How to Fix AI Overload & Copy-Paste Chaos with Ray Wang

Speaker A: AI without productivity is just more cost. Um, and that's what some people are feeling, but it's because they're not culturally designed to get AI to work.

Speaker B: Why are your employees actually more exhausted since AI arrived? Because most enterprises are running a copy paste economy, where disconnected AI tools flood workers with noise while the real work still has to be done manually. Welcome to the Future of Work podcast. In this episode, Workday's Jerry Ting and Constellation Research's Ray Wong make the case for why more AI tools aren't always the answer and what it takes to get AI working inside your systems instead of alongside them.

Speaker C: I'm Jerry Ting and I'm joined by Ray Wang, principal analyst, founder, and chairman of Constellation Research. Hey, Ray, thanks for joining me.

Speaker A: Thanks for having me.

Speaker C: Yeah, it's crazy, right? Because there's all this AI. VCs are talking about, this hype we're talking about across all the Alice that we're talking with. But the truth is, uh, a lot of employees are still working in a bunch of different solutions, and then they're having to copy and paste and do work between one and go to another. From your position where you're talking to CIOs and CEOs all the time, walk us through what you're hearing. What does the day of the life of an employee look like today with all this AI, has there really been less busy work?

Speaker A: Okay, so a couple of things. First, this is the best rule we've heard from our clients. AI without productivity is just more cost. Um, and that's what some people are feeling, but it's because they're not culturally designed to get AI to work. Uh, and so part of it is, in the old days, let's say maybe like three years ago, in the old days, you send someone out to come back with research, and two weeks later, they filled everything. They tell you what happened, right? You might send another team out, they'll come back by the end of the day, they'll catch up with you, might send someone out, they'll tell you what's happening at lunch. Well, the challenge is it's all coming back to you now in minutes and in seconds, you're like, oh, we did all this automation. This is great. We can let go of these people. No, you can't. Because what people are forgetting is that there's a timescale crunch. What that means is if you want to get more work done, you need more people. Um, so even when all this information is coming back, the human is now in the critical loop, and that Means that you are the bottleneck. So, Jerry, do you remember the first time you were an email, like ccmail, maybe Lotus Notes email, and you hit reply all, uh, and then everybody replied and said hi or said whatever, and suddenly you had an avalanche of work. That's what our workers are feeling at the moment, because they send an agent out, they come back, and then they gotta go enter something in a different system. And not that they're just doing one of these, they're now doing like 10 times the amount of what they were doing before. And they're wondering why they're so tired and they're so unable to get any work done. And then someone has this bright idea. It's like, oh, we can get rid of all these with automation. No, no, no, no, no, no. It depends what you're doing to be able to get there. And so there's two things you have to think about. There is deterministic processes and there's probabilistic processes. Wherever you can, you want to automate the deterministic processes, but then you have to understand the downstream implications of the probabilistic processes. And you need more precision in the downstream processes. And that means, for example, if you're okay with 85% accuracy in customer experience, you might be okay with it. Are you okay with 85% accuracy and getting your employee benef? And so you're going to adjust the amount of human labor based on the decision trees you've collapsed. And every time you collapse a decision tree, you have to say, is there a human on the other end? And do we need more humans to solve this? Because the volume just 10x on you, that's the thing. You've got to build that culture so that you're hitting the capacity that you're actually achieving so that you can then get the work done. But if you don't do the capacity planning or the orchestration, the work coordination, you are going to be miserable.

Speaker C: It's such a good point in the email analogy. Makes a ton of sense, because right now there's just a lot of noise, right? And you got clock code writing here. You got some other solution like Gemini or Copilot sometimes there one company we talked to has three, all three, and they're all bringing you back signal, right? And so before you can even get to benefiting from these AI solutions, uh, what are we automating? Are we doing the right thing? Right. Any advice you had for customers? Because I think it's easy. Right now, CEOs are pressuring CIOs and CIOs are saying, let's just go buy these tools. Any advice on how to actually think about planning for that AI first organization before throwing a bunch of tools at,

Speaker A: you know, like, you know, and what you've been doing with your customers for years is you map the value chain, you map the processes, you map the journeys. And then as you're going on long lists and you're doing the automation, you have to figure out, do I insert a human here for regulatory requirements? Do I assert a human here? Because oh yeah, we're only 33% accurate, we better have someone here. Do we insert a human? Because I want more touch. It's part of my experience, my employee experience, my brand experience internally. And so that's how you get things done. But you can do all this without buying the technology. Do this first before you buy the technology so that you can actually be right when you bring the technology back in. Now, let's talk data. Let's talk workflows. The data sets upstream downstream implications. You better get that right. And sometimes it's not mapped the right way in this copy paste economy portion. Like, yeah, I'm getting structures at different levels. And this is almost like the SOA conversation we used to have. This is what the peanut butter jelly problem was. You typically map a process and say, let's make a peanut butter jelly sandwich. Here's the bread, let's put some peanut butter, here's some jelly, let's put it together, let's slice it in half. So that would be what we call a classic CRM process. It's simple. But then there's the workday process, which is here's the bread. Oh, uh, by the way, is that organic or not organic? Would you like to toast the bread? Okay, let's toast the bread. Is that jam? Peanut butter. Well, hey, do you want organic? Non organic. Is that crunchy? Not crunch, that creamy? Hey, do we cut this in quarters? Right? Granularity becomes very important. And when you don't have the matching granularity and the data comes back and the metadata comes back and it is not the same. That's why you're doing a lot more work.

Speaker C: Couldn't agree more. In for a workday, when you talk about payroll data by country, by region, by regulatory standard, you go to the finance side and you talk about accounting. That semantic data layer that you're talking about is so critical. So let me ask you a different question. I mean, there's a startup for every single thing right now, right? And I used to Be a founder. And, you know, I, you know, was maybe guilty of this as well, but because of the excitement around AI, there is 10 startups doing every single part of the workflow is when you talk to your customers, what advice do you have for, hey, when should I go get a point solution or when should I work? Where should I work with my vendor that has a platform? How do you make that decision between best of breed versus going with a trusted partner?

Speaker A: Classic equation, build, buyer, partner, and how do you want to get to that solution? And if it's a core thing, part of, you know, okay, actually let me not use the word core. If it is not differentiated and it's something that is not within your strategic mission, you should automate the crap out of that and get onto a bigger platform so that you can take care of all that. But if it's something so unique to your business that you're doing it differently, then I would definitely buy or definitely try to put some investment in build so that you can get the differentiation you're looking for. Because there is a startup for everything today, just like there's an app for everything in the app Store. But the point being is the bigger the platform you have, the less headaches you're going to have over time to manage that. And the control plane is going to be important, depending how you want to navigate that. So if a lot of the workflows and the work processes you're doing right now are HR finance related, I want an HR finance platform. But as soon as you start getting out of the HR finance world and you're integrating to other parts of the business where you might have a differentiation because you're the world's best oil and gas exploration company, or you're the best person at wealth management, identifying clients and building the right pipeline. Okay, that's a different story. Focus on that. Get that custom. Buy the one product that does that better than anyone else. Um, but it's the traditional argument. Platforms always win, suites always win. It's because the cost of integration, the cost of data movement, and the knowledge that you might lose in the AI that's the important piece.

Speaker C: Yeah, I was talking to one CIO and he was saying it's not, um, fun to vibe code your own ERP system. It's just not because they're not an ERP company. Right. And so all the nooks and crannies, all the security privileges, you talked about integrations, these things are hard. Right? So for us, you know, that is our core business, maybe that's a good synergy between two companies.

Speaker A: Oh, just take payroll right there. I mean nobody wants to manage their own payroll. I think brutal. I mean so yeah, I have a

Speaker C: friend who's working at a large bank in IT AI department and he doesn't want to do that. Like it's not, you know, they're a large bank.

Speaker A: Right.

Speaker C: To your point, focus on your customer work and your product, uh, versus back office. Uh, let's talk about data silos. You mentioned that earlier. Right. These things, they could be talking about the data at different levels that can create conflicts. When you think about the enterprise fabric of somebody who was a CIO of a company and they're thinking across both customer G and a their product orgs, all the above. How do they think about a data strategy that makes sense so that it's not just working for today, but it actually scales for tomorrow? Yeah.

Speaker A: So the foundation of all these activities has been getting the right data strategy in place. And the right data strategy means you basically map everything out to understand where your dependencies are, impact of what data needs to be updated, when, what data needs to be secure, when, how is it going to be available, what's the context surrounding on this. And as you start thinking about this data strategy, you start realizing that you have so many data sources you have to manage. And so when you're making a decision and at that point of decision, was it the workflow, was it the business process you're in, was it the location, the security level, the role, um, time of day, maybe it was raining outside, like what's your heart rate? So those contextual and the semantic layer becomes very important. Right. So the data is going to be there, the workflows have to be at the right granularity. And now the question is what data sets do you need to make a decision? Because in the world of agentic, as you know, and you've been in this for a long time, it's about outcomes. It's about end and outcomes on a workflow. So if you don't have an action on the back end or you don't have any kind of like financial metric on the back end, it's what we call vanity metric. A ah, vanity workflow. Like what did you accomplish? There's no value exchange. Why even bother with this? And so once we get the outcomes in place now you can work your way backwards. And so working your way backwards is a very important tooling. As you think about the end to end processes and the dependencies that are there and that drives your Data strategy.

Speaker C: That's super helpful. One thing that I also hear about from customers is, uh, roi. And you touched on it. What is it actually doing? What is it actually automating? Right. ROI is, I think it feels like common sense, but sometimes when you look at these AI products, they don't have an roi. It's kind of like AI for, for AI sake. Right. It feel, it feels like slop. When you talk to executives, how do they calculate roi and then how do they calculate. Well actually that's actually just adding. You mentioned earlier, more cost, but you're keeping all the people, it's just actually more expensive. How do you guide your customers through that conversation?

Speaker A: You're right. So we have a framework called return on transformational ah, impact. And what you get on that return on transformational impact really is that investment process. We break it into seven types of projects. Uh, there's regulatory compliance and those are easy to measure. It's about mitigating a risk of getting a fine, not getting sued. That's very important. There's a second set, that's what we call cost reduction. And cost reduction is the hard cost reduction that you see, not like the soft fake paper cost. It's like how much money did you really take out of a process? So did you have less phone calls in the contact center for self service help and that kind of mitigated that issue, or did you have less error rates, uh, in terms of people coming to you for enrollment of benefits or on the finance side, did you have less missed payments when you did three way matching? So all those things become important. So you get the cost side down. Uh, then we can actually talk about areas around operational efficiency. And the operational efficiency pieces of course are, can I do you know, every dollar I invest here, did I get a return back? Uh, that's your classic ROI metric. But we also have other metrics which are really about revenue and growth. And did you grow the business? Um, was a dollar invested giving you a 3x return or forex return? Right. And those revenue growth initiatives are a little bit different. Where you use AI enablement to drive more sales, you're enabled to drive more customer or employee satisfaction. You were using that to actually get more claims processed or more drive down your accounts receivable days. So your error days went down, your cash on hand went up. Right. Those are the things that you want to start thinking about. Uh, and those are very manageable. Then we get to a little bit more, I'd say squishy, which is talking about you did an AI transformation. You've completely rebuilt the business, you have a brand new business model and then we get to the very, very high end is a brand transformation instead of the business transformation where you're seen as an AI first company.

Speaker C: It's interesting, Ray, as I, as I listen to you talk about that at uh, workday, I feel like we're going through all those ourselves as a large organization. And so it's exciting. Right? I've been here for uh, almost two years now. And even just from when we started to now, we've actually up leveled our thinking to get closer to where you are talking right now. But it wasn't intuitive overnight. The framework you laid out is actually super helpful. Uh, let's assume that our leadership team believes in that framework. How does it change where they invest? How does it change? Like where do you procure technology? When do you think about platforms versus point solutions? How does it actually show up in the buying cycle?

Speaker A: Yeah, so one of the things that we discovered, we had a large conference, um, by means of a number of public CEOs and board of directors. ANIL was there as a Constellation Futures forum. What we learned at that conference was that every CEO is now the chief AI officer, good and bad. They're now involved in every part of that decision. That change happens at the top. You've got to at least try the tools, you've got to at least use them. You have to know what doesn't work. And we're going to say this many times for the next decade, AI expertise is now a commodity. But experience is not your domain knowledge, your understanding of what's failed in the past, your understanding of what works and uh, connecting the dots across industries and processes, that's right here. And that's not going away for a while. Because AI only trains on the success, it doesn't train on failures. And because it only trains on success metrics, it doesn't know what a failure looks like. And so culturally we have to continue to push these systems and find the false positives, the false negatives. We gotta have teams that try to beat the AI. So every time they beat the AI, the AI learns something and we've got to bring that knowledge out. Um, and that's how you make that change. It's a cultural shift that starts from the top and it's also going to come from support from the bottom where people say, you know what? That AI is so off. Oh, why is it so off? Tell me what's wrong. And they'll be like, oh, this thing's missing. This thing's missing, this thing's missing. And this is why it's so important to have FTEs like, around all the time, because that's what they're capturing. FDEs are really like sales consultants, sales engineers meet product managers. And their ability to actually get that back and that feedback loop back into the product and the platforms, that's what allows you to continue to innovate. And so what you want to be able to do is work with a partner that will help you continue to innovate on the product line, especially if you're working in platforms. And that cultural change, not going to underestimate that. That is people just to be comfortable that it's the right decision.

Speaker C: You know, it's interesting. I think the technology is actually easier now than the culture change. Um, you and I both have been doing this long enough, uh, to remember when AI was the hard part. Having to label data, put it into neural networks, train the ML models. My goodness. Uh, now a lot of that's gone away. But now, uh, the people having to change the way they adapt in the culture, that's. That's extremely hard.

Speaker A: Am I going to lose my job? That's the number one question. Am I training the system that's going to replace me? And that's always been the fear. But culturally, you have to set that example. I'll use this because it's, you know, I've said this a couple of times. Public actually have to be careful. Let's do it this way. There's a CEO of a large manufacturing company. I gave the keynote at their opening, um, kickoff to like 30,000 employees. And in that event, the CEO basically said, every one of you will be using AI by the end of the year. And we're like, oh, uh, come on, it's a manufacturing company. How's that going to happen? Well, apparently quality management, there are cameras on every line. And so now you're going to find defects and train defect rates better than anyone else. Global marketing. It was six months to do a global campaign. We can now use AI to get that down to six weeks. Product innovation and simulation. Right.

Speaker C: The.

Speaker A: I have to be careful what I say. The product that they have could put certain amount of items in there, and they have to simulate what they can carry in that item. And can you put more of one thing, less of one thing? Does the design shape change? Well, that process used to take them three months for a new product introduction. Now they can do it in three weeks. And so pretty soon everybody is using parts of AI and understanding how it impacts their life and of course, handling customer service and so support. It's going to be different with agents and more like what you do in terms of the finance side. Right. Instead of you can do a lot of finance automation, you can do before the fraud management now is available. So you can look at detection and that's where you get the excitement. We're going this progression, right? First tell me what the hell's happening. That's great. Then just tell me. Just notifications that make sense. Then you get to the next level. Can you make a recommendation, a suggestion? Then can we automate? And then we suddenly get to prevent prediction and prevention. Right. And that's where you see a lot of interesting opportunities where instead of doing the mundane stuff, you can become more strategic. And that's what we've been talking about is how do you elevate what's happening on the AI side? And more importantly, how do you actually avoid burnout on the employee side as well?

Speaker C: Yeah, I think it's a really good point because I think, you know, a lot of the news media is about job, job loss. Right. But, uh, I don't think that's actually just a, ah, that's too simple. I think jobs will change. To your point. Right. Every part of that company you discussed, uh, there's a change in how you do work. Um, but hopefully the outcome of that is, um, there's less copy and paste. So, you know, for me, I used to be a lawyer. I used to read contracts. My goodness, I tell you, uh, I didn't read every word because I couldn't. It was physically impossible. And so using AI to help me with contracting as an example was a good use of time. Right. So my hope is that with guidance from you and from Workday and from all the folks in industry, that we can actually make a, a positive impact. And AI, I think is a good thing versus just automating copy and paste.

Speaker A: Great point and great way to tie it back. I mean, this copy paste economy is here, but it's not really here. And I think that's what we have to learn is that there's so much nuance in how you make AI work. I ask you a quick question, and it's really about, you're in the middle of all these deployments, like you're the center of the world. This action is happening. What are two or three lessons that clients should be doing before they jump into an engagement?

Speaker C: I think the first thing is they should understand what are they doing today. Business process Wise. There's so many times I've sat in a deployment where I'm watching two different svps or C level organization leaders debate what is the right process. And we as a vendor in the room, helping even be a therapist in some ways, but they don't know where they want to go. And so, uh, it's hard to implement a change when, um, we're the ones having to change the software to match a business process. I think a better way to do it is what should be the ideal business process. And then let's match the software to that. I think that's not the tail wagging the dogs. The first one is make sure you know what you want to get to. And I think that's, that's hard. And consulting partners and our services can help on that. I think the second thing is, um, the rollout of the technology. You know, Ray, you've been doing this a long time. You see these tools, you like them. There are customers of ours where they have hundreds of thousands of employees. They don't want to do anything with AI. They just don't. Right. Some of them have told me they're just waiting to get, uh, their retirement package. And hopefully this doesn't make it there before their retirement package. And so we have to do a better job as an industry getting folks excited because I think if they understand what this can do for them, it can actually make their life better. A lot of jobs out there are really bad. Honestly. If you think about the job of an accountant, the job of a payroll, uh, expert, a lot of it's taking the dying data. And so if you can make that go away, your job is actually better. I think maybe to ask you another question, because I think we were touching on this, uh, AI tools, you know, everyone's looking at it. Is there one question that you would give your CIO friends and customers to ask for their vendors to actually prove out? Is this thing just another feature or is it actually transformative? What's one silver bullet question?

Speaker A: I think the question you should ask is that how often can I reuse that AI tool or AI platform? Uh, we tell clients to focus on highly repetitive tasks, lots of volume, places where there's a lot of coordination, nodes of interaction. So you and I can multitask three things, five things we can't do, 500 things. And then especially think about where you can use that tool to actually create some leverage in terms of digital arbitrage, labor arbitrage, where you can see it's where you have massive scaling Issues that you have to get to. And that's where I tell people to focus, because that tool solve that. You can start stack ranking. Which tools apply to which purpose?

Speaker C: Yeah, I saw an interesting study by Anthropic recently where they talk about the surface area of where AI, uh, can logically automate. You know what I'm talking about? That spider chart. That's cool, right? Um, there's things on there, like legal has been. I think one of the main areas. Computer coding has been a big area. Customer support we touched on earlier. Um, but I think also high on that list was finance and business transformation around the erp, including hr. So, um, these are areas that we know the technology works.

Speaker A: Yeah, no, finance. Hr. Actually, uh, related to that. I finally watch Suits because of Harvey AI. I didn't know it was RV Spector. I'm sitting here like, oh, uh, now this makes sense. I had missed that show. And I was like, okay, this totally makes sense now.

Speaker C: That's funny. I watched it before I went to law school. And then when I went to law school, I realized it was totally different than the show. The show was much cooler. Uh, my job was actually a bunch of copy and paste. And so I said, oh, goodness. There was literally. I was in a law firm. You literally read a contract and you will summarize provisions into a. An Excel spreadsheet just so that somebody else can read. The Excel spreadsheet I created as a. As a younger guy. I mean, this is kind of a weird business model.

Speaker A: Like, my first day of consulting. Like, your experience in law, coming out of getting into your first job and, you know, like, as a lawyer. Um, I was sitting around. Someone's like, what are you doing? I'm like, oh, you know, like, I'm waiting for something to do. I'm on the beach, on the bench, and the guy's like, well, do you want to go to New York? I'm like, of course. He's like, okay, we got a job for you. Just call, get a ticket, get to New York. I'm like, oh, what's going on in New York? Like, no, no, we'll tell you when we get there. And suddenly it's like, you need to get to the 15th floor of this office building, go meet there. Someone's going to hand you an envelope. Okay, what do I do when I get there? You grab the envelope. Okay, what do I do after I get there? He's like, we'll tell you when you get the envelope. I called back. I got the envelope. It's like, what do you do? He's like, you bring the check back, we need to cash it tomorrow.

Speaker C: There's so many stories like that, right? There's like so many stories like that where if you look at what your job is on the outside on LinkedIn, it looks like really, this must be a great firm too, right? I mean, big company, smashy stuff. Um, but when you actually look at the actual job, it's, it's very mundane, especially for folks who are working their way up.

Speaker A: You know, I was non billable, so that didn't help either.

Speaker C: Not billable. I don't like that. I just feel one more for you. And, um, I think the empathy that you and I have for this transformation is real. And so I'm really excited to keep this conversation going. But just one more big one is, um, we talked about transformation data, roi, this, this framework that you've laid out for a legacy enterprise, Right. And, and probably no enterprise identifies as legacy. But for the ones that are maybe not tech product companies, um, how do they go from the copy and paste economy where, you know, you're getting an envelope, I'm reading a contract. These are the same things, or these are things that hopefully AI can do in the future. Um, what's the first step they should take to think about where do they start tomorrow morning?

Speaker A: Yeah, to make it from AI luddits to what we call AI exponentials. A lot of it really starts with understanding what that mundane work is, um, and really where the value is supposed to be. Uh, we've seen a shift in terms of how people are hiring, and we're seeing people hire for more years of experience and less. And then we saw a shift where people are hiring for less experience and more. And so we have the shift that's going back and forth. We're actually seeing a new type of ratio that's occurring. It's like you need one or two people with a lot of experience and you also need a good number of people that actually know how to use the technology. Pairing them together is probably one of the best things you can do because the tribal knowledge and the experience is up here and the appreciation for the tools and what you can't do is at the next level, which is the folks that actually are getting the AI implementations done, that is going to be one of the most important things to be able to do.

Speaker C: Yeah, I think what we're seeing is sometimes when we're building the products, let's say we're doing a payroll product, we have an AI engineer and AI uh, product manager who loves, you know, all the AI tooling. But we pair them with somebody who's got 20 years of experience actually running payroll. Yes. They get in a room and like don't come down and tell on the whiteboard you guys agree it was actually valuable. And you know, when we watched that happen, we watched the AI team really learn a lot about payroll in Europe. And we hear the payroll person say, well I didn't realize that technologies actually come so far. But that uh, cross disciplinary you mentioned FTEs earlier for Deploy engineers, getting these types of folks in the customers actually, uh, sitting with the teams to the implementation, uh, I think is the only way. It's throwing over software over the wall. It's not going to work.

Speaker A: What's more fun than payroll in Europe? Tax rules in Poland and Brazil.

Speaker C: 100% listening to the show. Yeah, but those are the things you mentioned earlier. Deterministic versus probabilistic. You need to know the domain knowledge. So from an engineering perspective, you know, where do I build in hard rules and where do I let the AI, uh, cook a little bit, so to speak. And so uh, Ray, those are the questions that we had. Anything you wanted to close on with audience things.

Speaker A: Don't be afraid. Get started now. Now is the right time to get to know what's going on. But you're right, Jerry. Know your business processes inside and out. If you don't know your business processes, don't even get started. Like you're in the wrong game. Figure out how it all works and then figure out how it all really works. Because there's a lot of band aids going around internally that are not documented. And that's why it's so important to have everybody sit there in the workshops that you guys are talking about. I think the last piece of is, you know, here's the reality. We don't have enough workers to do the work that's out there. And even with AI, it's still a long road. I don't think you're going to lose your jobs for a lot of the reasons of the job being replaced. I think what you're going to have to do is know where you're going to use AI to do your jobs better and more importantly where AI is going to take the job into the future. It's not like a zero sum game. We're actually creating more jobs because of AI. We might need fact checkers for AI that are here human to check the AI, uh, and things like that. So you're going to see a law of unintended consequences. A brand new set of, you know, jobs that are out there, but if you don't try it, you're not going to be able to play in that game.

Speaker C: Yeah. And trying it could be as small as, you know, I'll give you an example from home. My wife and I don't love cooking, uh, to be honest. And so what we do is we ask ChatGPT, give us a recipe book and then hook it up to Instacart and, you know, help me make the order and the food shows up and then we cook it when it's here. Right. So, so it could be small, things like that, just to get your feet wet.

Speaker A: That is actually really cool. Yeah.

Speaker C: I give them to you for free

Speaker A: menu creation with order optimization, and then you got to get someone to come and cook it for you. See, now that would take it to the next level.

Speaker C: That's the next step. And I hope, hopefully our friends working in robotics are helping on that. Ray, thanks for helping us cut through the noise today. I mean, there's just so much that's happening out there and your. Your expert walked experience is so super interesting. And thanks to everyone for listening. If you enjoyed today's podcast, please subscribe to the Future Work podcast. Wherever you get your shows.

Speaker A: Uh,