# Engineering Evolved. The Rest Didn't. That's Why Your AI ROI Sucks. > Companies keep complaining that AI isn't paying off. Engineering rebuilt how it works around AI, while product, sales and marketing kept working the way they did in 2015. That's where the ROI goes missing. Karim Traiaia · 2026-10-26 · https://kerno.io/blog/why-your-ai-roi-sucks/ In 1881, New York and London both had electric generating stations. Factories started replacing their big steam engines with electric dynamos. And for the next few decades, productivity barely moved. The trouble was everything around the dynamo. Factories of that era were multi-story buildings arranged around one central power source, with belts and shafts driving every machine at the same speed. Owners swapped the steam engine for a dynamo and kept everything else exactly the same. The gains came only when factories were redesigned around what electricity made possible. Single-floor plants laid out around the flow of work, with a small motor on each machine that the worker could start and stop. Workers needed new training and took on more responsibility. In the 1920s, about four decades after electricity was commercialized, US manufacturing productivity grew more than 5% a year.[1] The economist Paul David made this comparison in 1990, between the dynamo and the computer.[2] I think AI is following the same pattern, only faster. Every few weeks a new study lands saying companies aren't getting ROI from AI. It gets shared around, executives nod along, and someone repeats it in the next board meeting. In McKinsey's 2026 State of AI survey, 80% of respondents said AI made them more productive. Only 37% said it had contributed to their company's earnings, the same share as the year before.[3] In PwC's 2026 survey of 4,454 CEOs, 56% said they've seen neither higher revenue nor lower costs from AI. Only one in eight said it had delivered both cost and revenue gains.[4]
Bar chart from McKinsey's 2026 State of AI survey. 80% of respondents say AI improved their productivity, while 37% say AI contributed to their company's EBIT.
What I disagree with is the conclusion people draw from them. When you dig into these companies, the failure usually sits in the parts of the business that never changed. ## Follow the tokens Start with where the money goes. Menlo Ventures estimates that of everything companies spent on AI tools for specific teams in 2025, coding took $4.0B, more than half.[5] No other function comes close.
Bar chart of 2025 departmental AI spend from Menlo Ventures. Coding 55%, IT 10%, marketing 9%, customer success 9%, design 7%, HR 5%.
That spending changed how engineering works. Engineers today can build features, fix issues and handle almost any customer request faster and better than ever before. Engineering teams are redesigning their processes around it, from how they plan to how they review and test. They've adapted. ## Meanwhile, down the hall Now walk over to product, strategy and go-to-market. Most marketing and business people I know use AI in a pretty basic way. They write emails with it. They prepare presentations with it. That's the smaller problem. The bigger one is that these functions never rethought their job for a world where engineering can ship this fast. Try asking them about their marketing or go-to-market strategy, and you get the same answer you'd have gotten in 2015. Engineering picked up capabilities that would have sounded crazy two years ago. The rest of the company went stale and changed nothing to cash in on them. PwC's data shows how lopsided this is. Only 22% of CEOs say they apply AI to a large extent in demand generation, and 19% in their products, services and customer experiences. The companies seeing both revenue and cost gains look different. 44% of them apply AI to their products, services and experiences, compared with 17% of everyone else.[4]
22%of CEOs apply AI to a large extent in demand generation
19%apply it to their products, services and customer experiences
44%of companies seeing both revenue and cost gains apply it to their products
So you end up with two halves of the same company living in different eras. One half rebuilt how it works around AI. The other half added a writing assistant. Engineering got its dynamo. The rest of the factory still runs on belts. ## Shipped, and then nothing Here's what happens next. Engineering spends heavily on AI and ships a new feature. The feature leaves engineering. And that's it. Nobody builds a campaign around it. Sales doesn't pitch it. Customer success doesn't take it to the accounts that asked for something like it. Nobody goes back to existing customers to upsell them. Nobody asks whether it opens a new market or changes the pricing. Then the quarterly numbers come in, revenue looks the same, and someone asks why the company is spending so much on AI for so little return. Of course there's no return. Nobody did anything with what engineering built. ## The bottleneck moved Operations people have a name for this. Eliyahu Goldratt's theory of constraints says a system moves only as fast as its slowest step, and speeding up any other step changes nothing.[6] For years, engineering was the slow step. AI sped it up. Now the slow step is everything that happens after a feature ships. It's the factory all over again. The investment worked. The rest of the system was never built to use it. ## What changing actually looks like Getting ROI from AI means rebuilding the rest of the company around the fact that engineering can now ship much faster. That starts with the teams who talk to customers. Deciding what to build is their job. Sales, product and customer success should be working out what customers will pay for, getting it to engineering quickly, and planning around how much more engineering can now deliver. When you can ship faster, a lot opens up: - Sales and customer success can take a customer's request, get it built, and turn it into an upsell quicker. - Product and strategy can test adjacent markets that used to cost too much engineering time to even try. - Marketing and product can run more experiments. The question for every department is the one engineering already answered. If we designed how we work today, knowing what AI can do, what would we do differently? McKinsey sees this in the companies getting real returns. Its high performers "fundamentally redesign workflows that are enabled by AI rather than insert AI into existing ones."[3] ## The fair objections > "Shipping more features doesn't mean shipping the right ones." I agree, and that's exactly the point. Choosing the right features depends on the people closest to customers. If sales, product and customer success have no strategy for figuring out what people will pay for, engineering ends up building random features, because suddenly they have time on their hands. That's an organizational problem, and you fix it in the organization. > "AI-written code creates more review and testing work." Also true. That's an engineering problem, and engineers are on it, with a lot of work and new tooling going into better review and testing. > "Some engineering teams just burn tokens." Yes, some engineering orgs use AI inefficiently and spend far more than they need to. That's a real cost. I don't think it's the main reason companies are complaining. ## Before the next ROI complaint If AI only made it to engineering, of course you don't see ROI. Your engineers are spinning their wheels and burning tokens while the rest of the company talks to customers the way it did before any of this existed. You aren't closing more deals. You aren't upselling customers on what's new. So before you forward the next study about AI failing to pay off, ask a different question. Engineering changed how it works. What has everyone else changed? ## Sources
  1. History News Network, Paul David on the decades-long economic impact of electricity
  2. Paul A. David, "The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox," American Economic Review 80(2), 1990. Summary
  3. McKinsey, The State of AI, Global Survey 2026
  4. PwC, 29th Annual Global CEO Survey, 2026 (PDF)
  5. Menlo Ventures, 2025: The State of Generative AI in the Enterprise
  6. Eliyahu M. Goldratt, The Goal, 1984