The hardest work in AI adoption isn’t building the technology, it’s changing how organizations operate.
When enterprises implement AI, the initiatives that succeed rarely spend the most on algorithms or infrastructure. Instead, they invest most heavily in something companies typically underfund—people and processes.
The framework BCG describes breaks the work of AI transformation into three categories. Algorithms and models account for roughly 10 percent of the effort. Technology infrastructure represents about 20 percent. The remaining 70 percent comes from redesigning roles, managing organizational change, and training teams to work alongside intelligent systems.
This observation isn’t philosophical. It reflects where the real work of AI implementation actually occurs. As BCG notes in its research on AI agents, “AI agents fundamentally change how work gets done and by whom; as such, the lion’s share of the leadership team’s effort must go into redesigning roles, managing change, and training your workforce to provide the right oversight and guidance to AI agents—and to reinforce expertise where humans make a difference.”
The 70/20/10 rule describes the effort distribution in transformations that succeed. When enterprises invert that allocation and spend heavily on technology while underfunding organizational change, they fund innovation initiatives that fail to transform how the business actually runs.
If the majority of the work in AI transformation lies in people and processes, it raises an obvious question: why do most enterprise AI budgets look almost exactly reversed?
Part of the explanation is structural. Technology investments are easier to fund, easier to procure, and easier to measure than organizational change. A platform purchase produces something tangible. Infrastructure gets deployed, models run, dashboards appear, and the organization can point to visible progress. These milestones provide a sense of momentum even if the surrounding workflows remain unchanged.
Organizational transformation does not offer the same clarity. Redesigning workflows requires examining how teams actually operate, not how leadership assumes they operate. Training employees to collaborate with AI systems often involves redefining decision rights, accountability structures, and performance metrics. Governance models must evolve so employees understand when to rely on automated outputs and when to override them. None of these changes arrive as a single deliverable.
Technology projects tend to have defined timelines and completion criteria. Organizational change unfolds gradually, often revealing deeper structural issues along the way. Because of this difference, enterprises often prioritize funding the part of AI initiatives that feels most concrete: the technology itself.
The problem emerges once that technology is deployed. Without the surrounding transformation, it struggles to integrate into existing workflows. The AI system produces outputs, but those outputs rarely influence decisions.
The system works. The organization does not.
When executives hear that most of the effort in AI transformation lies in people and processes, they often interpret it as a call for more training programs or change management initiatives. That interpretation dramatically understates what the 70 percent actually represents.
The bulk of the work is not teaching employees how to use new tools. It is redesigning the work itself so those tools can matter.
Most enterprise workflows were designed in environments where humans performed every step of the process. Employees gathered information, interpreted patterns, made decisions, and executed tasks sequentially. AI changes that structure.
Information gathering may now be automated through integrated data systems. Pattern recognition may occur within machine learning models rather than through manual analysis. Decision support may arrive instantly rather than after hours of human investigation.
However, if the workflow itself remains unchanged, the AI becomes an advisory system rather than an operational one. The model generates insights, a report is produced, and someone reviews it during a meeting. Eventually a decision is made through the same process that existed before the AI system was introduced.
In that environment the technology produces intelligence, but the organization continues operating as though that intelligence were optional.
The real work of AI transformation lies in redesigning workflows so AI outputs become part of operational processes rather than an additional layer of information.
The emergence of AI agents makes this challenge even more visible. Earlier generations of enterprise AI primarily focused on prediction. These systems analyzed data, surfaced patterns, and produced recommendations that humans would evaluate and act upon.
Agentic systems behave differently. Instead of simply producing insights, they can initiate and execute sequences of tasks. They retrieve data, interact with systems, trigger workflows, and coordinate actions across multiple tools.
This capability means AI is no longer simply informing work. It is beginning to participate directly within it.
That shift forces organizations to reconsider how responsibilities are distributed between humans and machines. In customer service environments, for example, conversational AI does more than automate responses to common questions. It reshapes the structure of the contact center. Routine inquiries move toward automation, while human agents increasingly focus on complex cases and exception handling. Operations teams shift from managing queues toward managing automation performance and escalation logic.
Deploying the technology can happen relatively quickly. Redesigning the operational model that surrounds it takes far longer.
Without that redesign, AI agents operate inside workflows that were never built to accommodate them.
When enterprises underinvest in the organizational side of AI transformation, the symptoms tend to look remarkably consistent across industries. AI tools launch successfully but remain confined to pilot programs. Employees continue relying on legacy workflows rather than incorporating AI outputs into their daily decisions. Automated recommendations are viewed with skepticism, or treated as informational rather than actionable.
None of these issues originate in the algorithms. They originate in the absence of operational redesign.
Without new processes, AI outputs rarely influence real decisions. Without training and role clarity, employees lack confidence in how to interpret automated recommendations. Without governance frameworks, responsibility for maintaining and improving AI systems becomes fragmented across teams.
Over time the technology begins to feel underwhelming—not because it lacks capability, but because the organization never adapted around it.
Successful AI transformations rarely resemble technology deployments. They resemble operating model changes.
Workflows are redesigned so AI outputs feed directly into operational decisions. Roles evolve as certain tasks shift from humans to machines. Governance structures emerge to monitor system performance and manage exceptions. Teams learn how to collaborate with intelligent systems rather than simply using them as tools.
This work is slower, more complex, and far less visible than deploying software. It requires sustained leadership attention and coordination across multiple functions.
It is also where the majority of value is created.
Learn how AI can streamline physician referrals, reduce leakage, and improve patient experience across complex healthcare systems.
Explore how modern managed services shift from reactive support to outcome-driven ownership and continuous improvement.
