About This Episode
Frank Cottle sits down with Dave Wright, Chief Innovation Officer at ServiceNow and co-author of Infinite: How Visionary Leaders Transform Today’s Businesses into AI-Forward Companies, to dig into why so many companies are pouring money into AI and getting almost nothing back. Dave’s argument: most companies aren’t failing at AI because of the technology — they’re failing because they’re using it to automate the same broken workflows they already had, instead of rethinking the workflow itself. Frank pushes back with his own company’s playbook: rewriting 130 job descriptions from scratch, doubling productivity with the same headcount, and giving guardrails to the people instead of the AI. The two also get into revenue-per-employee at AI-native companies, Ford’s very public AI-quality-control stumble, why five generations of workers are adopting AI so differently, and whether “human-in-the-loop” is already outdated thinking.
Key Takeaways
- Using AI to simply do faster what you already did produces cost, not ROI — the ROI shows up when AI changes the workflow itself, not just the task.
- AI strategy should follow business strategy, not lead it: define what the company is trying to become first, then figure out what AI frees up to fund it.
- Frank’s company rewrote all 130 of its job descriptions around AI-augmented decision-making, then reported roughly 20-25% revenue growth in a year while adding only two headcount (per Frank’s own account — see fact-check register for sourcing caveat).
- Ford rehired roughly 300-350 veteran engineers in recent years after AI-driven quality-assurance tools underperformed — a widely reported example of AI removing human judgment from a process that still needed it.
- AI-native companies are posting revenue-per-employee figures in the millions of dollars, compared to a couple hundred thousand at a well-run traditional enterprise — a gap public reporting on companies like Anthropic supports directionally.
- Dave’s take: guardrails belong in the workflow before AI is deployed, not bolted on after; Frank’s company instead puts the guardrails in the job description and lets the tools serve it.
Why This Matters
Most “AI strategy” conversations right now start with the tool and work backward to a use case, which is exactly the trap both guests describe. The distinction Dave draws between AI-enabled, AI-embedded, AI-first, AI-native, and AI-driven companies isn’t just branding; it maps to how much of the organization’s actual structure — job descriptions, decision rights, workflows, governance — has been rebuilt around AI rather than layered with it. Ford’s reversal is a useful real-world check on the hype: rehiring hundreds of experienced engineers is a costly, public admission that automating a process without preserving its judgment layer doesn’t work. And the generational split Dave describes (Gen Z treating AI with more skepticism about its limits than millennials do) is a live workforce-design question for any company betting its AI rollout on younger hires adopting it uncritically.
Advertisements
Introduction & Welcome
Dave Wright [00:00]if you look historically at how big enterprises were, I think people were always touting those figures “Hey you know we’ve got 100,000 people 110,000 120,000.” I don’t think you’ll see another company get to 100,000 people I I don’t think there’ll be the need for it anymore
Frank Cottle [14:10]: Dave, welcome to Future of Work podcast. Gosh, I can’t tell you how excited I am. Your background, the knowledge you’ve got, the things you’ve done — if you don’t deliver to our audience today, I’m going to be shocked. We’re expecting a lot. So thank you very much for joining us.
Dave Wright [34:13]: I will do my best to keep up with your expectations, Frank. Thanks for having me on.
Frank Cottle [38:29]: Well, our expectations are not unfounded, considering your background. So thank you very much for joining us.
Why AI ROI Is Failing: The Automation Trap
Frank Cottle [47:26]: We hear every day about companies heavily investing in AI, but reporting very little measurable return on investment. You’ve just written a really interesting book, and you argue it’s not a technology issue — it’s a structural or implementation issue. Can you explain that, and give us some guidance on what works and what doesn’t in your view?
Dave Wright [01:15]: Yeah, absolutely. This probably goes back to our history — we’ve both been involved in automation and artificial intelligence for a long time. I think the biggest problem people make is using AI to do the wrong thing. They use it to automate the things they can already do — take a task they already perform and ask, “how can I do more of it, how can I do it faster?” But they don’t make any other change apart from that. They’ve paid for the AI, they’re doing what they were doing before, but all they’re doing is incurring cost. You get much… go on
Frank Cottle [01:53]: Shouldn’t they have automated their processes already, in advance of looking at AI?
Dave Wright [02:09]: You should do, but most people — I think this is where some struggle with the ROI, because a lot of people I see are treating it as RPA 2.0: just using it to do basic automation or integration you could have done with APIs or RPA. That’s where it’s an expensive solution for automation. It’s a fantastic solution if you use it to do things you couldn’t do before — but if you just use it to automate the past, you’re not inventing the future. You’re not going to get much ROI from that.
Frank Cottle [02:29]: It’s funny — because they should have automated before, but that’s not very interesting, not very exciting. And driving new processes and procedures through companies always gets a lot of kickback. Do you think AI is exciting enough, and offers enough opportunity, that it’s actually causing people to do the basics of what they should have done already? And if they do that, do you think they’ll go another step, or do you thin it’ll stop there?
Dave Wright [02:55]: what it reminds me of digital transformation — when everyone did “digital transformation,” it was a lot of digital and not much transformation. It feels the same now: “we need to deploy AI” becomes the strategy, rather than “we’ve got to have a bigger strategy than that.” You’ve got to be trying to achieve something more from a business perspective than just implementing AI as a tool. You’ve got to think about changing everything — organizational structure, workflows, what you measure, your governance capabilities. Everything has to change for it to be truly effective, or for you to maximize how effective it is.
AI Strategy Should Follow Business Strategy, Not Lead It
Frank Cottle [03:42]: If I want to change everything — which is a challenge — where do I start, to prepare for AI?
Dave Wright [03:52]: When you’re thinking about going on this journey, the way I always describe it to people is: tell me what your business strategy is. What do you want to achieve as a business? Your AI strategy shouldn’t be about AI — it should be about the company. Say you’re making automobile parts, and you want to get into aeronautical engineering — making parts for planes. If that’s your business strategy, your AI strategy is how you automate enough process to free up enough time and resources to fund that true business initiative. AI should be an enabler, not the strategy — you should be trying to achieve something else. So first you define the strategy, then you work backward: what do we need from a resource perspective to do that? Then, how do I choose a function to apply AI to, and how do I choose the workflows within that function that matter most? I encourage everyone, when they first go on this journey, to choose one workflow — and make sure it crosses multiple silos within the business.
Choosing Your First Workflow
Dave Wright [05:13]: Make sure it’s important. Don’t apply AI to a mundane task no one cares about — do something highly visible, like employee onboarding or your supply chain. Something people will see and understand the challenges of, but that lets you learn how to think about those workflows and processes. The history of workflow design — which has been my area for a long time — has been about: how do I define a workflow that optimizes how work is done, and makes it repeatable, based on a limited number of resources? I define workflows around scarcity, and that scarcity is people — how do I optimize my people? So what happens when you go from scarcity to abundance, and you have an almost infinite capacity to execute that workflow? The chances that workflow was designed to work optimally in this new world are almost zero, because it’s the opposite of what it you’ve actually designed the work for originally.
Frank’s Approach: Rewriting 130 Job Descriptions
Frank Cottle [06:32]: The issue then is where to start, and I agree — start with something visible. In our own company, we talk about companies as AI-enabled versus AI-embedded, and we use the term “AI-driven” for ourselves — maybe a blend of the two, or maybe we’ve whipped it further than embedded, I don’t know. Strategically, we looked at what was going on in the world and the change AI could bring. We saw a lot of people taking the mundane tasks and recognizing they could let people go by automating them — automate them and never hire those people in the first place. That’s not us. That doesn’t fit our company values, which are about supporting our people. So we rewrote every job description in the company — 130 different job descriptions — and then found the tools to let people fulfill that job description. My mandate was: I want to see a band of angels on everybody’s shoulder. Our strategic view was, I want to double our productivity with the same resources, and then give the staff a raise — reward that increased productivity. We had about 20-25% growth last year and hired two people.
That was it. So our approach was non-technical, a little strategic like you say, but it drilled down not into “what could you do with AI” but “using AI, how could you change the job descriptions themselves” — and then enable people with the right AI and other tools. We were already very process-driven. Overall, we felt it was really a people thing rather than a structural or technology thing. I’m 76 years old — I’ve been through every tech change since the invention of steam — and it’s always about change. This is a change management project to me, and it’s always about the people. You’re just applying another tool. Am I overly simplifying it?
Dave Wright [09:26]: I just want to ask a quick question based on that. When you created that hybrid environment — a mix of AI and people — did you tend to use AI to augment and automate tasks, or were you trying to use AI to do a complete job role?
Augmenting Decisions vs. Automating Tasks
Frank Cottle [09:55]: More of the latter than the former, because almost all of our tasks had already been automated, or close to it. What we hadn’t done was apply enough intelligence in that automation to let the person fulfilling the task make a higher level of decisions — and that’s what we wanted to do. We felt it’s not the big that beat the small, it’s the fast that beat the slow, and speed is dependent on decision-making, not processing, in corporate growth — at least that’s my belief. So we wanted to augment each role so it could make higher-level decisions, so everything could move faster.
The 600 Heads Thought Experiment
Dave Wright [10:55]: That’s a good approach. There’s a statistic I use a lot with CEOs to explain the benefits of AI. If you run a 10,000-person organization and automate 30 minutes of work for everyone, and you do nothing else — no reorganizing — then all that happens is people take longer coffee breaks.
Frank Cottle [11:17]: Exactly. Though we do have to serve better coffee these days too, so that encourages it as well.
Dave Wright [11:26]: I used to be all about the coffee and the foosball tables — that’s changed. If you take that 10,000-person company, half an hour per person, and concatenate that time — that would mean roughly one fewer person needed for every 16 people, freeing up one person’s time during the day. Concatenated across 10,000 people, 30 minutes each gives you 600 heads. If I went to a CEO and said, “I’ll give you 600 heads for free — what would you do with them?” — that’s when they start to understand what the AI strategy is really about. Do I want to open more buildings? More service offerings? Move into new countries or regions? These are things people would then have the capability to do — but you have to take that step and think about what you’re going to do with the resources, before you go blindly deploy it.
Frank Cottle [12:27]: I agree with that — our decision was: double productivity with the same number of people. That’s our simple goal. Reward the people if we accomplish it. People had to shift positions, we rewrote things, we moved people from one department to the next — but basically we shuffled—
Dave Wright [12:51]: You didn’t cut.
Frank Cottle [12:52]: Right — and that’s very important to the well-being of the individuals, but it’s massively important to the esprit de corps of the whole company. When people know you’re making a change for their good as much as the company’s good, they get behind it.
Small Companies vs. Big Companies: Who Really Benefits
Frank Cottle [13:14]: Agreed. One question — when you mention a 10,000-person company, it’s really hard to say “I’m going to save half an hour per person.” But it’s very easy to look at a hundred-person company and say “I’m going to save eight hours a week per person.” So is AI equally accessible and adoptable for smaller companies, where they might get larger relative benefits than what we see in published data from large public companies? We hear about big companies laying people off and spending all this money — but they only employ maybe 20-30% of the employees out there. The masses are employed by smaller companies. Is AI equally accessible and being equally utilized by them?
Revenue-Per-Employee and the End of the 100,000-Person Company
Dave Wright [14:06]: Absolutely. One of the things you see now — we were talking before about the different types of AI companies and hoe people define themselves as to where they are AI-enabled, AI-embedded but obviously you have AI-first, AI-native. And one of the things is…
Frank Cottle [14:24]:AI-driven.
Dave Wright [14:27]:When you look at those companies — born using AI from inception — their revenue-per-employee figures are huge compared to traditional organizations. You’ll see organizations with RPEs of six, seven, eight million dollars per employee, versus a traditional enterprise doing well at a couple hundred thousand. So there’s a trend of companies saying, “we don’t need to grow at the same rate we’re growing.” Some companies have taken the attitude of just wanting to cut costs and reduce headcount. But the more creative ones are taking the attitude you described — not cutting heads, but growing the company without growing headcount at the same rate. As revenue grows and employee count doesn’t grow as fast, your RPE figure grows. Historically, people touted “we’ve got a hundred thousand, a hundred and ten thousand, a hundred and twenty thousand employees” — I don’t think you’ll see another company get to a hundred thousand people. I don’t think there’ll be the need anymore.
Frank Cottle [15:56]: I absolutely think you’re right. If you look at history — the invention of steam created factories and railroads. The first billion-dollar company had around 20,000 people in it — the first billion-dollar-revenue company. The last billion-dollar-revenue company I saw, I think, had seven people in it. Pretty soon, you and I could go start a company tomorrow with two people. We can do that. And it’s not just AI, and not just technology — it’s our ability to create, communicate, and distribute, technologically and through every process out there today — supply chain management, intermodal systems, everything we can use to create almost any style of company. One of the issues is that when we talk about these big companies with huge revenue-per-employee, they’re mostly tech companies. You don’t see Ford Motor increasing their revenue per employee hugely, the way a two-person, billion-dollar-a-year company does.
The Ford Story: When AI Removes Human Judgment
Dave Wright [17:21]: Just cause you mentioned Ford — they actually have a great AI story. They let go 350 quality engineers and replaced them with AI to do quality assurance, but ended up rehiring them after customer satisfaction suffered.
Frank Cottle [17:41]: Because they hired something to assist with quality assurance without decision-making — and that’s the key. I think AI’s purpose is to allow faster, better decision-making. Everybody in the world makes about ten decisions a day, and probably four, maybe six or seven, aren’t the right decision. But we let them run — we defend them, we let them run long, and it takes a week to turn around on something. If you could make just one of those ten decisions right instead of wrong, and cut one of those bad decisions short the next day instead of letting it run for two weeks, two months, or two years — the difference in your company’s optimization is staggering. To me, that’s the purpose of AI: to help us make decisions.
Tribal Knowledge and the Fear of Questioning AI
Dave Wright [18:47]: The other thing that comes out of digitizing workflows and applying AI to them — Ford’s probably a good example here too — is you become aware of how many decisions are made based on tribal knowledge rather than based on documented facts. That’s a challenge…
Frank Cottle [19:09]: Yep. The old “what are the seven most feared words in business? Because that’s the way we’ve always done it.” I think there’s a lot of that. And it speaks to part of the challenge — in today’s workforce, we have five generations of workers. You represent a generation younger than me. There are two, three, or four generations below you, too. And each generation has different values, different decision-making processes, and different things they want from their careers. How is AI going to contribute to that, rather than frustrate it?
Five Generations, One Workforce: How Different Generations Use AI
Dave Wright [19:56]: It’s interesting — my oldest daughter is 20 and recently did an internship. I asked her takeaways, and she said the millennials’ use of AI is unbelievable — people use it for everything, whether they need to or not. I think Gen Z and Gen Alpha have a much more respectful approach: they understand when to use it, but more importantly, when not to. My kids appreciate what it can’t do — they know there’s no point asking it for style advice, because it has no taste, and no point asking it for emotional support, because it doesn’t truly understand what you’re feeling.
Frank Cottle [21:05]: And yet there are AI therapists out there.
Dave Wright [21:08]: There are, but I’d guess they’re not being used by anyone of that generation. Some people are just more relaxed talking to a machine than to people; some, oddly, think it’s more private — we can question whether that’s true — but the end result is it mirrors so much of what you feed it. I’ve done this with my kids: the same prompt, fed into their ChatGPT and mine, gives radically different answers. Totally different.
Frank Cottle [21:48]: Totally different. One of the things I don’t think people don’t fully realize is that AI does gain a perspective — like the algorithm on Facebook: you look at one ad, it feeds you more like it. AI does the same thing. The algorithms haven’t been taught yet, and maybe don’t have the data yet, to create a truly independent perspective. One thing we found — I come from a generation where if you had imaginary friends, you were probably going to get committed. But today, we actually encourage the people using AI — those “angels on their shoulders” — to name them, because if they can identify with them personally, they’ll be much more effective. Everybody has their George, Sally, Harry, or whatever they’ve named their AI, and they interface with them actively rather than just passively querying. I find that interaction has actually helped a lot.
Naming Your AI: Trust, Identity, and Adoption
Dave Wright [23:05]: I’m sure Anthropic didn’t accidentally call it Claude by mistake — I’m sure there was a lot of research into that.
Frank Cottle [23:11]: You know, we’re playing with that — how does one redesign work in a workforce around AI so people are creating trust and faith in decision-making, rather than just having automated automation?
Dave Wright [23:32]: I think this is key to getting AI adopted, and to keeping the culture of the company stable. First, you have to be transparent: “we’re going to deploy artificial intelligence, and this is why, this is what we want to achieve” — hopefully tied to that bigger business story. And this needs to come from the top down, because — as you said at the start — there are going to be changes that reorganize the company. If I automate a number of jobs in sales, those heads don’t necessarily stay in sales — I might move them to marketing or engineering. People need to be aware that’s going to happen. But you also need an open framework of trust where people can question things.
Frank Cottle [24:30]: But wouldn’t you need to do that no matter what?
Dave Wright [24:34]: You would, but I think AI is a specific use case, because it suffers from Emperor’s New Clothes syndrome — no one wants to say “hang on, that sucks, that isn’t helping with this job,” because questioning AI in any way makes you look like a doomer, or like you don’t want to move with the times. But there are times AI gets deployed and it isn’t producing as good results as before, and you need an environment where subject-matter experts can say “this isn’t helping” or “we’re doing this the wrong way” — or that instead of using AI on the existing process, you need to rethink the process itself, rather than layer AI onto what’s already there. That’s why you saw company leaders recently saying “everyone has to use AI, we’re an AI company, let’s get out there and use it” — and six months later, “my god, these tokens cost a fortune, why is everyone randomly using tokens?” We went through the whole token-maxing era.
Frank Cottle [25:50]: Build your own LLM and raid your own database — that’s a solution, and it can be managed pretty easily these days.
Dave Wright [26:02]: So when you talk about trust and transparency — and yes, it should always be that way in a company — you’re now introducing an element people aren’t used to. Parts of the workforce can’t see it, aren’t sure of it — it’s not like a person whose personality you can judge. That causes a natural unease in people. So that openness, trust, and transparency becomes even more important.
Guardrails: Should They Constrain the AI or the People?
Frank Cottle [26:30]: That’s a good thing — guidelines, guardrails, maybe a cage at times — that’s part of the theme of your book, Infinite. When does a company — or even an agency or government — need to put guardrails in place? Let’s stick with companies, to depoliticize it a bit. When is the case for guardrails, and what sort should a company put in place so it’s not AI gone wild — doesn’t look like a bad Spring Break movie?
Dave Wright [27:17]: [laughs] Now I’ve got “AI gone wild” stuck in my head. The honest answer — and it’s hard for people to accept this — is you should put guardrails in place before you deploy the AI. We’re in a kind of transition phase where people implemented AI and are now thinking, “okay, now we need to apply governance to it.” If you were doing this correctly, you’d apply governance before implementation: define what you’ll allow AI to do, what you’ll allow it to do with human intervention, and what you’ll never allow it to do. Then look at what areas to deploy in, what model and platform and harness to use, how you’ll measure ROI, and what ROI you want. You should have all those metrics in place before you even start planning what you’re going to do.
Governance Before Deployment, Not After
Frank Cottle [28:22]: It’s funny you say that — we reversed it a bit. That’s why we rewrote the job descriptions: the job descriptions carry the guidelines and guardrails, and then the tools work within those guardrails. We felt the people needed the guardrail, not the AI.
Dave Wright [28:42]: So they understood how they could use it.
Frank Cottle [28:45]: Right. If I give somebody a paintbrush, I might say, “stay in an impressionist style” — but I won’t tell them what color or subject matter. People then make their own decisions about how to use their tools to accomplish the goals already set out — the transparency, the goal, the strategy everyone’s working toward as a company. If they do that, we’ve found that few people break the guardrails, but they usually ask first: “can I run this experiment, can I have a separate sandbox to try something?” We’ve found that experimentation outside the lines can be very beneficial — some of our best creativity has come from that — but we’ve never broken the train running down those guardrails. It’s a people-first philosophy. Technology, to us, is still just a tool — we don’t minimize its power, but it’s just a steam engine, an electric motor, a laser. It’s just a tool.
Human-in-the-Loop vs. Human-on-the-Loop
Dave Wright [30:35]: I think we’re still going through the phase where, from an enterprise perspective, people don’t want to fully hand the reins to AI, because they haven’t defined all the guardrails yet. If you define a task for AI without giving it guardrails for how it does it, the risk is like building a self-driving car and saying “get from LA to San Diego” without specifying it has to stay on the freeway — it might just try to go in a straight line. Same with agents: a security agent detecting a vulnerability might think the immediate response is to close down all network connections to every server.
Frank Cottle [31:25]: But isn’t that the guardrail — isn’t the guardrail the person, not the technology?
Dave Wright [31:29]: It depends what route you’re on. If you’re going for a true agentic workforce, where you’re nondeterministic in outcome and just define what you want done — that’s where everyone would like to get to, that’s where people think the optimization will be, but that’s a scary red button to press. I think in the short term people still want — perhaps not human-in-the-loop, but human-on-the-loop — where people retain control and a decision on what they’ll allow AI to do. I think that’s the safety valve we need right now.
Lessons from Self-Driving Cars
Frank Cottle [32:07]: I see it as allowing human decisions to be made faster, not made for us. Using your car example — I bought a new car recently with automatic steering and so on, and when I started using it I thought, “wait, I’m not in control here.” But after adjusting the settings, after a couple of hours, I thought, this is pretty cool — it actually does what it’s supposed to do. I don’t know whether it’s better than I am until one of us crashes. Statistically, it’s supposed to be way better than me — but is the data deep enough yet? Not yet.
Dave Wright [32:59]: The car example is interesting — when I first got a self-driving car, I felt there was a design flaw: when there was a car in front, it started braking way too late for my taste. I thought, if I were designing a self-driving car — not that I have the skill — I wouldn’t let it fully engage for the first week, until it understood how I like to drive, what I’m familiar with, and could emulate that a bit more — assuming you’re not a lunatic when you drive.
Frank Cottle [33:37]: Having a self-driving car you can adjust settings on — you don’t really train it, you adjust it — I think is important to that comfort level.
The One Thing to Do Thursday Morning
Frank Cottle [33:53]: Let’s move along — we’re starting to run long. For people listening today: what’s the single step — it’s Wednesday — what do they need to do Thursday morning when they walk into the office? What’s the one thing to change in how they approach AI?
Dave Wright [34:21]: I think we’ve covered a lot of it already, but I’d say: the first thing is get everyone together — everyone on that core management team.
Frank Cottle [34:34]: Strategic.
Dave Wright [34:35]: Strategic, yeah — because what you’re doing isn’t deploying a piece of technology, you’re defining what the future of the company is going to be. You’re going to be thinking about organizational design, business model design, revenue flow through the company. You need everyone on board, everyone knowing what you’re doing from a leadership perspective. Then be honest with that strategy, and make sure the company knows why you’re doing it. But you have to start somewhere — look at where you feel the most pain in the company, or where you think you’ll gain the most benefit, and choose a workflow in that region to start automating. One of the interesting things this gives you is the best free audit you’ll ever get — there are so many processes where people know the workarounds, know what they need to ask whom, or know that if they leave something till Wednesday it’ll go through quicker. AI knows none of that. AI will stop when it finds a break in the process — and when you remove that bottleneck, it shows you the next one. That exercise of actually thinking through how you do things is probably the most valuable thing a company can do.
Frank Cottle [35:53]: So: strategic, transparency, and then commitment. That’s a great summary. And the commitment is initially experimental—
Dave Wright [36:14]: Experimentation with purpose.
Frank Cottle [36:16]: With purpose, right — that lets you learn the processes you need to expand.
Dave Wright [36:19]: And once you’ve established that, you need to look at how you manage demand — now that you understand the process, how do you manage the demands and priorities of where you apply this across the business?
Closing
Frank Cottle [36:31]: We’re having a strategic meeting in early November — do you want to stop by? You could help us out.
Dave Wright [36:38]: I could, depending where you are and what day it is.
Frank Cottle [36:42]: I’m really grateful for the time you’ve shared today, and the tremendous knowledge you have. I’d encourage everybody to take a look at your latest book, Infinite.
Dave Wright [36:56]: Thank you very much.
Frank Cottle [36:57]: I really would — there’s a huge amount of value there, no matter where you are along the path. So thank you for that.
Dave Wright [37:13]: Well, thank you for having me on, Frank — thanks for a great conversation.
Frank Cottle [37:16]: I’ll look forward to the next time. Take care.
Dave Wright [37:19]: Alright. Bye now.
Advertisements
