Here’s a number that should make every CTO pause: global enterprise AI investment has crossed $400 billion.

Here’s the one that should keep them up at night: fewer than 10% of enterprises report measurable ROI, according to Draup.

So where’s the money going? Not down a technology drain. The tools work. The models are improving quarter over quarter. The problem is somewhere most leadership teams aren’t looking: inside the org chart itself.

The hiring is happening. The redesign isn’t.

Draup’s Mapping Tech Skills Across Software databook tracked a striking acceleration: AI-related software job postings grew roughly 50% in the U.S. from Q3 2023 to Q2 2025. AI exposure in software roles climbed from 14.3% to 21.3% in that same window. Seventy percent of enterprises have adopted generative AI in at least one function. AI engineering is now the number-one skill priority for 74% of U.S. engineering leaders.

Companies are clearly spending on AI talent. What they’re not doing is rethinking how work actually flows once that talent is in the building. When you automate a task, you don’t eliminate the work around it — you redistribute it. Oversight grows. Validation queues get longer. Exception handling lands on senior people who weren’t budgeted for it. And none of that shows up in anyone’s efficiency dashboard.

“Enterprises are learning that AI performance is directly tied to the quality of human guidance around it. Human-guided AI consistently outperforms AI-only systems, which means demand for skilled oversight isn’t a transitional cost — it’s a structural one.”
— Vijay Swaminathan, CEO & Co-Founder, Draup

The costs hiding in plain sight

Most AI business cases are built on a comforting fiction: that automated work is deleted work. The ROI model says “200 hours saved.” What it doesn’t say is that 80 of those hours reappeared somewhere else; in prompt refinement cycles, in human review of AI-generated outputs, in judgment calls the model can’t make. Research from Workday puts a number on it: enterprises are losing nearly 40% of expected productivity gains to employees fixing low-quality AI outputs.

And the people doing that fixing aren’t cheap. AI engineering salaries in North America grew 56% between 2023 and 2025, according to Draup’s data. At the same time, roles are converging fast — developers are embedding ML models, data scientists are managing deployment pipelines, IT teams are orchestrating cloud-native AI infrastructure. Two years ago, most of these responsibilities didn’t exist in anyone’s job description. Today they’re table stakes, and they’re expensive.

The reskilling math most leaders haven’t done

There’s a lever here that most companies are underleveraging. Proprietary Draup data shows that internally reskilled employees are roughly 50% more likely to stay beyond 18 months compared to external hires. That’s a soft HR metric but it’s also a compounding advantage in recruiting costs, ramp time, and institutional knowledge.

But the industry response so far has been lopsided. Deloitte’s 2026 State of AI in the Enterprise report found that 53% of companies are investing in AI fluency training — but only 33% are redesigning career paths, and just 30% are restructuring their orgs around AI. That gap matters. Teaching someone to use an AI tool is not the same thing as redesigning their role so the tool actually changes the economics of their work.

The uncomfortable truth is that most AI ROI projections are built on best-case assumptions that collapse on contact with reality. The enterprises getting this right aren’t just buying better models, but they’re rebuilding how work gets done, who does it, and how they measure whether it’s working. The technology is the easy part. The workforce equation is where the returns actually live.

Read the full Forbes article:
The Hidden Costs That Are Undermining Enterprise AI ROI

Checkout the full Databook:
Mapping Tech Skills Across Software, IT, and Data Science Functions

‍