Forgetting change management undermines AI ROI - SDxCentral

TEXT START: One statistic jumps out in the MIT Media Labs report on the state of generative AI: 95% of enterprise AI pilots never reach full production.

The Dissection

The article reframes a structural labor conflict as an implementation problem. It treats employee fear, executive uncertainty, undocumented workflows, and weak adoption as the main reasons AI stalls. Its proposed solution is process intelligence: build a digital twin, map the real organization, identify automatable work, clarify future skills, and normalize perpetual transformation.

The commercial logic is clear. The text creates demand for process-mapping and AI-governance tools while reassuring executives that the transition can be managed. It also reassures workers that roles will change rather than disappear, with humans retained for judgment and exceptions.

The Core Fallacy

The article mistakes today’s automation boundary for a permanent division of labor.

“Boring stuff” is merely AI’s entry point. Once repeatable workflows are mapped, the exceptions, judgment calls, coordination, and quality checks become the next targets. Under P1, improved models do not preserve the human-in-the-loop; they progressively reduce the loop to a servitor, rubber stamp, liability shield, or temporary transition layer.

The article also confuses organizational friction with technological limitation. A digital twin may help move pilots into production, but that success makes labor more legible and therefore more removable. The better the map, the easier it becomes to eliminate handoffs, consolidate roles, and concentrate output in AI capital.

Reskilling does not solve P3. If AI makes each remaining worker more productive, fewer workers are required. Training people for disappearing or shrinking categories of work is not productive participation; it is a queue-management strategy for displacement.

Hidden Assumptions

• AI will remain a complement to human labor rather than an increasingly capable substitute.
• Exceptions and contextual judgment will remain beyond automation.
• A human in the loop will remain economically necessary rather than merely legally or cosmetically useful.
• Executives can make credible long-term promises while AI capabilities and competitive conditions change monthly.
• Employees will cooperate once expectations are clarified, even when cooperation accelerates their own replacement.
• Reskilling will create enough valuable positions to absorb displaced workers.
• Process intelligence can capture organizational reality faster than AI can alter it.
• Firms can preserve human-heavy domains despite competitors that automate more aggressively, contradicting P2.
• AI ROI means successful adoption for the organization, rather than output capture by owners of AI capital.

Social Function

Primary classification: transition management and ideological anesthetic, with a substantial element of elite self-exoneration and prestige signaling.

The article contains a partial truth: messy processes, poor incentives, and worker resistance can absolutely kill current AI pilots. But it uses that truth to conceal the larger mechanism. It presents displacement as a communication failure, suggests that better guidance will reconcile workers with transformation, and shifts attention away from ownership, bargaining power, and the collapse of labor demand.

“Perpetual transformation” is not a humane solution. It is a description of permanent economic insecurity dressed up as organizational culture.

The Verdict

Change management can improve the conversion of AI pilots into production. It cannot preserve the mass employment-to-consumption circuit. The digital twin is not a durable human moat; it is an inventory of the carcass, making labor easier to automate. Under P1–P3, this article is operationally useful to Sovereigns and psychologically useful as transition management, but it offers no defense against the terminal decline of wage-centered capitalism.