In this episode, host Stéphane Zanoni sits down with Marc Low, Director of Innovation, Growth, and Emerging Tech at KPMG, to explore the rapidly evolving world of AI in the enterprise. From the overnight transition to conversational AI to the challenges of implementation and adoption, Marc shares his expertise on how businesses can leverage AI to boost productivity and stay competitive. The discussion covers a wide range of topics, including building trust in AI, addressing bias, and the future of AI agents in business processes.
Subscribe and listen to the Rev-Tech Revolution podcast series on:
 |
 |
 |
| Spotify | Apple Podcasts | iHeart Radio |
Prefer reading over listening? We’ve got you covered!
Stéphane Zanoni: Hello, listeners. I’m Stéphane Zanoni. And today, I’m pleased to have Marc Low, the director of innovation, growth, and emerging tech at KPMG joining me on the podcast. Marc and I first met at the upperbound AI conference, and then again had a chance to meet when we spoke at our annual Riva company event. And since then, it seems like the innovation around AI, Gen AI, and AI agent has continued to skyrocket.
Stéphane Zanoni: I really look forward to today’s conversation. Thank you. Before we get started, Marc, why don’t you share a little bit about your career path and how you ended up in your current role?
Marc Low: Sure. Well, thanks so much for having me, first of all. I’m great to be here. Looking forward to the conversation. I spent the last decade and a bit in this type of space.
Marc Low: I was fortunate enough to start my career in technology. So working for technology companies here in Vancouver and really learning the game as technology shifted from traditional on prem architectures to cloud and the evolution of software as a service and all these type of emerging models, and then did a transition from there into professional services and to consulting. Worked for another one of the big four for a long period of time, and ended up running the innovation group, uh, out of there. And then moved back to, uh, moved back to Canada where I ultimately met the managing partner here at KPMG. And I’ve been building this innovation group inside KPMG for the last 3 years, um, which has been an amazing adventure and gives me the opportunity to touch and experiment all the stuff that’s on the bleeding edge, uh, and really work with clients from a place of empathy, helping them understand this fantastic landscape of solutions, try to articulate how fast the world is moving and evolving, and really helping them understand how they should position their organizations in order to at least ride the wave and certainly to avoid being disrupted themselves.
Marc Low: And so, um, yeah, the work is challenging. Every day is different, really varied, which is fantastic. You can’t ask for anything more. Um, and gives me the opportunity also to work with clients across all kinds of industries from public to private sector. Uh, yeah.
Marc Low: So, the work is challenging, rewarding, exciting, um, and that and gives us the opportunity to really be on the bleeding edge of everything that’s that’s happening.
Stéphane Zanoni: And you talked a little bit about where you were before. And before we talk about the present a little bit, uh, maybe just talk to us a little bit how you’ve seen how you’ve seen AI transition from maybe more of that machine learning predictive model for anomaly detection to really what we see today where it’s much more conversational. What was that transition like in the
Marc Low: Uh, almost overnight. Insofar as it became topical, like, just from almost from one moment to the next. You know, historically so I’m coming to you this afternoon from our Ignition Center, which is our innovation space space at DeepMind. Um, and one of the groups that is housed inside the Ignition Center is our Lighthouse function, which is our data and analytics group.
Marc Low: And these are all PhD level data scientists, statisticians, right, software engineers. And that really was the domain of AI for the longest time. Um, and probably much to their regret now, it’s been democratized for everybody’s access. Right? So everyone thinks that they’re a data scientist and an AI professional now.
Marc Low: Right? But it really was that transition from highly technical, uh, solutions really that sat in the background. Like, one of the analogies that or use when I’m speaking to clients, right, is your cell phone is packed with AI solutions, almost entirely which have been developed by very, very smart people. And these solutions sit in the background. They just they help inform how you interact with technology.
Marc Low: And, you know, that fall of 2022, right, when Chat GPT became the kind of had achieved Kleenex status, right? It became a brand that all of a sudden everyone was talking about. It really was that thing. And all of a sudden, we literally I remember having the conversation with my managing partner saying, This feels like a thing that’s happening. And literally, we went into Christmas break and by January, every CEO, every board of directors wanted to talk about it suddenly, right?
Marc Low: And it’s not like the topic broadly was new, it’s just that because it had morphed into this conversational interface where you’re packing all these traditional paradigms into something that anybody can interact with, all of a sudden that unlocked 100 and 100 and 100 of use cases, and every organization suddenly sprinting to figure out what that meant for them. So, if you had worked inside the space before that, it probably felt like this was a multi decade kind of evolving story. But I think for most non technical folks, it was kind of like an override, flip the switch, and all of a sudden everybody needs to be thinking about it, which is challenging, right, in terms of helping for organizations trying to understand, like, What do I do with this thing all of a sudden? And trying to separate sort of hype, like signal from noise, right? You know, hype from reality.
Marc Low: But then also very exciting, right, because it opens up a whole new landscape of opportunities. So it was when we talk about paradigm shifts, it sometimes can feel a bit hyperbolic. This really did feel like a paradigm shift that you were experiencing in real time, which was really exciting.
Stéphane Zanoni: And that timeline you described of leaving for the Christmas holidays then coming back and everything being different. I know with with the team, we basically took the entire product roadmap for the next 2 to 3 years.
Marc Low: Bundle it.
Stéphane Zanoni: Just throw it out and not to reimagine it.
Marc Low: Yeah. There are it’s and it hasn’t stopped. Right? Um, you know, I was listening to a podcast yesterday, and they were talking about workflow automations. And I won’t mention the particular organization, but, there was a we had built some low code applications on it.
Marc Low: And it’s a workflow orchestration piece of software, right? And that was cutting edge 3 years ago, and all of a sudden, all that’s been usurped you know, by these platforms that are now, you know, conversational. They’re helping you stitch workflows together in agentic workflows and so on. And so, organizations are, you know, literally rewriting their entire, uh, you know, um, kind of code base on the fly. Um, very large CRM based organization.
Marc Low: Again, I won’t mention this particular brand, but had a big conference, their famous annual conference this past week, right? And they had to do a full pivot, right? Their entire architecture, the entire kind of reason for being for that platform rewritten overnight. So every organization’s struggling with it and wrestling with it and trying to figure out what it means. So, again, I’d say in empathy to your team at Riva, you’re certainly not alone.
Stéphane Zanoni: Well, it’s a new I’m really excited about the shift in maybe the traditional user interface of maybe having a user interface with a dozen predefined rules with checkboxes to then just having a prompt that has guardrails. And now the end user can really ascribe the some of the behaviors and some of the logic flow that they want to see the system take. Uh, are there real life examples of organizations or of technology where we really see that applying in a tangible way in the enterprise? My answer is it
Marc Low: will be in in a couple of parts. So one is, um, uh, the good news is that, uh, in the so in our experience, right, the work that we’ve been doing over the last sort of year and a half, specifically around generative AI, we’ve documented 900 and counting use cases for generative AI inside the organization. In consulting speak, we would talk about front office, middle office, back office, meaning front office, stuff that touches your customer, back office, typically things like finance, your middle office things like HR and legal and others. Across that spectrum of functions inside the organization, there are 100 and 100 and 100 of use cases. That’s kind of point number 1 to the answer.
Marc Low: The second point is that, uh, there’s a great tweet that I came across the other day and the tweets I won’t claim it, the delay, which is mine, but the tweet was that there are no AI shaped holes lying around waiting to be filled. What the gentleman meant by that was you have to think differently. So, your point about this conversational interface, what the tools are doing is taking the way that we think about information, process information, the logic flows that we kind of, uh, that we learn and develop over time as we come up in our career and gain experience. You’re taking that and codifying it in software. And so you need to think differently about what that means for your workflow.
Marc Low: And the first comment that we make to clients is you need to start thinking in Visio, right? These students are thinking about the steps that you take from, I have to do this thing to get to this output, right? And what are the steps that are happening in between? Where might generative AI play a role in accelerating bits of that? What do I need to think about as a consequence of using that technology, right?
Marc Low: Because it doesn’t work like a calculator. It’s a At the end of the day, it’s a predictive model. It’s spitting out things that it thinks is going to make you happy and keep engaged in the tool. And so that has implications in terms of the work that you get done. And so you need to start thinking in these kind of workflow oriented ways.
Marc Low: And so it’s not you know, the headline is, hey, we can use it for our accounts receivable process in finance, but that’s kind of only the headline view of that. The subheading of that is all the steps that happen in there. And some of that is generative AI. Some of that is going to be other technology solutions, and some of that’s going to be good old fashioned humans. And so the work is really to figure out, I need to get this thing done and the output looks like this.
Marc Low: What are the steps along the way and how do I need to reimagine what tools I use to get there? And we’re not used to thinking about technology in this way. We’re used to thinking about technology like Excel or like a calculator, right? Where I know the need to add two numbers together, I put it in the machine, the machine then spits out 1 +initively equals 2, and generative as is not like that. And so, again, it’s a different paradigm in terms of how you interact with technology.
Marc Low: It’s a different paradigm in terms of what you’re asking the technology to do, and it’s a different paradigm in terms of making sure that the end product that you get out of the bag is predictably the same so that it’s useful in a business context. Because most of the work that we do inside organizations in those 900 processes that I’m describing that we’ve mapped some Gen AI to, most of that is not just like, Hey, do a first draft of something and publish it. There’s a whole bunch of steps that go along the way. And so, that’s really where organizations, I think, are still struggling in some ways is to, uh, you know, map that landscape of 900 and growing use cases to the organizations, and then how do you reimagine the workflows that those sit on top of.
Stéphane Zanoni: And so if I was at Riva, we’ve kind of jumped in with both both feet into the deep end. Uh, but if I was to get started, where would I start to learn about assessing the areas that have the greatest impact to those 900 use cases? You use the 900. That feels like an infinite possibility when you’re looking at the entire business spectrum of all these use cases. How do you isolate that?
Stéphane Zanoni: And where do I get started as a leader in in my organization?
Marc Low: Yeah. Great question. The $64,000 question, where do I start? Typically, what we advise is you start in an area of the business that is really what you’re assessing at an organizational level is feasibility and risk. And so, typically we would say like, don’t start with a thing that interacts directly with your consumer, as an example, right?
Marc Low: Where if you’re learning how these systems work and the types of outputs that they create, and that has impact on your brand directly in terms of how you interface with a consumer, that’s probably not the best place to start for most organizations. Most organizations are starting in that middle or back office and picking, uh, in terms of selecting the kind of area of the business. That’s a little bit of an art in terms of, uh, you’re looking for stuff where you think you’ve got a significant amount of homogeneous process. So that’s really what you’re looking for is you’re trying to pattern match the ways that those in those 900 use cases I’m articulating, there’s a bunch of kind of patterns of usage. So you’re trying to pattern match that usage to your processes.
Marc Low: To have value, you want to have a lot of similar styles of work being done inside that group. And that’s kind of top down, like, executive overview of where do you start. So, finance tends to be one place. Uh, HR tends to be one place. Legal tends to be one place that organizations are kind of starting to think about it.
Marc Low: And then the second part of the art of that is then each organization is different in terms of their that group’s appetite to experiment. And so, that’s a little bit of a only sometimes you need to understand how the different groups will think about their appetite to experiment. And then the last part of the process is that to take that group that you’ve identified as your kind of you know, your test bed, right? So, for example, the legal your in house legal counsel. Um, then the work is then to spend some time understanding like, Hey, what is a day in the life of this person inside that group?
Marc Low: What are the things that they’re doing on a day to day basis? How can we kind of quantify that and start to bucket workflows again to pattern match? And what you’re looking to do is identify really effort impact and feasibility. So, when I say effort, like how hard is it to make that behavior change? How hard is it to stand up that solution?
Marc Low: How impactful would that be? So, if we identify a solution and you’re doing it once, like if it’s one thing that you’re doing outside that group, it better be very, very high value. And in some cases, like in your legal group, it can be. So we work with a client where multiple 7 figures worth of legal work is flowing out the back door to external counsel because their internal teams can’t get to enough stuff. So, that might just be like one thing that they need to do, right?
Marc Low: That’s a very high value, low volume, great use case, right? And then the last piece of it is the feasibility, which means how readily accessible is the data, how clean is the data that it sits on, and so on. Because all these systems are being driven from the data inside your organizations. So, if you, for example, are sitting on Microsoft SharePoint and the SharePoint’s a disaster, well, guess what? The outputs from your generative AI tools are also going to be a disaster.
Marc Low: And so, that’s the last piece of it is that, really effort, impact, and feasibility in terms of things like data. Once you’ve done all of that, then there’s an assessment around the risks and how do you identify them and mitigate them. But the first piece is pick a group where you’ve got a lot of similar types of processes. Ideally, it’s not customer facing, so you can isolate the sort of experimentation of it. And then, once you pick that group that has high appetite for experimentation, now you start to think about, okay, let’s start to map those workflows and identify where we can start, itemize those, and then prioritize those.
Marc Low: And pick off the ones that are the least effort, highest impact. That’s typically where we start. Experiment, put some runs on the board to prove that it can be done. And then you can start to build your roadmap of things that you want to do next. And our view is that ultimately it’s going to touch every part of every business.
Marc Low: So it’s not like it’s a one and done. Uh, it’s really just about how do you start, learn a lot about, uh, what’s required to stand these solutions up, and then you can start to scale that out across your org.
Stéphane Zanoni: And when we had a an economist come in and speak to the team, and one of the things that they had identified was that in that there’s a a productivity gap in in the industry, or not just our industry, but there’s a productivity gap in the world economics. And when I look at some of the stats that I had read, something like 10% of Canadian businesses were leveraging generative AI today. And when we think about automation and productivity, what do you think is needed in order to to transition that, to get it over to get that number to double in the next couple of years? What’s what’s required in order to see that number increase to get that, uh, adoption rate up?
Marc Low: Yeah. That’s a tough one for sure. Um, a couple of thoughts. 1 is the technology is evolving very, very quickly. So, um, you know, there’s that I think there’s an accessibility issue.
Marc Low: You can go and, you know, any one of us, uh, can take our credit cards and sign up for chat gpt tomorrow or today, right? You can start to use it. Um, but that doesn’t really help you in a business context because you got to be thinking about the safety of your data, what you’re inputting there and how it’s training the models. And also, of course, all the things around, can you trust the output in a business context, right? And so, right now, there’s a gap between, I want to do this thing, and to stand it up at an enterprise level in a way that the organization can trust both the system architecture and the outputs of those tools, there’s a gap there.
Marc Low: My view is that that is shortening very, very quickly. And that as the tools become more sort of prolific, the large platforms are going to figure out how to make it more accessible and reduce the barrier of entry cost wise to stand the system up. That’s kind of point number 1. Point number 2 is that they’re going to proliferate across every major platform. So the big, uh, you know, any big enterprise platform will have some version of generative AI baked into it from one of the providers.
Marc Low: Um, and then the last piece is that, uh, is really the skilling around it, right? So the challenge, the tension with putting these systems inside organizations is, uh, this bit of a joke not joke, right? But if you go, Hey, there’s a 30% productivity uplift, right, to use these tools. Well, in a team of 3 people, you’re kind of looking to your left and looking to your right and trying to think, Oh, that’s going to impact one of us, right? Not quite sure which.
Marc Low: And so that’s part of the story is A little bit of the narrative I think is that there’s a justifiable understandable fear that it’s disruptive to, you know, the world of work and that’s all of us feed our families. So, that’s a natural reaction. I think it’s unfounded, uh, in that that really what we’re seeing is that the use of these tools radically accelerates the work that people do. And that has big, big implications in terms of if you can lean into that as an individual, uh, and really get the benefits of these tools, massively accelerates you as a professional, and in turn, you know, accelerates your organization. That’s really the main kind of headline there.
Marc Low: But because we’re, you know, we’ve got tens of thousands of years of survival baked into the backs of our brains, right, You get introduced something new and all of a sudden your fight or flight mechanism kicks in. And so, there is resistance at an individual level sometimes to implement these tools. And so, I think as people get more familiar, that barrier will come down and really the users will take that list of 900 use cases that I’m articulating, and that’ll double, triple, quadruple in very short order as people figure out, hey, I could do it for this. I could use it for this. I could use it for this.
Marc Low: And it’s benefiting me personally in terms of the work that I can do. So, uh, I think it’s that combination of things. So, as that happens, I think the usage of the tools will accelerate and you’re going to start to see the benefits in terms of that productivity outlook, or uplift rather. And in Canada, especially, we can certainly use it, right? There’s a there’s a notable productivity gap per per capita in Canada relative to our, uh, neighbors to the south.
Marc Low: And so that’s this is part of the story of is, you know, how can this act as a lever for all of us?
Stéphane Zanoni: And so when we look at AI in the enterprise, I think what I heard you say is, a, we need a place that can store confidential information and can be gate some of the training so that we’re not training public models with corporate data. Are we seeing an increased adoption in in financial services across industries? Or are there certain sectors that are more eager to jump onto these technologies?
Marc Low: Well, the ones like the types of clients that read the services, right, are, uh, you know, especially in financial services, that’s one of the places we’ve seen huge adoption. Let me get on and say that’s partly a reflection of, um, is the sense of, uh, as I’m describing, the large groups with a lot of homogeneous type of processes. The big banks, the big financial institutions are going to be very ripe for that type of implementation, right? Where you can see very clearly that if I put it inside this part of my organization, I’m going to get a big productivity uplift. It’s a big lever for the organization.
Marc Low: So they’ve been one of the first, for sure, financial services. The distribution outside of a couple of core industries has been, I would say, fairly consistent. Um, and the laggards really are the ones where the core business is not technology. So, think about things like, um, you know, some of the extractive industries in Canada, right? Forestry, whatever, where the main biz the core of the business is, uh, you know, grow a grow a product in the ground, harvest it, and, uh, and, you know, and put it into market.
Marc Low: And so, uh, you know, those types of organizations where, uh, the core of the business is still in the legacy type of industry. Now, those are probably the ones that are a little bit on the slower side. And the rest, I would say, you know, between those two poles, uh, is a lot of experimentation and people trying to figure out how to how to scale it. And that’s partly a, you know, a discussion around the cost of the technical effort to do that and the change required inside the organizations to really get it to scale. Uh, and so those two challenges are everyone’s everyone’s in the same boat trying to figure out how to solve for that problem.
Marc Low: But that that those are the sort of barriers to to seeing more kind of widespread adoption when you’re in that middle kind of cohort.
Stéphane Zanoni: Is there anything, Marc, that you wanted to talk about or specifically call out that that we haven’t yet touched on?
Marc Low: Well, what I would love your listeners first of all, I think, you know, the markets that Riva serves, uh, are the ones where this is the Riva solution and the types of conversations that you’re going to have with clients are right in this wheelhouse. So I think, you know, super exciting for, uh, for ReVet to be at this particular point in time is is comment number 1. And really, um, the solutions, uh, to reduce the barrier to entry for organizations to leverage us, us. That’s the name of the game in this space. It’s kind of, I think, point number 1.
Marc Low: Point number 2 is, you know, for the on an individual basis for the folks that are listening is really to get your mindset at a place where these tools are a lever for you. Right? They’re that Archimedes famously said, give me a lever long enough and a fulcrum and a place upon which to stand,
Stéphane Zanoni: I’ll move the world.
Marc Low: Right? And generative AI is that tool and you are the lever. And really what that the call to action is to lean into experimenting with the solutions, to try to figure out where it can personally benefit you as an individual. And it requires a high degree of agency. It requires everyone to take a leap of faith, lean in, and get a little bit uncomfortable.
Marc Low: But if you approach it from the perspective of how can this benefit me and how do I you know, rethink how I need to position myself in a world where knowledge and reasoning is ubiquitous because everyone’s going to have access to the same tools. The question you have to ask yourself in terms of, you know, of your, uh, how do you get you know, how do you have value in the marketplace of talent, right, is your ability to apply the results that are coming out of these tools. Your ability to create context around the results. The ability for you to, uh, you know, to really do more and push beyond the boundaries of what you typically would have done, right? And so, you know, me, I’m a strange one.
Marc Low: I like being on stage. I like my parents would tease me, right, that I haven’t met a microphone that I didn’t like. Um, but I’m not particularly good at analyzing financial statements. Generative AI is a lever for me to be able to do that. Right?
Marc Low: Now, what does that allow me to do? Well, if I’m looking, you know, my part of my job is innovation inside KPMG and working with clients around, uh, you know, understanding what solutions can, you know, do you need to bring into the fold and how do you think through the financial implications of that generative AI is a tool for me to be able to, uh, you know, contribute to that conversation in a way that I couldn’t before. And so, um, if you start to think about the skill sets that you have, you know, you used to think about it as I get really deep expertise in the thing that I do and that’s what creates my personal moat, right? That’s what makes me valuable in the marketplace. You need to rethink how you’re positioning against these tools.
Marc Low: And so, you know, the only real message for individuals and organizations is you can feel any kind of way about the technology. All of your emotional reactions are valid. The only thing you can’t do is put your head in the sand and pretend it’s not happening. Uh, because the tools are proliferating in the marketplace inside organizations. And so the gig or the goal is, uh, is to experiment, to get comfortable, and to figure out how to drive value with them.
Stéphane Zanoni: And a lot of what you’ve just described has to do about experimentation. And in many cases, some of these enterprise organizations have an aversion to experimenting with some of this data because it’s such an important set of data or corpus of information. How do you work with customers to let them understand or with clients, help them understand the value that’s in the data, but then also exposing some of that data in ways that traditionally would have had answers of no. You wanna do what with my data? Uh, is there building trust and building trust in the data, building trust with trusted vendors?
Stéphane Zanoni: Uh, any recommendations there on how for the first time, we’re looking at fine tuned models. We’re looking at vertical AIs. We’re looking at micro LLMs trained on your corpus of information that never would have happened before?
Marc Low: Part of the answer is a technical answer, which is, uh, you know, you architect your system and solution in a way where your data is secure, the model is not being externally trained, uh, and so on. So that you know, there there are relatively straightforward ways to protect the data. The harder question I think is, um, you know, we used to talk about at KPMG, we used to talk about responsible AI. That’s how we would brand that view of, um, you know, how do you leverage these tools responsibly? We’ve actually rebranded that to trusted AI, which I much, uh, I much prefer because it articulates what you’re, uh, sort of a conversation in multiple parts.
Marc Low: 1 is, can you is your data secure? Can you trust that technically speaking, the sensitive data that you’re referencing, right, is not being pushed into the outside world? So that’s step number 1. Step number 2 is the stuff that’s harder to to get to, which is what you’re what you’re referencing in the first part of the question. Can you create the space to experiment?
Marc Low: Can you then, uh, understand what the tools are doing and create process and rigor around being able to trust the output of what these tools provide in a business context. Much, much, much, much harder. And so, um, the conversation, our the conversations that we have at a sort of C suite or board level is really more trying to articulate, listen, the world around you is changing. It’s changing very, very quickly. So to refuse to experiment is to really sign your own metaphorical death warrant as an organization.
Marc Low: Right? Because your competitors and a whole bunch of emerging competitors are going to be using these tools and leveraging them, and they will competitively eat your lunch. And so, it’s really the mindset shift around, can you create the space to experiment knowing that a whole bunch of stuff you can experiment and build, you’re going to throw away. Right? And that’s that’s okay.
Marc Low: That’s not that’s not failure. It’s a it’s a mindset shift trying to figure out, like, you’re in search of, uh, search of value. You’re in search of a new way of doing things. And so you got to create the space to, um, to do that. And, really, that’s the that’s the, um, the sort of invitation for clients in that conversation.
Marc Low: So, yeah, first part of the answer is really it’s about technically ensuring that you can experiment safely. The second part is creating the mindset and the space to experiment, and then figuring out what roadmap to go from experimentation to enterprise hard end so you can start to put these solutions into production and rewrite the processes that your organizations, uh, use to then deliver that end end product or end end value.
Stéphane Zanoni: Well, I I think we’ve I’ve heard it coined either responsible AI or ethical AI. The trusted AI is a is a great way to guess to summarize it. I feel like a lot of it comes back to explainability. And how prevalent would it be or is it going to be for the language models to have to include explainability in the enterprise space. And one of the things if I’ve got the document repository with all of my contracts and I ask it, which contracts have these articles in it?
Stéphane Zanoni: Uh, please summarize them. To have all of those documents referenced is also another great way of showing this is where I found the evidence to make this claim. Um, where does explain ability fall into creating trust?
Marc Low: I it was part and part and parcel of it. Absolutely. So, um, you know, the retrieval augmented generation solution, right, is is you’re hoping that it’s going to leave that trail of bread crumbs. You want to build it in a way where it’s creating those references. Commercially, the solutions like Perplexity, which have done that, try to replicate a Google style experience, right, of giving you the answer and then creating those references.
Marc Low: And really that’s, um, in an enterprise context to build a similar type of solution, you want to have that same story. And there’s a nuance in that explainability story, which is also I think the idea of, yes, references are part of it. So how did I What data sources did the tool use to create the output? And how did tool think through that process? But then have you also ethically thought through the implications of using these tools in the context of your business operations?
Marc Low: So, um, the metaphor attendees with clients is summing to the effect of, imagine a dating app, right, which is driven by algorithms and it’s a that’s sort of a version of what’s happening inside one’s generative AI models. Right? You can engineer the algorithm for all kinds of outcomes, not all of which are ethical. And so it depends. You got to be very clear about what business outcome you’re trying to drive.
Marc Low: And the explainability piece comes to me, also comes back to this idea of if we live in a world where these AI systems are driving business decisions and the outcome then is being pushed out into the world, you better be able to articulate what was the logic behind how we leveraged these systems? What was the output that we created? And can we articulate how we got from point A to point B, right? Um, and so that again comes that’s part and parcel of that trusted AI story is, ethics and explainability are really key to thinking through, what do I want to do with these solutions? Am I doing that in a way that’s ethical?
Marc Low: Am I being mindful of the end state or the solution that’s being driven? And being able to explain not only how we got there in terms of the technical sort of steps of the tool, but what was the thinking that drove that outcome as well? And so, especially in large organizations where you’re thinking about, um, you know, publicly listed entities and so on, right? There will be a whole bunch of regulatory reporting requirements around, uh, the use of these tools. And so organizations, you know, need to be thinking in that context, um, as well.
Marc Low: Right? And so, any good deployment of these solutions inside organizations like that takes that into account. Um, and right now it is, you know, still a little bit of the wild, wild west in terms of regulators trying to catch up to where the technology sits. But ultimately, it’ll form part of it. In the same way that ESG reporting requirements are starting to come down the pipe for regulators, the same thing will cascade around the use of AI systems.
Marc Low: So, we think that that’s that’s kind of the next phase of where this is all headed.
Stéphane Zanoni: The very next element in that journey for me really is around the bias and bias in data, bias in the intention. Any suggestions for some of the listeners as how you can interrogate the data to suss out any kinds of biases or think about the problem in a way that might eliminate some bias in in the teams that are solving these problems?
Marc Low: Yeah. Again, as a two sided answer, uh, that that it’s a it’s a good news, bad news. Um, the good news is, uh, I I think the bad maybe let’s start with the bad news. Let’s start with the bad news first. Bad news is we’re all biased.
Marc Low: And so, uh, you know, what the tools do is they codify that that bias. Um, and so that’s that’s the point number 1 on the on the sort of tool and solution side. Interrogating the data for bias is more of a data science problem. I would probably defer the technical, um, you know, parts of that to to colleagues of mine who are more clever than I am on that on that topic. But, you know, I think the good news is that what the conversation has forced us to do is to have is to interrogate in the first place.
Marc Low: So until we were sort of talking more broadly about the use of data, the use of AI in stitching it together into every facet of business, um, that the discussion was the domain of a relatively small number of folks. And so what we’re doing now is actually having an open discussion about, Hey, what data sets are we using? What bias is baked into those? How can we be aware of it? How do we interrogate it to try to minimize that impact as much as possible?
Marc Low: And then when we layer a tool on top of that data set, how are we thinking very carefully about how that tool was trained, How that bias that has been now baked into the tool will influence our outcomes and so on? Um, again, the bad news is that bias exists all around us. Uh, it’s baked into what we do as humans. The good news is we’re having the conversation and interrogating the data and the tools in a way that we can then stand behind what that output’s going to be. And again, that is coming back to that sort of sense of trusted AI to interrogate that bias both in the data and the tools.
Stéphane Zanoni: And we can’t talk about trusted AI without thinking about some of these autonomous vehicles. And I spent the week in San Francisco last week, and I took my first driverless car experience. And it’s wild.
Marc Low: I haven’t had the pleasure yet personally, um, but I certainly watched all the videos. And I mean, this is exactly what you’re, you know, what you’re driving at, right? Pardon the pun. Is, uh, yeah. How do these systems in real time interrogate what they’re sensing, and then how do they drive decision making on the back of that?
Marc Low: So that’s a very real life example of the challenge of exactly that. And so, you know, there’s a a whole bunch of, uh, sort of AI ethics questions around how do you think through the that problem set and so on. But in that example, it’s really, uh, it’s really life and death, literally.
Stéphane Zanoni: And that’s that feels like one end of the spectrum where you’ve got a system that’s making decisions on our behalf. And then the other end of the spectrum seems to be maybe a more traditional recommendation type system. Is is there you’ve got recommendations on one end of the spectrum and then you’ve got a system that’s taking action on behalf of its participants. Um, it is there what is the lifeline or what is the what is the spectrum that we should be considering for the trusted AI or for trusted execution at the enterprise?
Marc Low: Uh, well, all I mean, all the way all the way through, I would say. And in fact, I think the stuff that that you’re articulating there around the recommendation engines, in some ways that’s almost that can be almost more insidious if you’re not careful. So I think of an example of, you know, basic one, right? But credit applications in a financial institution, right? How are decisions being made about what to approve, what not to approve, right?
Marc Low: And to coming back to this idea of bias, if you’re not very careful about how do you think about what data sets are being driven into that decision making engine, even to recommend, yes, no, you create unintended outcomes in terms of what the output is. And so, the immediate consequence is more readily apparent when it’s a system that’s autonomous or semi autonomous. And that will become more and more important in the organizational context as these tools move from recommendation engines to sort of agentic workflows or, you know, autonomous, semi autonomous workflow agents inside an organization. Um, but the conversation is the same all the way across, which is, um, you know, thinking very, very carefully about are we accounting for what’s baked into the data that’s going into it? And then, what outcome are we trying to drive as a result of the tools?
Marc Low: And so, I think it’s more a function of what’s the immediate impact, but the lens should be the same across that spectrum.
Stéphane Zanoni: And so when we start to think about these autonomous flows and we start to consider this concept of the agent or the AI agent. It feels like there’s been a transition in the industry where we went from Jenny I type copilots to now transitioning to an agent based an agent based experience where you now have a, maybe, a highly tuned model that has very specific guardrails that’s looking to solve for a specific set of problems. Where are you seeing that that shift happening, and how quickly do you think that transition from these horizontal language models, the OpenAI’s of the world, to much more vertical solution fit, uh, outcomes?
Marc Low: I my sense of it is that it’s going to be a hockey stick kind of adoption. Um, it’s slow at the moment. And part part of the reason is that if we’re oops. So where we’re seeing a lot of early AgenTek workflows is in things like, uh, you know, customer service interactions or sales sales and marketing type interactions where you wanna you know, you’re getting a piece of input. So imagine, you know, customer lands on your website.
Marc Low: They provide some, uh, you know, they provide some inputs to you. Right? They give you their email address or their telephone number. And, you know, we would have talked about previously robotic process automation of taking just that input, driving it into a CRM, and then cascading some actions out of that. And where people are layering these tools is to say, great.
Marc Low: I’m, you know, I’m getting this piece of customer information. It’s going it’s I’m just automating. I’m putting it directly as a workflow into my CRM. That’s triggering some research that informs an email or some type of customer service interaction and so on. Because the tools are still relatively early, I would say that the quality of that output is mixed.
Marc Low: And so to scale that into large enterprises where the appetite for risk in those interactions is lower, that’s where it’s gonna be slower. So I’m just giving one example, but that’s where we’ve seen kind of the early, early stuff. My sense of it is that as the technology improves, so the results get more predictable, the interactions and the ability to build those workflows becomes, uh, better, and the ability to the cost to stand them up becomes easier. You’ll see them proliferate. You’re already seeing a few of those, um, in, for example, Microsoft Copilot has built a workflow tool inside, uh, inside Copilot itself, inside Copilot Studio.
Marc Low: And so I think those type of solutions will start to proliferate. As the technology gets better, it gets more easier to consume, and therefore gets pushed into more and more of the org. And the results of or the outputs of the tools become more predictable and more useful at an enterprise scale. And so, it’s relatively easy to build proofs of concept. Take that again.
Marc Low: It’s relatively easy to build proofs of concept for these type of flows to make them enterprise ready. At enterprise scale, that’s where the effort is still significantly higher. And that’s why the adoption hasn’t hit quite as hard yet, but definitely coming. To me, the next part of the discussion is in a world where it’s easy to build workflows, the results are predictable, and effectively a piece of software is driving a lot of those interactions, what role do humans play, or rather what type of role does that create? And that’s the harder question, or the thing that’s a little bit chunkier is, if we if organizations can flood the zone with content and interactions that are all software driven, software initiated, software created, you know, how do you how do you stand out competitively in that landscape?
Marc Low: And that’s, I think, where ironically, the pendulum’s gonna swing the other way. And, uh, the future’s largely gonna be driven. It’s like a the future is human. The future is not AI. The future is driven by humans that can take those tools and then layer value on it, apply your sense of, uh, your sense of context, your experience, your, um, for lack of a better word, taste, right, as a human to then take those tools and outputs and then drive some human oriented value, right?
Marc Low: Because at the end of the day, we’re having we’re just 2 meat bags walking around sucking up oxygen and having a human experience, right? And really what these tools are the implication of these sort of AgenTek workflows is that you start to really attack the nature of what it means to come together as an organization. And really, it’s about humans coming together, creating something that has value to other humans, and the buying experience is, at the end of the day, a human to human interaction. And so, the question is going to be, how do humans then layer into that process? Even though it’s driven or fueled by these AgTech workflows, at the end of the day, it’s really about humans having a human experience.
Marc Low: And so, that’s where that that’s the chunkier, grittier conversation, I think, of these of the implication.
Marc Low: Well, and I I wonder what the scenario would be like if you’ve got 2 agents representing the buyer and the seller. How how that interaction becomes humanistic, and how do you prevent the human element from being removed from that experience?
Marc Low: Or do you even want that? Right? These are the fundamental, like, you know, philosophical questions of, like, what does it what yeah. What does it mean? Yeah.
Marc Low: What does it mean for an organization to have to to be represented there? There’s all kinds of really I mean, we can rabbit hole this one big time, but there’s all kinds of really interesting implications, I think, around, uh, what does that mean for organizations to be represented by an AI agent, for example, in a procurement discussion? Right? What is that what are the ethics and outcomes of that? Do you even want that?
Marc Low: Nevermind the idea of I think there’s a lens here where it’s kind of the venture capital view of the whole opportunity where it’s like if you Humans are an extraneous distraction to organizations creating stuff and selling to one another. But that’s not typically how the world is actually structured, right? That at the end of the day, business has always been about how do you optimize the delivery of that thing. But at the end of the day, I’ve still got to sell you, Stephan, something for a dollar, and I’ve got to produce it as efficiently as possible, obviously, to create as much profit. That’s the goal of a capitalist system.
Marc Low: Right? I’m going to sell you something for a dollar, and I’m going to produce it as efficiently as possible to maximize my return on it. But it’s not just 2 solutions autonomously interacting with one another doing something, right? To me, that’s sort of antithetical to, um, yeah, to what it means to come together and do it in the first place. Anyway, these are weird and wild, um, yeah, nebulous philosophical discussions about what our world is likely to look like on the back of it.
Marc Low: So mostly what I spend my days doing is prognosticating with organizations about, you know, how do you take that first step and, uh, and how do you think about, um, kinda getting started on that journey.
Stéphane Zanoni: But we’re all figuring it out. Right? We’re building we are literally building a plane as we fly it. When I I look at it as being AI, maybe AI powered, not necessarily AI driven.
Marc Low: Yes. I think that’s the I think that’s exactly the right, uh, the right lens, um, that we’re all gonna have these tools, um, at our disposal. And every organization’s going to have them driven into their in the work they do on a daily basis. But that we are going to be AI enabled or AI powered and not AI driven is my, uh, prediction for where it where it’s all headed.
Stéphane Zanoni: So thank you very much, Marc, for your time today. I think we both agree that it’s a better together situation when it comes to leveraging AI. And appreciate everyone joining us for our RevTech podcast.