The Exchanges

Q So how big is this industry? And also maybe talk a little bit about how long it's been around too.

A So the industry is valued at three hundred billion today with expectations to grow to over five hundred billion by 2030. And it's just because there's, there's just so much work that needs to get done for large enterprises to be able to function. As well, the industry's actually been around for a long time. Some of the oldest players in the space were started in the 19 forties to help manufacturing companies manage their operations. And today really touches on all the major industries that we think about when we think about like the fortune 500. So It includes retail. It includes travel, telecom, logistics, manufacturing, healthcare, insurance, banks. It's just a huge, huge swath of industries who all in some way, shape, or form rely on BPOs to be able to function.

AI assessment note: “industry is valued at three hundred billion today with expectations to grow”

Q Kimberly, you wrote an article with a pretty fun title, RIP to RPA. So let's jump into that. But first, what is RPA?

A RPA stands for robotic process automation. Um, and it's a way of basically automating very manual tasks within an organization. So things like data entry or invoice processing that basically every business has to do, but it's nobody's core competency. It's just one of the, like, Dirty, messy internal things within an organization that everyone has to do. So historically it's been done very manually. Like you would just hire a data analyst or you would hire a back office operations person. Um, and there was this like, I would say innovation in the last 20 years where people were like, is it possible to automate these tasks? And so the historical way people have done it is through robotic process automation where you basically build like a little software bot that mimics the actual clicks that somebody would be doing. It's very deterministic, meaning, like, they're literally clicking the different, like, boxes that I would be clicking as a human. But, you know, like, organizations are messy, and the work we actually have to do is not perfectly delineated by a very specific, like, process. So oftentimes, if something veers a little bit off course, like maybe someone misspelled a name, or maybe a website changed where the sign-in box physically is on a page, then historically, that would break the RPA process. And as you can imagine, there's, like, An infinite number of small littl…

AI assessment note: “RPA stands for robotic process automation. Um, and it's a way of basically automating”

Q Normally I ask the question, why now? But I feel like, you know, listeners know that AI is coming, it's here. LLMs are maybe the term that a lot of people use. But is there a deeper why now or specific technological advances within the sphere of LLMs that you can point to that actually make this possible?

A Yeah, I think one thing that we're really excited about is, you know, people use the term AI and they're like, oh, everything's going to change now because of AI. But like, what does that mean? You know, there's a lot of very distinct technological breakthroughs that make different applications possible. And specific to intelligent automation, I think one of the things that makes it much more possible than before Is a lot of the fundamental research coming out of the large labs. So for example, recently, Anthropic announced computer use, which is basically a browser agent that is able to intelligently understand what is happening on the browser level of any sort of desktop and be able to take actions accordingly. So, you know, we talked about how historically RPA basically understood at a pixel level, hey, I should click this thing and then I should click that. But with something like computer use, or I think open AI has something called operator that they're gonna release soon. Agents are gonna be able to browse the internet and browse the web in a much more sophisticated way, which is gonna open up a lot of possibilities for what intelligent agents can do before. So we think a lot of these intelligent automation startups, they're not gonna be doing fundamental research on their own. You know, there's still tech that needs to be done to make a browser agent fully work at scale…

AI assessment note: “Anthropic announced computer use, which is basically a browser agent”

Q What can intelligent automation, or what you refer to as these LLMs in action, what can they do that RPA couldn't?

A Let's use the example of, um, a company that we were actually invested in called Tenor. Um, Tenor does referral management. For healthcare practices. So if I'm a primary physician and I need to refer a patient to a specialist, historically, the way that that would be done is I would literally write something out on a piece of paper. I would fax it to the specialist. The specialist front desk person would take the fax, look at it, look at all the information on it, and then input it into my own database, check, you know, like the insurance policies, check prior history, et cetera, and then decide whether to accept the patient or not. And that was a very manual task that there's just a little bit too much complexity in the way that it's done for RPA to be able to handle. So it had to be some sort of administrative person like human who was going to do it. And with now like intelligent automation, um, tenors come up with a very sleek solution that is basically able to automate that whole process. And it's much more self-serve. Yeah. Because the way that RPA would historically work is you would have to hire like an implementation consultant or something, and they would sit Next to whoever was doing the task, and they would basically just watch, like, what are the clicks that you are doing?

AI assessment note: “too much complexity in the way that it's done for RPA to be able to handle”

Q Kimberly, you wrote an article that's been going pretty viral across social called Unbundling the BPO, How AI Will Disrupt Outsource Work. What is BPO? And then also, where did it originate? And also, maybe how does it look today in terms of categories or the breakdown of, of the industry?

A I think a lot of people actually don't know what a BPO is, so you're not alone. BPO stands for Business Process Outsourcing, and it is a large component of work that Really large companies like Accenture or Tata or Wipro, Cognizant, Infosys do. What it means is, sort of as the name implies, if you are a large enough enterprise, there's just a large amount of work that is unsustainable for you to manage in-house. And so you outsource that to one of these businesses to do for you. This includes some of the obvious things that maybe we've interacted with before, like customer support, customer service. It also includes a lot of back office functions that we don't see as much like Outsource IT, HR, finance and accounting for invoice processing and such. Some sort of like knowledge management and outsource research functions as well. So it's really a large catch-all bucket in some ways for work that big enterprises need to do, but for some reason feel like it is more cost-efficient or more scalable to give to someone else to do versus do it in-house.

AI assessment note: “BPO stands for Business Process Outsourcing, and it is a large component of work”

Q Why is it that software to date has not been able to solve that challenge?

A In a lot of ways, I think it's because software historically was very good at doing very clearly defined processes that did not have a lot of variation, didn't have to use tons of different data inputs, did not have to contextually really understand what it was going on and be able to like make In some cases, like judgment decisions and actions off of that. So a lot of times in which you do have to do one of those things, which, you know, in a customer service question, you have to understand what the customer's asking, or if you're processing an invoice, you have to really know what are the different inputs in that invoice. That sort of work software just couldn't handle. This is actually the type of work that AI is really good at handling. It is really good at taking very disparate amounts of information that is often unstructured in different formats across different systems. Synthesizing and structuring it, making sense of all that information, and actually being able to output some sort of action against that. And so what we're really excited about is seeing that this capability is really enabling net new use cases for software that historically just couldn't be handled.

AI assessment note: “software historically was very good at doing very clearly defined processes”

Q think one of the most interesting aspects of this industry is that it's pretty horizontal to your point about, you know, call center technology that's not, you know, specific to the finance sector. It's also in shipping and logistics. It's in healthcare. It's in insurance, right? Where are we seeing, I guess, the most disruption? And then, yeah, where do you see the most potential as well for future disruption?

A I think where we're seeing the most disruption today Is industries that have very high call volume. We're seeing it a lot in logistics in particular, because if you think about how many different nodes are in a supply chain, there's so many people who have to call between the different nodes to be able to just manage communication and collaboration across the, the supply chain. We're seeing a ton of it there. We're starting to see a lot of innovation in healthcare, where either you are a consumer who's calling about some healthcare question, Or it's actually between, let's say, the, the hospital and the insurance provider, or between the insurance provider and somebody else. Anything in which calling is a huge function, um, we're seeing a lot of. And then we're starting to see a lot of early innings in a lot of back office work, where calling is not the primary function, but there is some kind of automation that has to happen on the back end.

AI assessment note: “We're seeing it a lot in logistics in particular”

Q Why is it that software to date has not been able to solve that challenge?

A In a lot of ways, I think it's because software historically was very good at doing very clearly defined processes that did not have a lot of variation, didn't have to use tons of different data inputs, did not have to contextually really understand what it was going on and be able to like make In some cases, like judgment decisions and actions off of that. So a lot of times in which you do have to do one of those things, which, you know, in a customer service question, you have to understand what the customer's asking, or if you're processing an invoice, you have to really know what are the different inputs in that invoice. That sort of work software just couldn't handle. This is actually the type of work that AI is really good at handling. It is really good at taking very disparate amounts of information that is often unstructured in different formats across different systems. Synthesizing and structuring it, making sense of all that information, and actually being able to output some sort of action against that. And so what we're really excited about is seeing that this capability is really enabling net new use cases for software that historically just couldn't be handled.

AI assessment note: “software historically was very good at doing very clearly defined processes that did not have a lot of variation”

Q think one of the most interesting aspects of this industry is that it's pretty horizontal to your point about, you know, call center technology that's not, you know, specific to the finance sector. It's also in shipping and logistics. It's in healthcare. It's in insurance, right? Where are we seeing, I guess, the most disruption? And then, yeah, where do you see the most potential as well for future disruption?

A I think where we're seeing the most disruption today Is industries that have very high call volume. We're seeing it a lot in logistics in particular, because if you think about how many different nodes are in a supply chain, there's so many people who have to call between the different nodes to be able to just manage communication and collaboration across the, the supply chain. We're seeing a ton of it there. We're starting to see a lot of innovation in healthcare, where either you are a consumer who's calling about some healthcare question, Or it's actually between, let's say, the, the hospital and the insurance provider, or between the insurance provider and somebody else. Anything in which calling is a huge function, um, we're seeing a lot of. And then we're starting to see a lot of early innings in a lot of back office work, where calling is not the primary function, but there is some kind of automation that has to happen on the back end.

AI assessment note: “where we're seeing the most disruption today Is industries that have very high call volume.”

Q but now that we're here, how do you see the next five, 10 years evolving? Because there is kind of like a shift that people have to do intellectually as well as they're thinking about their software budget to labor budget, and as they almost have to re-gear their brain to say, oh, we actually can do this automation, which we previously couldn't. So how do you see that trajectory?

A I definitely think it's gonna be an evolution, and I think it'll depend, you know, on the technology spectrum, like how technology, uh, savvy or how at the forefront that industry is, but for a lot of these older industries that we're talking about, like the larger ones that are a little bit more on-prem, a little bit more based, um, like in the physical world, I think it, it, it will take, take time, which is why I think doing the vertical end-to-end automation solution is so exciting, because You can actually build something that is very tailored for their specific workflow, where it's almost a no-brainer to use it. Like everybody, nobody wants to do data entry. Like nobody wants to sit in the back and read a hundred faxes and try to input that into a system. And that's no, no company's core competency either. So if you're able to build an intelligent AI agent specifically for that industry that is tailored to exactly how they do their business, It's almost a no brainer to do it. And then the folks who were doing that before can now focus on much higher value, either customer facing tasks or much more complex tasks. And then over time, let's say in the next five to 10 years, you know, the technology wave will continue to get adopted by more and more companies, people that will become more knowledgeable about what these agents can and cannot do, more comfortable with the techn…

AI assessment note: “over time, let's say in the next five to 10 years”

Q Totally. And so Obviously, we're kind of early in this arc, as you mentioned, but there's a lot of interesting, exciting things to come. What would you like to see builders focus on? What kind of builders would you like to hear from as well?

A I would be really excited about people who are thinking about, ah, what was not possible before. I mean, like, you know, we've talked a lot about, like, what RPA does today, and the types of customers it's able to target today. When you think about, like, the world of work that could be intelligently automated away, and the amount of time and savings Both employees and companies can get. It's just like an order of magnitude larger than what, what is currently possible. And so I'd be really excited about people who are thinking about the bucket of types of tasks that were automatable that RPA historically could not handle and types of industries that it currently was not able to tackle and really thinking about like, what are those first flows or first automations within those industries that are possible? And really thinking about what are the clean UI or UX paradigms that you could bring to bear for those solutions.

AI assessment note: “I would be really excited about people who are thinking about, ah, what was not possible before.”

Q And in your piece, you kind of talk about this difference between front office and back office, and with new founders coming into the space, how would you advise them on attacking this opportunity and thinking about maybe the difference between those things, or is there some other way that they should be thinking about attacking this market?

A These BPOs are very large businesses, and they understand the opportunity of AI the way, you know, lots of people who are paying attention to the news understand the opportunity of AI. So, They should not assume that these BPOs will not try to leverage it themselves. We do think in the short term, there's actually still a really exciting opportunity. These BPOs, the, the business model they have is fundamentally about labor, and it's fundamentally about having humans execute on a lot of these tasks. And it's quite a big shift for any business, but especially large public businesses with tens of billions of dollars of revenue on the line to be able to shift that work. Into product immediately. The second thing that I think a lot of people underestimate or don't quite realize is just how difficult it is still is to work with these AI systems. There's a lot of work that needs to be done to make sure, you know, hallucinations don't happen, to be able to actually evaluate the responses of the AI agents, to know as the models get better, which model to swap in and swap out. I think you have to be a really AI native technical founder to be able to understand how to leverage that, and that's actually just not a widely distributed skill set yet. So we think that the best types of opportunities for people in that domain is Just really thinking about situations in which the ROI is so incr…

AI assessment note: “best types of opportunities for people in that domain is Just really thinking about situations”

Q What does this enable in terms of the kind of new business that can be done, right? Can, for example, instead of only working with large companies, does this enable smaller companies to leverage some of these resources? Or is there something else at play here when you think about the longer term of again, Not just replacing the old, but actually starting something new.

A I think the advent of AI solutions, which are much cheaper, much more scalable, et cetera, is that you can not only maybe offer this type of work to a new subset of the population that BPOs never handled, but even for companies that use BPOs, you can now expand the surface area that that type of work covered. For example, going back to customer service with Yeah, you can offer it across the entire gamut of your product surface area. And that opens up just a ton of net new areas that BPOs did not cover historically. The new industries or new types of companies question. Yeah, I think there are a lot of companies that maybe would have wanted to outsource their invoice processing and not do it in house or who did want to offer support but couldn't. And then now you'll see that with AI agents, They'll actually be able to make their own internal operations much more efficient. Today, maybe the BPO's, the large BPO's won't see directly impact their business because this wasn't the core function.

AI assessment note: “there are a lot of companies that maybe would have wanted to outsource their invoice processing”

Q Are there any parting thoughts that you have in terms of where this goes, or also maybe gaps that you see where, you know, founders haven't quite found their mark in that industry yet?

A So I think one common question that I've gotten about this is distribution of what these BPOs actually do is not super clear from the outside. Um, like if you, you read the reports, it's actually quite opaque. And I think a lot of what these businesses do is not just a lot of this outsource work that we were talking about, but it's also this like amorphous, I would call it like outsourced IT or outsourced application development where these businesses Also might build these small internal tools or just like small applications for companies that don't have the IT resources or engineering resources in house to do. And one thing that we haven't really talked about is I definitely think this is on the earlier curve of things that software can handle because, you know, building a full application is quite different than responding to a customer service inquiry. But we're seeing a lot of at a more horizontal level, like Coding agents just get a lot better and be able to empower people who maybe weren't as technical or maybe who weren't technical at all, be able to build full formed applications. And so I think that's actually going to be a very interesting orthogonal attack vector against a lot of the type of work that did get outsourced, even when the way in which you would do that is not just I'm going directly after this BPO spend, but it's just I'm enabling every individual perso…

AI assessment note: “I think that's actually going to be a very interesting orthogonal attack vector”

Q And in your piece, you kind of talk about this difference between front office and back office, and with new founders coming into the space, how would you advise them on attacking this opportunity and thinking about maybe the difference between those things, or is there some other way that they should be thinking about attacking this market?

A These BPOs are very large businesses, and they understand the opportunity of AI the way, you know, lots of people who are paying attention to the news understand the opportunity of AI. So, They should not assume that these BPOs will not try to leverage it themselves. We do think in the short term, there's actually still a really exciting opportunity. These BPOs, the, the business model they have is fundamentally about labor, and it's fundamentally about having humans execute on a lot of these tasks. And it's quite a big shift for any business, but especially large public businesses with tens of billions of dollars of revenue on the line to be able to shift that work. Into product immediately. The second thing that I think a lot of people underestimate or don't quite realize is just how difficult it is still is to work with these AI systems. There's a lot of work that needs to be done to make sure, you know, hallucinations don't happen, to be able to actually evaluate the responses of the AI agents, to know as the models get better, which model to swap in and swap out. I think you have to be a really AI native technical founder to be able to understand how to leverage that, and that's actually just not a widely distributed skill set yet. So we think that the best types of opportunities for people in that domain is Just really thinking about situations in which the ROI is so incr…

AI assessment note: “the best types of opportunities for people in that domain is Just really thinking about”

Q I mean, one natural question that comes up, I think, for many people, especially as they think about things like hallucinations, is where is the technology in this arc? Are we able to really achieve this idea of intelligent automation today? Are there barriers? Like, where do we sit in that trajectory?

A The way that we've seen it work best is when there's one very specific automation flow, at least to start, that a company can just nail. Meaning, um, it's often industry specific. Um, so you can integrate into all the core systems there. You can understand the context for that industry, and it's one very repeated, um, but very manual flow. So for example, like data entry, it's I get on a phone call. I hear the update on where an order is. All the information from that order, it can be parsed through that call and input it into my main system. That probably happens like thousands of times a day for the largest organizations, um, all manually done. And that is one very specific flow. And that's just to start. And then once you get there, you can build deeper into other flows. But I think that is a much more successful path where you can actually understand the constraints and build around them, make sure that, like, the agent performs correctly versus tackling, let's say, like, everything within healthcare, everything within legal and logistics to start.

AI assessment note: “The way that we've seen it work best is when there's one very specific automation flow”

Q What can intelligent automation, or what you refer to as these LLMs in action, what can they do that RPA couldn't?

A Let's use the example of, um, a company that we were actually invested in called Tenor. Um, Tenor does referral management. For healthcare practices. So if I'm a primary physician and I need to refer a patient to a specialist, historically, the way that that would be done is I would literally write something out on a piece of paper. I would fax it to the specialist. The specialist front desk person would take the fax, look at it, look at all the information on it, and then input it into my own database, check, you know, like the insurance policies, check prior history, et cetera, and then decide whether to accept the patient or not. And that was a very manual task that there's just a little bit too much complexity in the way that it's done for RPA to be able to handle. So it had to be some sort of administrative person like human who was going to do it. And with now like intelligent automation, um, tenors come up with a very sleek solution that is basically able to automate that whole process. And it's much more self-serve. Yeah. Because the way that RPA would historically work is you would have to hire like an implementation consultant or something, and they would sit Next to whoever was doing the task, and they would basically just watch, like, what are the clicks that you are doing?

AI assessment note: “too much complexity in the way that it's done for RPA to be able to handle”

Q I mean, one natural question that comes up, I think, for many people, especially as they think about things like hallucinations, is where is the technology in this arc? Are we able to really achieve this idea of intelligent automation today? Are there barriers? Like, where do we sit in that trajectory?

A The way that we've seen it work best is when there's one very specific automation flow, at least to start, that a company can just nail. Meaning, um, it's often industry specific. Um, so you can integrate into all the core systems there. You can understand the context for that industry, and it's one very repeated, um, but very manual flow. So for example, like data entry, it's I get on a phone call. I hear the update on where an order is. All the information from that order, it can be parsed through that call and input it into my main system. That probably happens like thousands of times a day for the largest organizations, um, all manually done. And that is one very specific flow. And that's just to start. And then once you get there, you can build deeper into other flows. But I think that is a much more successful path where you can actually understand the constraints and build around them, make sure that, like, the agent performs correctly versus tackling, let's say, like, everything within healthcare, everything within legal and logistics to start.

AI assessment note: “The way that we've seen it work best is when there's one very specific automation flow”