an exponential, sort of factor of improvement, alertness, etc.

Amaresh Tripathy

Exactly, exactly. And the tooling is not there. I mean all of these things are developing right now. We have piece parts of it. Like we have the pen, and we have the nib, but not the whole pen. We don’t have ink sometimes. We’re trying put it all together so all of this of flows in the way we want it to. And it’ll take some time to get there. But it’s undoubtedly, that’s where it’s going.

Dave Cowing

I’m sure one of the biggest culprits is disparate, you know, data sources kind of all across the enterprise, you’re trying to try to start bringing them together and get access to them in a meaningful way.

Amaresh Tripathy

Absolutely, that’s going to be a big part of it, or keep it distributed where they are, but being able to go and find information from them when you require them. There are architectures that will evolve on that front also.

Dave Cowing

From that discussion, I think it’s very clear on how it provides value for the enterprise. I’m curious, you get down to the individual level, that nurse, that underwriter, how do you see value accruing to them as an individual, how it impacts their job and what they do?

Amaresh Tripathy

I mean, the salespeople we work with they get more commissions. You go after the easier ones. The nurses, we do work in prior authorization. So you go to a doctor, and you just say you need an MRI. And suddenly, the doctor will say, “Hey, listen, I need to go and check with your insurance company, whether this is covered and what I need to submit.” To cover that actually takes a PA or a nurse a fair amount of time to put the justification together, send it and everything. So, every time they are doing all of that work, they are not actually seeing a patient, or they are not building a relationship. We have a prior authorization copilot, which basically simplifies that entire process for the nurse. So, they are able to go do these prior authorizations faster and much more accurately. It drives more revenue and makes it easier for the nurse. We already talked about salespeople. All these underwriters who are reading contracts versus actually doing the underwriting and understanding what the business context is. So, for high leverage workers, the benefit is they get to practice at the top of the license. And everyone, the whole Maslow theory, we want to be practicing at the top of your license and not doing stuff that you don’t want to be doing, but it’s kind of part of the process. That’s very clear. When we go to the hyper automation side of things, there are a lot of processing people who will become investigators, because I think there will be a lot more exception-based workflows that will happen, there will be a lot more straight processing. But that problem has been around for some time, around how do we upskill this thing. At an individual level, depending on where you are with how critical you are and how much leverage you have in the organization, it’s either amazing news in my mind, because from a skill set perspective or practicing on the top of the license to the problems that you always had around upskilling into the next wave of things, which will just get accelerated. So depending on where you are, I think it will impact differently.

Dave Cowing

Reading between some of those, it sounds like when you think about the enterprise, adopting AI, and really demonstrating to the rest of the enterprises, real business cases, focusing on those high leverage roles that have clear cut and straight forward benefits that accrue from them is the way for those organizations to go.

Amaresh Tripathy

Yeah, which is the case. This is where we have to start focusing because it’s a win win win for everyone and there’s value and I think those tend to be normally in the front office, so there’s all kinds of things going on the design up there.

Dave Cowing

What do you see is the sort of the biggest challenge that you’ve seen enterprises have when they’re looking at getting into AI and starting to implement it in their business?

Amaresh Tripathy

Just two things. One, there is a lot of confusion and noise. And I can see that. We do this for a living right now and we literally are 24/7 thinking about this. And it’s hard to keep up ,for us. So someone whose job is to run a sales team or job is to run a pricing team, or even if you are the CIO, you have your subsystems down, or something like that. And you have to deal with this. It’s just impossible. That’s number one. There are so many vendors, so many technology companies, everyone wants to have a point of view. And that sounds very confusing. But if you’re on the other side, if you’re trying to figure it out; what’s there what’s real, what’s not? That’s one I think.

Second, it’s an early stage technology. Which early stage technology that’s moving at an exponential speed. There’s only two times to get in, either too early or too late here. So that’s the second. And the third is I have the org design that is being set up. I mean, the technology, think about it, Open AI released GPT 3.5, which is kind of starting this whole era, last November. Just around a year ago? The budgets for ‘23 have already been frozen. And we are cutting and everything right there. And suddenly, this thing becomes a big deal. People can’t say “where’s the innovation funds?” So there’s this whole how organizations work and how do I prove value, do all of that stuff, that itself is a barrier.

So, confusion and noise, early stage technology and org structures and the budgeting cycles that we’re dealing with, combine all three together and it’s like, okay, come on, you want me to do things, where do I start and what do I do?

Dave Cowing

Yeah, the technology outpacing the budget cycle is a particularly interesting one. Some firms well figure it out, other firms won’t. And you could start seeing bigger gaps between competitors.

Amaresh Tripathy

You can start seeing that. I mean, some people are like, “OK, let’s kind of make it there.” Or some people are hiding it under, data transformation or other programs. Let’s kind of do this a little. You have to be creative in this thing. Normally you would have you send these things to your head of innovation, and say, “OK, we’ll talk to you three years later and see what you find out.” Here, three months later, someone is releasing something. And you’re like, OK, I’m already behind. So, there’s a FOMO and a fear factor kind of going on. These are all natural things and things will probably generally settle down and we’ll probably find a rhythm. But that’s where we are, at least for the last 12 months and maybe the next 12.

Dave Cowing

One of the things that I’ve seen in talking to heads of innovation, it seems like they’re almost exclusively focusing on AI right now. And excluding almost everything else.

Amaresh Tripathy

If I am the head of innovation, there’s no other technology, which is going to become as real, as quickly that I don’t have to sell to the organization. Anything on my portfolio. So of course he is.

Dave Cowing

It impacts your agenda pretty quickly. Just pivoting a little bit. If we think about financial services for a moment, big complex, IT driven industry, lots and lots of data, but still large patches of it have significant amounts of unstructured data. Especially in the lending space, the credit space. There certainly other spaces like the front office, alpha generation, financial advice. What do you see is the way your solution can help solve big problems in the financial services space.

Amaresh Tripathy

I think you laid it out. This broader notion that people will develop some version of the brain or the platform. That’s kind of one which we can accelerate that journey, or we can just get you started on that journey much faster. That’ll be all these use cases around, unstructured data. If you really think about it, whilst financial services have been very ahead of managing structured data and they’ve just generally been ahead of data analytics, unstructured data is completely unsolved. A big chunk of all workflow and processes, because of regulation and other things, has a lot of unstructured data. That’s one big area. We just hired someone from one of the largest banks who was building a non-structured data gen AI platform for them. That’s going to be a big, big area to reduce cycle times, whether it’s in lending, whether it’s in structured products, anywhere where end-to-end automation hasn’t happened.

Second, on the Alpha Generation piece of it, which is what we do, like the sales copilots that we have done. A lot of the sales effort is understanding the same unstructured data from a client-based perspective. One of the examples we have is, we have a company that essentially sells whatever there is an ESG agenda for the client and something gets triggered. They are in the waste management business. So, whether there is an EPA violation that actually happens or in the 10K, they go and start committing to a carbon net neutral thing or a wastewater effort that they start somewhere. So, we have essentially built an ESG GPT. Which is always scouring their potential client base and you can think of the same thing from investment perspective, the same idea. And you are looking for that. Basically, on the right timing, on the right triggers, they are able to go on to make an outreach. What you’re trying to do is unstructured data combing to get demand signals or investment signals. Which will kind of create alpha.

And the third one, we’ve always talked about personalization, one-to-one personalization. Whether it’s in financial advice, like kind of your wealth manager, we talked about all these robo advisors. Now think about robo advisors, which can be like so super like one-on-one kind of a conversation that you could have and you could create all these portfolios at the back, which are tax loss harvested and everything at the back. A lot of the innovation is already happening. Now you can connect it to the one-to-one advisor on front, which is a great use case. I’m sure we are going to see, if already not there, some version of advisor GPT either supporting advisors, right now, to raise the quality of all the advisors to the next level or directly to the consumer. I think we’ll see both versions fairly soon. Or any marketing customization, like offers, whether it’s current offers, all offer customizations. There was a great example, even in Open AI dev day, you saw Coca-Cola had a Diwali greeting which is auto generated for India versus different festivals for a different place. So the way you are able to go and do these auto-generations and move to one to one targeting, that’s what would be there. These are some of the obvious use cases. And if you go down to the more nitty gritty areas where there’s a lot of value in the very small things.

Dave Cowing

A lot of those really play well into the high leverage users and personas and certainly financial advice, alpha, unstructured data, being able to accelerate cycles, I’m interested as well maybe do you see what sort of opportunities see if any, and more the operations side, so client onboarding, investment operation, trade clearing, settlement, that sort of thing.

Amaresh Tripathy

Any process related stuff. Contact centers is going to be one big one that you see. I mean, that’s it’s obviously going to see all of it, in terms of suddenly opening up omni channel contact centers. And the thing is that some of the theory we’ve been talking about for some time that technology is kind of making it easier and better. IVR’s will go away. It’ll be a different time of IVR interfaces that I think we’ll see. But you’re right on the onboarding kind of use cases, KYC, think about all those kinds of operational use cases, where a big piece of it is actually a lot of financial setup processes or data flows. And a lot of it is unstructured data flows, actually. You can just go put two and two together, and we’ll see all kinds of all kinds of workflow platforms. Now we’ll hook into these task level Gen AI copilots all over the place to make things faster, better, easier.

Dave Cowing

A last question, I’ll put a constraint on this question, just to make an interesting, if you were to give an enterprise sort of one bit of advice, if you’re thinking about Gen AI in their enterprise, what would it be?

Amaresh Tripathy

Start working on it, on use cases. Basically, experiment. I mean, production level experiments – I’ve never seen so many production level experiments going on, and at the same time. It’s intuitive that it will make it will make things better, we don’t exactly know how and we also know there are some organizations structure that might need to change to accommodate that. Some of the technology might not perfectly work. And there are ethics and risks things that are associated, so choose use cases that are more internal facing, put human in the loop. Make it easier for humans to do certain jobs which matter. And test it and learn from it because the tech stack will evolve, everything on all sides will evolve, people, process, tech all sides sites will evolve. But as I said, if you don’t do it, someone else is. The analogy I have is I don’t think organizations will feel it like a typhoon or a big storm, they will almost feel it like a termite. Things will just kind of go on and naw and naw and suddenly, it’ll collapse. I like the termite analogy a lot more, though it feels like there’s a lot of changes going on, I think a lot of small, small changes will add up and things will change. So, I think one piece of advice is experiment and test from wherever whatever vantage point you are starting with bigger or smaller. Because I’ve talked to $5 million $10 million revenue companies and have talked to Fortune 10 companies, and they all are kind of in the same place. Once you get started working in it you understand a lot and you will have a much better appreciation for what to drive and how to drive the change. The people side of it is going to be the most important thing eventually, technology will get figured out and if you don’t, if you’re not in the mix there it will become brittle, because you cannot short that time.