Why enterprise AI adoption stalls at scale
Enterprise AI adoption stalls at scale because the organization around the tools stays the same. Approval chains, legacy workflows and request queues keep the old way of working faster and safer than the new one. People follow the easiest path. Scaling AI means redesigning strategy, workflows and the operating model so the AI-first path becomes the easiest one.
This page draws on Hard-Wired, BOI’s report on the neuroscience of AI-first transformation.
Why do enterprises struggle with company-wide AI transformation programs?
Enterprises struggle because they add AI to an organization designed for a different era. Leaders keep their old mental models. Workflows keep their old steps. Governance keeps its old queues. Each layer quietly rewards the familiar way of working. People respond to that environment, so new tools end up speeding up yesterday’s operations instead of replacing them.
BCG’s Applied AI Index 2026 surveyed 1,330 senior leaders. Nearly half of their companies now generate real value from AI. The rest struggle to turn investment into impact. BCG’s 10-20-70 rule locates the gap: 70% of AI value comes from people and processes.
Where does AI transformation stall: strategy, workflow or operating model?
It stalls in all three, for different reasons. At the strategy layer, leaders protect the model that made them successful. At the workflow layer, AI stays optional, so people skip it under pressure. At the operating model layer, central teams become bottlenecks. Each layer needs its own redesign. The gains only compound when all three move together.
| Layer | What stalls | Behavioral mechanism | What to redesign |
|---|---|---|---|
| Strategy | Leaders fund use cases and optimize today's business model | Cognitive entrenchment | Practice through provocation. Every executive owns an AI-native future for their domain. Measure choices made. |
| Workflow | AI sits on top of old processes. People revert under pressure. | Anchoring and the default effect | Redesign from the outcome. Assign every step to AI, human or both. Make AI the default. |
| Operating model | Central teams drown in requests. Business teams build duplicates. | Present bias, not-invented-here, low self-efficacy | Quarterly portfolio reviews, co-funding, reusable capabilities, governance built into the workflow |
Why do AI pilots fail to scale?
Pilots fail to scale because organizations count them as progress. A pilot proves a tool works in one place. Scaling demands harder choices: what to stop, what to reuse, who owns it and who pays. When success is measured by pilots launched, nobody makes those choices. The portfolio grows. The business stays the same.
In 2025, S&P Global Market Intelligence found that 42% of companies had abandoned most of their AI initiatives. A year earlier, it was 17%. Respondents blamed cost, privacy and security. Behavior makes those worse. Teams prefer building their own tools to reusing someone else’s, so three teams with one problem buy three tools and share nothing.
Why do leadership teams keep optimizing today's business instead of reinventing it?
Leaders who built a successful business carry deep assumptions about how value gets created. Behavioral scientists call this cognitive entrenchment: relying on mental models that once worked after the environment has changed. Those assumptions feel like facts. So leaders ask where AI fits today’s business instead of which parts of it no longer make sense.
Mental models change through practice. Briefings rarely move them. An AI-native attacker exercise asks how a 50-person AI-native competitor would beat you. Then measure decisions made and initiatives stopped. As Laura Stevens, PhD puts it: “A strategy isn’t really a strategy if it’s almost impossible to disagree with it.”
Why don't employees adopt AI tools, even after training?
Employees fall back on old routines because the workflow still makes them the easiest option. Habits run on autopilot and respond poorly to new information. Training adds knowledge but leaves the friction in place. If using AI means finding a tool, checking a policy and requesting access, the manual path wins under pressure.
This is the default effect at work. A classic 2003 study found organ donation consent far higher where donation was the default. Flip the default in the workflow. AI-led execution becomes the standard step. Manual work becomes the exception that needs a reason.
Is change management enough for AI transformation?
Classic change management helps, but it can’t carry an AI transformation alone. Communication plans, training and adoption campaigns all assume people mainly need convincing. Behavior depends just as much on the environment. Think of how decisions get made, which steps a workflow demands and what gets rewarded. Change has to be designed into those systems.
Why is redesigning work with AI agents harder than it looks?
Redesign usually starts from the current process, and that anchors everything that follows. Psychologists call it anchoring: staying attached to the first solution you see. Teams map the as-is, fix obvious gaps and bolt AI or agents onto existing steps. Legacy approvals survive. The result is yesterday’s workflow, slightly faster.
Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027. It cited costs, unclear value and weak risk controls. “If you start with today’s workflow, you’ll almost always end with tomorrow looking remarkably similar to yesterday,” says Laura Stevens.
How should enterprises redesign jobs and workflows around AI agents?
Start from the outcome the customer needs. Then ask what the workflow would look like if AI had always existed. Assign every step to AI, a human or both. Let AI handle speed, consistency and scale. Keep humans where judgment, accountability and creativity matter. Make AI the default.
Question every approval and handoff. Explicit roles prevent overlap, and fewer steps make new behavior easier to adopt. “Automation shouldn’t need permission. Manual work should,” says Laura Stevens.
An illustrative scenario. Picture a logistics business where a delayed shipment needs three handoffs before the customer hears about it. The finance sign-off turns out to rest on a compliance rule that changed two years ago. Redesigned around a fast, accurate answer, three handoffs become one AI-flagged exception with a single human check.
How should we redesign our operating model to work with AI?
Move from a helpdesk to a platform. Central AI and IT teams stop building every request. They build shared capabilities, data and guardrails that business teams build on. Pair that with quarterly portfolio reviews, co-funding and graduated autonomy. Ownership spreads, reuse becomes the easy option and the central team stops being the bottleneck.
What makes it hard to redesign operations with AI at scale?
Inherited operating models treat technology like a ticket. The business orders. IT delivers. AI doesn’t work that way. It enters through every door, gets reused in ways no single team controls and keeps evolving after launch. Demand outpaces capacity. Shadow solutions fill the gap.
Present bias lets the loudest request win. Central budgets make asking for AI feel free.
How should AI governance work so it doesn't slow teams down?
Governance should run inside the workflow instead of beside it. Committees, approval boards and policy PDFs rely on people pausing to look up the rules. Urgency pushes them not to. Embedded governance makes the compliant path the easiest path. Think traffic lights rather than traffic wardens: the system keeps things safe without someone at every intersection.
Board of Innovation’s GATE framework puts this into practice:
- Guardrails enforced automatically, instead of written into a policy document.
- AI-native design that makes every experiment reusable from day one.
- Training delivered as just-in-time nudges in the flow of work.
- Enablement through self-serve sandboxes that show what already exists.
What should leaders do first?
- Pause new AI pilots. Make room for strategic reinvention before adding to the portfolio.
- Run an AI-native attacker exercise with the executive team.
- Ask every executive to define an AI-native future for their domain. No delegation.
- Pick the workflow people complain about most. Redesign it from the outcome.
- Make AI the default step. Require a reason for manual work.
- Replace continuous intake with a quarterly portfolio review and co-funding.
- Run a GATE session on one live workflow.
More resources on AI transformation
FAQ
What is an applied AI company? A company that builds AI into a specific business outcome for its clients. The output is a working system, such as a product, a workflow, or a decision tool.
Should a large enterprise build AI in-house or work with a partner? MIT NANDA’s 2025 research found that AI bought from specialized vendors and partners succeeded about 67% of the time, while internal builds succeeded about one-third as often. Many enterprises partner to build and prove a system, then take it in-house to run it.
How long does it take to see results? A focused prototype can be working within weeks. Scaling across an enterprise takes longer and depends on data, integration, and change management.
How is an applied AI studio different from a consultancy? A studio combines strategy and engineering in one team and ships working systems. A consultancy is typically stronger on enterprise-wide strategy and change programs, and often partners for engineering.
What does BOI do? BOI is an applied AI company for large enterprises. It builds working AI systems that help clients create new value with AI, redesign work, and build AI-native operating models. Clients include Walmart, Nestlé, Chiesi, and Coca-Cola.
