The 90% You Skipped: Why Most AI Investments Don’t Deliver

BCG research says 70% of AI deployment success comes from people, process, and context. Sauder says only 25% of leaders have a comprehensive AI roadmap. Grant Thornton says only 6% of executives recognize change leadership as the critical AI skill. The math tells you exactly where AI ROI is leaking. Here is what that costs, and how to close the gap.

The Question I Keep Hearing

A CEO called me last quarter. Her organization had just finished a six-figure AI rollout. New tools, new workflows, new vendor relationships. The technology worked. People were using it.

Her question to me was not, “How do we get more out of this?”

It was, “Why does nothing feel different?”

I have heard a version of that question from three different CEOs in the last 90 days. Same investment scale. Same vendor confidence. Same quiet disappointment six months in.

Until recently, I would have answered that question from experience. After 30 years leading enterprise transformations, I knew the pattern: the technology layer is rarely the limiter. People, process, and context determine whether technology delivers value.

Now I have a number for it.

The 70-20-10 Math

Last month I attended the AI Hub Summit at UW-Madison. Ramayya Krishnan, who runs the AI Measurement Science and Engineering Center at Carnegie Mellon’s Heinz College, gave a keynote that put research behind what experienced operators already suspected.

The keynote referenced Boston Consulting Group (BCG) research on AI deployment success. The breakdown:

• 70% of AI deployment success comes from people, process, and context.

• 20% comes from digital technology and infrastructure.

• 10% comes from the AI itself.

  
  

Krishnan visualized this as an iceberg. The AI technology you can see (the tool, the model, the interface) is the visible 10% above the waterline. Below the waterline is everything that determines whether the visible 10% delivers value: people and change management, process and workflow redesign, context and domain knowledge, governance and risk.

Most organizations invested in the iceberg upside down. 70 to 80 percent of effort and budget went into the visible 10%. Maybe 10 to 20 percent went into the layer where success actually lives.

The result is predictable. The technology works. Productivity gains are real. And the organization is not meaningfully different than it was before.

Three Out of Four Leaders Don’t Have a Plan

At the same Summit, Karen Sauder, President of Global Client and Agency Solutions at Google, gave a keynote that put a different number on the same problem.

Per Karen Sauder: only 25% of company leaders say they have a comprehensive, defined AI roadmap in place for their organization. 

Three out of four leaders are leading AI without a comprehensive plan to point to. Three out of four are buying tools and approving budgets without having defined what they actually expect AI to change about their business.

Stack the two data points together. Three out of four leaders have no comprehensive plan. The plan they should have is mostly about people, process, and context, not technology. Most organizations are walking into AI deployment without the plan, and the plan they need would mostly be about something other than the technology they are buying.

That is the gap.

And Only 6% Recognize the Real Skill

A third data point landed last week, and it makes the pattern impossible to miss.

Per Grant Thornton’s 2026 AI Impact Survey of 950 C-suite and senior business leaders: only 6% of executives say change leadership and workforce enablement is a top skill needed for thriving in an AI-driven environment.

Six percent.

Grant Thornton’s own conclusion from the data is the line that should land hardest: “AI is a change management initiative, not an IT project.”

Stack the three numbers together. BCG says 70% of AI deployment success is the change layer. Sauder says only 25% of leaders have a comprehensive plan for it. Grant Thornton says only 6% of leaders even recognize that change leadership is the critical skill.

That is the proof gap, written in three numbers.

Why This Pattern Repeats

AI is not the first time this has happened. ERP rolled out the same way in the 2000s. Digital transformation rolled out the same way in the 2010s. Every wave of enterprise capability arrives with the same trap.

The trap looks like this. A new capability arrives. The vendors are confident. The boards are pushing. The investment case is built around the technology. Budget gets allocated to the platform, the integration, the licenses. Change management gets a line item that is 5 to 10 percent of the technology spend.

Then the rollout happens. The technology works. People use it. Productivity gains show up at the task layer. And six months in, leadership is wondering why the strategic outcome the investment was supposed to produce has not materialized.

The diagnosis is almost always the same. The capability was built. The readiness to use the capability was not.

The problems that never go away are not unsolvable. They are undiagnosed.

AI Acts as an Amplifier

Here is the framing that has shifted how I think about this work.

AI is not a tool that enters an organization neutrally. It is an amplifier.

When you introduce AI into a system that already has unresolved issues, fragmented processes, unclear ownership, legacy technical debt, misaligned incentives, it does not clean any of that up. It makes the existing dysfunction faster and more visible.

In a healthy organization, AI accelerates good work. In a fragmented organization, it accelerates fragmentation.

That is why the 70% layer matters so much. The 70% is the readiness of the system the AI is entering. If that system is clear, AI compounds the clarity. If that system is dysfunctional, AI compounds the dysfunction.

This is why so many AI investments produce “productivity gains” without producing transformation. The technology amplifies what was already there. If what was already there was a clean process, you get faster cycle times. If what was already there was a tangled process, you get faster tangle.

The leaders getting real returns from AI did the readiness work first. They fixed their data governance. They clarified their processes. They aligned their people around what the technology was actually supposed to do. Then they introduced AI.

That sequence matters. Technology does not create alignment. Leadership does.

Three Mistakes I See Repeatedly

Every leader I talk to whose AI investment is underperforming is making one of three mistakes. Sometimes all three.

One. Building on a cracked foundation. Unresolved technical debt, unclear data ownership, fragmented workflows. AI gets layered on top. It does not fix the cracks. It pressurizes them.

Two. Mistaking activity for strategy. Every department spins up its own tools. Innovation feels like it is happening everywhere. There is no unifying strategy, no coordination, no shared infrastructure. Some call it “agent sprawl.” It creates more complexity than it resolves.

Three. Automating the past instead of designing the future. The instinct is to use AI to make existing processes faster. If those processes were built around silos, outdated workflows, and manual handoffs, you are just digitizing dysfunction. The opportunity is not speed. It is rethinking how work flows.

None of these are technology failures.

What the 1-in-50 Get Right

Per Gartner (cited via HBR, 2026), only 1 in 50 AI investments delivers transformational value. Most of the other 49 are not failures in any technical sense. The technology operates. People use it. Productivity gains are real.

“Transformational” is doing real work in that sentence. It means the organization is meaningfully different than it was before. Different decisions get made. The operating model has shifted. The competitive position has changed.

The 1 in 50 that get there share a pattern that maps directly onto the 70-20-10 framework. They invested as if 70% of success would come from the change layer.

What that looks like in practice:

The CEO and the operating model owner are at the table when AI decisions get made. Not just IT. Not just digital. Not just procurement.

Leadership engages with the tools personally before mandating them. They build a real intuition for what AI can and cannot do, instead of approving budget on faith.

The change management investment is sized relative to the strategic weight of the change. Not bolted on at the end as an afterthought.

Workflows get redesigned around what AI changes about decision-making and task flow. Not bolted on top of legacy processes.

Metrics evolve to capture what AI is supposed to enable, instead of measuring the same things differently.

Same technology. Different organization around it. Different outcome.

The Healthcare Case

Krishnan walked through a case study at the Summit that translates to almost any industry. EHR systems were supposed to make physicians more efficient. Instead they added 1 to 2 hours of documentation work outside the workday. Physicians were drowning in the technology that was supposed to help them.

Then Ambient AI arrived. The technology that listens to a clinical visit and produces structured notes automatically.

Here is what is interesting. The Ambient AI did not just make documentation faster. It changed the workflow shape entirely. The old workflow was serial: consult, dictate, transcribe, review, enter into EHR. The new workflow is parallel: consult while AI transcribes in the background, physician reviews and signs off.

Result: 7 to 8 hours per week per physician saved (Albrecht et al., 2025).

The lesson is not about AI. It is about workflow.

Most organizations deploy AI on top of existing serial workflows and book modest productivity gains. The transformational gains come from rewiring the workflow shape itself: identifying which steps can run in parallel, which steps the AI eliminates entirely, and which new steps the AI creates that did not exist before.

What to Do This Week

If you have an AI investment that is underperforming, the next move is not to evaluate the technology again. Start with these three questions.

What is one decision our team makes differently because of AI? Not faster. Differently. If the room goes quiet, the AI is operating at the task layer, not the decision layer.

Did we redesign the workflow, or just augment it? Adding AI to an existing serial process produces modest productivity gains. Rewiring the process around what AI can change produces transformation.

If you stripped AI out of your current strategy, would the strategy still make sense? If the answer is no, the problem is not your AI investment. It is what is underneath it.

The honest answers will tell you exactly where the gap lives. And it will not be in the technology.

Where I Come In

I work with CEOs, founders, and senior leaders navigating exactly this problem. Where capability ran ahead of readiness, and someone needs to close the gap.

Most consulting builds the system and leaves. Most coaching grows the leader while the structure stays broken. I do both at once. The operational reality and the leader navigating it. Because that is how the work actually sticks.

If the question being asked in your boardroom is some version of “why does this not feel different,” that is the conversation worth having.

I take a small number of engagements each year. If something here landed, send me a message.

Find the real problem. Do the real work. Build what lasts.

Sources

• Boston Consulting Group’s (BCG) 70-20-10 principle, presented by Ramayya Krishnan, Cooper Professor of Management Science and Information Systems and Director of the AI Measurement Science and Engineering Center at Carnegie Mellon’s Heinz College. Keynote: “AI in the Enterprise: From Capabilities to Deployment.” AI Hub Summit at UW-Madison, April 16-17, 2026.

• 25% AI roadmap statistic: per Karen Sauder, President of Global Client and Agency Solutions at Google, keynote at the AI Hub Summit at UW-Madison, April 16-17, 2026.

• 6% change leadership statistic and revenue growth comparison: Grant Thornton 2026 AI Impact Survey of 950 C-suite and senior business leaders, published April 2026. Available at https://www.grantthornton.com/services/advisory-services/artificial-intelligence/2026-ai-impact-survey

• Gartner data on AI investment outcomes: per Gartner, cited via HBR, 2026.

• Ambient AI healthcare workflow data: Albrecht et al., 2025.