Closing the AI Value Gap

Today we're focused on closing the AI value gap and I'm thrilled to be joined by an expert on the subject and a Verscell customer, Dan Martinez, managing director at BCG Platinian. So Dan, welcome. Thank you, Jane. Great to be here. Awesome. Uh well maybe to sort of set the table for us, BCG research found that only 5% of companies are generating substantial value from AI while 60% are still struggling. Uh what's creating this gap? Is it the technology problem and execution problem or something else entirely? Well, Jean, if we look into the last uh three years, right, since since Genai basically started in 2023, lots of companies started with use cases and pilots, right? Um, and I found that some companies were almost like competing for how many use cases they could get to and sometimes they would get through a 100 or 300. I've seen organizations have, you know, hundreds of these use cases. And I feel like ultimately people just got stretched too thin. Some of these ideas were very small. they were not what we consider process reimagination. They were not functional reimaginations of of the organization and and then people just I feel like they got lost in the shuffle and some of these ideas you know I think the you know the business was aiming too low. Um also we found that um some of these ideas were not they did not involve capability build. So people were developing these use cases but they weren't clear what is the change in job families, how does that change on upskilling, what is the impact on people, what is the impact on processes. So I feel like organizations were missing the bulk of the work which is what we at BCG we call the 10 2070 which is 10% is stack um 20% is data algorithms and then um 70% is really the bulk of the work. It's rethinking the businesses, rethinking tasks, how processes are different, who needs to be upskilled, how jobs will change. And I feel like 23, 24, it was just a lot of people, you know, experimenting, testing with these use cases, but not really thinking about look, they have to go into production, they need to scale, we need to think about a a whole bunch of stuff. So, I feel like companies now are building the muscle, the discipline, the attention span. The leadership is looking at this. It's no longer AI is no longer a technology project. AI is no longer a um you know a little experimental project. It's here to stay. It's existential risk. It's competitive advantage. Yeah, that makes a lot of sense. I think your point on 70% a lot of what I found in the work we've been doing in GTM is actually a lot of that is is even pre-production if you will of understanding what a best-in-class process ought to look like. Um, and do you have all the content for that having brought it through? So, piggybacking on this, there's a phrase that's been coming up in enterprise AI conversations, which is the shift from systems of record to systems of work. What does that mean in practice and why does it matter for how companies think about their technology investments? Yeah, I first saw this concept in a article from um u VC in the Bay Area where they talked about you know with the with the emergence of digital 20 years ago companies moved from on premise software to SAS and moving to large enterprise packages what we call systems of record right so if you think of Salesforce or service now or workday right these are systems that hold a lot of corporate data they have your customers your orders your uh deliveries right? Uh your financial data is in these systems. But then over time we felt that people um wanted to collaborate differently and we've seen the emergence of more modern systems of engagement for example uh you know slack or teams uh zoom for example and people are using these systems to engage collaborate internally collaborate externally so it's almost like the user interface to think from an enterprise architecture perspective the UI has moved from the systems of record to the systems of engagement and now what we're seeing with AI is is a new phenomenon altogether which is the business logic of some of these systems of record are now moving to systems of work and they're becoming agentic. Right? So what we used to see as rules-based deterministic types of features now they're moving to probabilistic system prompts in these multi- aent systems. Uh and of course the hyperscalers are moving in that direction. They're creating lots of platforms. I mean, Verscell is in in that scope as well, helping allowing companies to very quickly rapidly build these new agentic systems. And then we see companies like Salesforce, they're moving in that direction as well, right? They're they're building agent force as a capability and going to market with readymade agents, right? Um, and this is something that I feel like CIOS are starting to understand and grasp this new reality, right? moving from these systems of record, how do I invest in these systems of record going forward, but then how do I build capability that allows me to shift these business rules into agentic systems. Um, I feel like this is becoming more clear. 2025, 2026 is when we started to see organizations move to multi- aent systems. Um, start to go from experimentation to production. uh building more uh resiliency, governance, all the um architecture around it. Uh and that's what that's the pattern that we expect to see more and more in 26 and 27. Yeah, I mean I I can bring that to life pretty specifically with Burcell of how you uh describe it aligns exactly with what we've experienced here which is we've got Salesforce still a system of record. Um we started by building a singular agent to handle our inbound leads. So folks who fill out contact sales in building that agent we were to able to go from 10 sales development reps down to one that then formed the basis of a playbook platform where we're now have multiple types of the sales development function. So event followup um or you know hot p uh PLG leads that type of thing. So you've got all of those multiple agents running and then to your point system of engagement. So, a bunch of this stuff gets now piped into Slack um or built custom workflow UIs um because Salesforce front end didn't necessarily represent those um exactly how we wanted. Uh actually what what you just teed up is precisely what we've seen play out in um you know our first six months of of bring AI really deeply to to go to market. How do you help uh companies identify which workflows to prioritize? uh you know at Verscell we're we're very much avoiding working to avoid random acts of AI. So we found that the highest likelihood of success for agents comes from tasks um that are a little bit more on the repetitive and deterministic side. So not a ton of cognitive load. Um the lead agent I just get um is a good one. Does that match we're seeing um BCG, you know, to my knowledge's language is like stop with the use case mindset and sort of open that and pilot purgatory. I've heard a couple times. Yes. So I think you're you're spiritually aligned with Versell's random acts of AI AI but um you know again how do you go from that rapid prototyping to picking the use cases that are actually going to drive value? Yeah, I think we're super aligned there. I mean 23 24 everybody was stuck in pilot purgatory learning figuring out the technology solving accuracy hallucination problems building rag applications um but ultimately realizing that it was very hard to scale and I think people realized that it was hard to scale because the business there's a lot of work on the business side right retraining people rethinking processes etc and I feel like we shifted from that use case pilot mentality to focus on value pools and what are these big uh reshape opportunities for organizations, right? So how does my servicing organization will be different? How will my finance function will be different? How will my supply chain function be different? So I think people start to amplify the scope think process value chain level picking specific examples in the value chain to drive um but really focus on a much bigger scope and in a scope that's much more businessled um a scope that requires risk compliance legal to be involved to make sure that we understand all the ins and outs of this thing um and so we sort of moved away from use cases into value pools doesn't mean that companies are not using use cases I still see that lang language happening But we're moving to value pools and we see for example some very clear value pools in the market. So for example servicing customer service help task has been arguably number one area of where companies are using AI. We're starting to see an bigger emergence of startups in this space. Some are becoming well solidified in the market. AI for software engineering. I mean this is a huge value pool for organizations. This is exactly where Versail is squarely in as as as one of the leaders in the market driving the charge here, driving the journey. I mean, I feel like we're just scratching the surface there. Um, you know, the, you know, the tools are gaining adoption. The engineering teams are are are building on top of it. I mean, some of these tools are being becoming more integrated and embedded with the uh with the ecosystem in enterprises. Um, this is actually one of the things I really like about Overell, the fact that you guys already built lots of integrations, they're very thoughtful about, right? So, um, you know, if companies need to do this on a hyperscaler, you know, they have to work through lots of hyperscaler services to pick from, etc. I feel like again, we're just scratching the surface here. We're going to quickly move into using these technologies to build multi- aent systems, to build uh digital twins of organizations. And this is where we're we're starting to see the next future proofing of the organization, right? We're we're what's emerging at BCG is this ability to develop digital twins of processes of uh functions of the partners, right? This is such a scalable concept, right? If if I'm instead of focusing on use case, instead of focusing on value pools, can I create a digital twin of the organization and then simulate improvement ideas, right? And we're starting to like dip dip our feet in that in organizations where if organization comes to this comes to us with a specific problem. We create this rem it's almost like a reimagination AI that allows us to feed data into it and reimulate tasks and processes and what if scenarios right at the enterprise level. It's a really interesting experiment. I mean I feel like we're just now scratching the surface there as well. But hopefully that will inform how we find these value pools in organizations. Right? This isn't exactly the point you were making, but on the uh thought of a digital twin, uh we have a an internal data agent. You can think of it as like take um about like a data scientist analyst with about a decade of experience and it's sort of that level of capability. And this weekend someone added that agent the executive channel. So we were all joking that this was uh you know the first agent promotion company. Um but you know we absolutely are doing that. We're um pretty far along I would say on the uh data science side of things where you can actually see ways in which the agents that team is creating are in fact digital twins. you also started getting into sort of like, you know, how do you go from the prototype to production? Um, touching on things like integrations, all the types of things that folks um don't necessarily think about when you're prototyping, but uh you know, you don't want to have to go spin up 20 underlying services at uh AWS necessarily. Um, so what are what are the best ways you've seen folks uh bridge that gap? We're starting to bucket those gaps in in specific archetypes for organizations. We came up with these four archetypes of AI agents. Um the first one is people are going to self-service the development of agents, right? and they're going to use um and maybe some people call them agents or not but regardless um custom GPTs or um you know self-service tools where people are going to you know cloud skills for example and you know people are going to use these tools to develop um their own agents connect with systems like for example I have an agent that runs every morning reads my email sends me a summary of what do I need to do what what actions I need to take and sends me all the emails I need to respond on prioritize. Okay, I mean that's a self-service agent I run in one of the tools and uh and it's helpful for me personally, but then we're going to see other types of agents that are built still by employees and organizations where they're built uh in tools like Microsoft Copilot, running in enterprise systems, um running uh connected to tools like SharePoint, connected to data, etc. I mean a bit more sophisticated but still within the realm of employees developing them. Then companies are going to buy agents, right? And they're going to buy agents from agent force and what not, right? So we're starting to see we're starting to do more, for example, market scan of agents, right? Just like we used to do for digital apps and and SAS companies. Now we're doing market scans for agents. And then the next one is where it is going to come in and develop enterprise agents, right? And that's going to become you know much more uh you know science than arts. It's going to become uh they're going to be a lot of rigor around these agents. We we have to test them develop them well and there's going to be a lot more scrutiny around information security and you know policies legal rigor uh you know for example responsible AI is going to be a big uh important component there. guard rails. Um and then for these agents, we have an enterprise framework on how to develop these agents, right? This is where we see AI coding uh tools becoming um you know, a huge value for IT teams. We're going to I actually do think that that when we think about buy versus build solutions like Verscell and AI coding tools are going to enable AI team uh IT teams to become very proficient at building. Yeah, absolutely. I think we share a similar point of view on CIOS going from buyers of software to builders of software. Um I think a lot of the use cases we're seeing on Burst are internal applications just as much as external. So if CIOS are now becoming software builders rather than just buyers, what does that shift from a role perspective? What's going to be new about the role of the CIO? Yeah, that's interesting because on one side this is totally elevating the buy versus bill discussion and what does it mean for IT. Um we've seen companies um like consumer companies uh start to hire agent developers, right? So these are no longer your typical machine learning engineer that you know may have a PhD in data science and uh knows Python really well. and I've seen a a job job description for one of these companies and it didn't even require Python, for example. Right. So, it's it's a new strange world that we're getting into. Right. Now, people are enabled and self-sufficient to develop their own agents. Yeah. And so, a lot of what you're describing here is is really a central AI platform. And your research has shown that future-built companies are 3x more likely to operate a central AI platform. agents multiply across enterprise. What should that platform architecture actually look like? We've been having lots of conversations with organizations around how to design this platform, right? Um and and the design I'd say the design um you know two years ago focused a lot on building simple rag applications, right? So it's all about you know make a choice on a vector database make a choice on LLM that sits on your model garden build guard rails at the application level and you're good right and your biggest headache is you know accuracy problems um but that we've seen a departure from that thinking right and nowadays it's becoming much more complex right you need guard rails not just at the agent level you need guardrails at the orchestration level um you need to control not just for accuracy need to control for integration with uh core systems. Uh there's a multi-layer way of thinking about security on these agents. So there's a lot to think about, right? CIOS are having to to adapt um their IT teams up skill their architectural teams to be able to deal with this additional level of complexity. But that's what we need to think about when we go to multi- aent systems, right? Multi- aent systems is going to be a big step for organizations to be comfortable with. But that's what we see a big part of the value coming up in 26 and 27. So you you touched a little bit on the application layer there. Um if we're moving towards systems of work that we talked about, uh what role does the application layer play? Does the software that sits between AI models and users become more or less strategic? I mean for sure they uh they they certainly have a strategic role because they are the system of record. So this is so they are they they ultimately have the repository on data in the organization right so they become very they will continue to be very valuable in that sense um they're also very valuable because they're going to they're going to provide those enterprise APIs for agents use in uh in organizations um but the question is some of the business logic is moving from systems of record to uh systems of work so it begs the question what all happened to SAS, right? We've seen u some tech leaders saying that SAS is dead. I I'm not quite there yet, but I do think that they're going to become very strong databases for with a very specific structure with very specific uh control points and uh they will continue to be valuable that way, right? Some of these companies are realizing that this trend is coming and they're moving towards AI makes perfect sense, right? Some are more holding their fort and you know trusting that you know the you know wait and see mode a bit. Um but we'll see in next 12 24 months we're starting to see an emergence of these systems of work. Many of these are offering great opportunities for buy. I do think the SAS companies will need to become AI first right instead of digital first and that's going to take time especially for some of the big ones. So you mentioned there be a being a lot of opportunities, but you could also say that the AI vendor landscape is overwhelming right now. I think like most categories have have 10 players in them, which seems like more than will probably be supported long term, but what question should enterprise buyers be asking to separate real capability from marketing? And how how do they evaluate whether a tool will actually deliver value versus become shelfware? Well, for sure there's a technology fit, right? um how will this how will these companies uh how will those agents run on a uh enterprise infrastructure? How are they integrated into that technology stack? How are they integrated with the systems of record for example? Right? That's an ongoing discussion. Then there's then we ask questions around enterprise fit. For example, how do they manage compliance? Um how do they manage risk? How do they treat uh data privacy? Those are top of- mind questions. you cannot, you know, there it's not it's a non-starter at enterprise companies if they don't have a good answer for these types of questions. Um, we look at cost, right? So, that's the buy versus build cost. Um, you know, and some of these some of these solutions are very expensive, right? They charge at the uh user user per month uh level. Um, and you know, this is going to and companies will need to allocate budget for these types of solutions. I mean, the these solutions are coming. They're more expensive, but they're very valuable. And then um and then we look at maturity of the company. Um as you said, some of these are new entrance. Many of these are still in series A, series B. Many of these have maybe 100 to 100 employees, right? So they're younger companies and they're trying to get into an enterprise space. The enterprise space is very complex. Um and requires a lot of attention, requires, you know, it's a long sales cycle. Some of these companies, they take six to nine months to onboard a new um AI agent, right? That's pretty reasonable. I see that all the time. Um and companies are trying to figure out how do we fasttrack this process, but there's a lot of due diligence uh process to on board one of these vendors, right? Um but I'm starting to see interestingly some of these started with small to midsize companies. some of these vendors, these AI agents, they started with retail consumer and and I'm working with one of them and this is going to be the first quarter where where the enterprise revenue tops the retail revenue. So, we're starting to again we're starting to see the shift towards um enterprise becoming the biggest customer for some of these solutions. Yeah, same seeing the same thing over here at Verscell. Uh well, Dan, thank you so much for joining us. This was a great conversation for everyone watching. If you want to continue the discussion, please connect with Dan or me on LinkedIn. We'd love to hear what you're seeing in your own organizations. And if you're ready to move from prototype to production, check out the new V0ero at vzero.app. We just shipped some major updates that make it easier than ever to go from idea to deployed application. Thanks for joining our first shipped Q&A. We'll see you all next time.