Are We Automating the Business—or Redesigning It?

Most companies no longer need to be convinced that artificial intelligence can automate work. The harder question is whether automation is the right objective. An organization can deploy AI across forecasting, maintenance, scheduling, customer service, quality, procurement, and reporting and still leave the economics of the enterprise largely unchanged. Tasks get faster. The business does not necessarily become better.

Consider an illustrative multi-site industrial operation that has accumulated a portfolio of successful AI pilots. A planning model improves forecast accuracy. A maintenance application identifies likely equipment failures earlier. A scheduling tool produces labor plans in minutes instead of hours. A generative assistant summarizes operating reviews and prepares exception reports. Each tool has a credible business case, and each function can point to time saved.

Yet the end-to-end order cycle is barely shorter. Inventory remains high because commercial, supply, and operations decisions are still made on different cadences. Site leaders continue to approve routine exceptions because authority has not moved with the information. Planners spend less time producing forecasts but more time reconciling competing recommendations. Managers receive faster analysis, then route the decision through the same hierarchy that existed before AI.

The organization has automated pieces of the business without redesigning how the business operates. That distinction is becoming one of the most consequential choices on the executive agenda.

Automation Is Not the Same as Operating-Model Change

Automation improves the execution of an existing activity. Redesign asks whether the activity, handoff, role, approval, or organizational boundary should continue to exist in its current form. Both can create value, but the economics are different. Automation typically captures task-level savings. Redesign can change cycle time, coordination cost, management capacity, working capital, service, and the amount of organizational complexity required to produce an outcome.

This is why technology adoption is a poor proxy for transformation. McKinsey's 2025 global AI survey found that only 21 percent of respondents using generative AI said their organizations had fundamentally redesigned at least some workflows, and workflow redesign was the factor most strongly associated with reported EBIT impact. Its July 2026 work on AI operating models goes further: top performers were three times more likely to pursue broad operating-model redesign and twice as likely to redesign workflows before selecting AI tools.

The implication is not that every AI deployment should trigger a reorganization. It is that executives should decide deliberately whether they are buying efficiency inside the existing system or changing the system itself. The mistake is expecting the first to produce the economics of the second.

A Seventy-Five-Year-Old Theory Fits the AI Problem Surprisingly Well

Sociotechnical Systems Theory offers a useful foundation for that decision. The field emerged from research by Eric Trist and Ken Bamforth on British coal mining in the early 1950s. Their work showed that the effects of new technology could not be understood through engineering performance alone because technology changed the social organization of work: roles, coordination, autonomy, relationships, and the way teams controlled variation.

The enduring principle became known as joint optimization. The social system and the technical system should be designed together so that neither is optimized at the expense of the performance of the whole. A technically elegant system can underperform if people, authority, incentives, skills, and workflow are poorly matched to it. A socially attractive design can fail for the opposite reason if the technical system cannot support the required performance.

A later review by Pasmore and colleagues examined 134 sociotechnical experiments and connected design choices with outcomes including productivity, cost, quality, and satisfaction. The theory is not a predictive formula for ROI, and 'joint optimization' should not be interpreted as a mathematically perfect state. Its value for executives is more practical: it disciplines the organization against treating technology and work design as separate decisions.

AI makes the sociotechnical problem more consequential because it does more than mechanize physical work. It can synthesize information, recommend actions, generate content, allocate resources, coordinate workflows, and increasingly execute decisions. The technical system is therefore moving into territory that was historically part of the management system. Once that happens, adding AI without reconsidering roles and decision rights is not neutral; it can create a mismatch between what the technology is capable of doing and what the organization still requires people to do.

The Economic Trap of Bolt-On AI

Task savings can remain trapped inside the workflow

A planner saves three hours producing an analysis, but the decision still waits two days for approval. A maintenance model identifies risk earlier, but parts authorization remains slow. An AI assistant drafts a customer response instantly, but three functions still review the exception. The task becomes cheaper while the end-to-end outcome barely moves. The organization records productivity at the point of automation but fails to capture the larger economic value because the surrounding workflow is unchanged.

Coordination cost survives the automation

Large organizations accumulate coordination mechanisms for legitimate reasons: specialization, risk control, local accountability, and the need to reconcile competing objectives. Over time, however, those mechanisms can become expensive. Meetings, reports, approvals, handoffs, reconciliations, and management layers consume capacity. AI can lower the cost of information and analysis dramatically, but if the same people still have to assemble around the same decision in the same sequence, coordination cost remains embedded in the model.

This is one reason AI can produce impressive user-level productivity while enterprise financial impact remains modest. The organization has accelerated activities without changing the architecture through which those activities become decisions and results.

Control can expand when technology should reduce it

New technology can paradoxically create more bureaucracy. Leaders concerned about model error, compliance, cybersecurity, or reputational risk add reviews, parallel checks, and approval gates. Some of that governance is essential. But when risk controls are layered onto an unchanged process rather than designed into a new one, AI can produce a faster task surrounded by a slower system.

The executive challenge is therefore not to minimize human oversight. It is to place oversight where the economic and risk consequences justify it. That question becomes especially important as AI moves from recommendation toward action, because human review added everywhere can eliminate the speed advantage while still leaving accountability unclear.

When Automation Is Enough—and When It Is Not

Sociotechnical thinking does not require every technology change to trigger operating-model redesign. Bounded, stable, well-understood work is often a good candidate for straightforward automation. If the task has clear inputs and outputs, limited downstream ambiguity, established controls, and little effect on authority or coordination, redesign may add complexity without creating additional value.

The threshold changes when the technology alters the work system around the task. If AI changes who has the best information, the speed at which a decision can be made, the number of handoffs required, the role of a manager, the skills needed by the team, the location of expertise, or the economics of centralization versus local control, then the organization is no longer dealing with a simple automation decision.

The same is true when value depends on crossing functional boundaries. An AI forecast can improve a planning task. Capturing enterprise value from that forecast may require commercial leaders to change commitments, finance to change inventory assumptions, operations to change capacity decisions, and supply chain to change allocation rules. When the benefit requires multiple functions to behave differently, the operating model becomes part of the technology business case.

A useful executive test is therefore straightforward: if the benefits remain attractive when the surrounding workflow, decision rights, measures, and roles stay largely unchanged, automate the task. If the benefits only become material when those elements change, leadership is evaluating operating-model redesign whether it calls it that or not.

Redesign Means More Than Moving Boxes on an Org Chart

Once the decision moves from automation to redesign, Jay Galbraith's Star Model provides a useful supporting lens. Galbraith's framework treats organization design as the alignment of strategy, structure, processes, rewards, and people. That matters for AI because changing the technical system can disturb every one of those choices without changing the formal reporting structure at all.

Strategy determines where AI should create economic differentiation rather than merely convenience. Structure determines where decision authority belongs when information becomes more available and execution becomes faster. Processes determine how human and machine work moves across the enterprise. Rewards determine whether leaders optimize enterprise outcomes or protect local metrics. People policies determine which capabilities should be built, which roles should change, and where human judgment creates more value than automation.

If those elements remain aligned to the pre-AI organization, technology can become an expensive layer sitting on top of old assumptions. The organization may have better tools while preserving the same bottlenecks, incentives, and coordination requirements that limited performance before the investment.

A Practical Example: From Better Planning to a Different Planning System

Return to the multi-site operation. Its AI forecasting and scheduling tools are technically successful, but the planning system remains fragmented. Demand is translated through one process, inventory through another, capacity through a third, and labor through a fourth. Each function receives better analysis, yet reconciliation still happens through meetings, spreadsheets, escalations, and managerial negotiation.

The automation approach is to keep improving each tool. Forecasts become more accurate. Schedules run faster. Reports become easier to produce. That may be entirely rational if the company's constraint is analytical effort.

But suppose the real constraint is decision latency. The organization routinely identifies a demand change on Monday, spends Tuesday reconciling inventory, Wednesday evaluating capacity, Thursday negotiating transportation and labor, and Friday approving the response. Improving the forecast does not solve that problem. The information was never the primary constraint; the coordination model was.

A redesign would start with the end-to-end decision rather than the existing functions. AI could continuously integrate demand, inventory, capacity, transportation, and labor signals; generate feasible scenarios; and route only economically significant exceptions to human decision makers. Routine adjustments could occur within predetermined authority and risk limits. Planners would spend less time assembling information and more time challenging assumptions, evaluating exceptions, and improving the system. Managers would own fewer routine approvals and more consequential judgments.

That design could compress cycle time, release management capacity, reduce buffer inventory, and improve asset utilization. It could also create new risks: model dependence, poor exception design, loss of local knowledge, unclear accountability, and excessive centralization. Joint optimization requires pricing both sides. The objective is not maximum automation. It is better system performance.

The Counterargument: Redesign Can Destroy Value Too

There is a legitimate danger in treating every new AI capability as a reason to rebuild the organization. AI capabilities are changing quickly, many use cases remain immature, data quality varies, and some decisions carry risks that justify deliberate human review. Redesign also imposes transition cost: new roles, new skills, new governance, new interfaces, and the temporary performance loss that often accompanies major change.

There is also a strategic risk in designing the organization too tightly around today's technology. A company can hard-wire processes to a tool or vendor that becomes obsolete faster than the operating model can adapt. Sociotechnical theory's open-systems heritage is useful here: good design must remain adaptable to changes in technology, environment, and organizational purpose.

The executive answer is therefore not 'AI-first.' It is purpose-first. Automate where automation produces attractive economics without destabilizing a sound work system. Redesign where the technology changes the fundamental relationship among work, people, decisions, and coordination. Preserve human control where judgment, accountability, trust, or downside risk justify it. The goal is fit, not technological maximalism.

The Decision

For senior leaders, the most important AI question is not how many use cases have been deployed. It is whether those use cases change the economics of the enterprise. License counts, pilot counts, hours saved, and model accuracy can all improve while cycle time, working capital, service, management capacity, and return on invested capital remain stubbornly similar.

Sociotechnical Systems Theory gives executives a durable principle for avoiding that trap: do not optimize the technical system independently of the social system that must use, govern, challenge, and benefit from it. Galbraith adds the organizational consequence: when technology changes how value should be created, strategy, structure, processes, rewards, and people may need to move with it.

The decisive question is this: if we were designing the work today with AI available from the beginning, would we recreate the same workflow, approvals, roles, and boundaries? If yes, automate. If no, redesign.

Selected Sources

Trist, E. L., and Bamforth, K. W. Some Social and Psychological Consequences of the Longwall Method of Coal-Getting. Human Relations, 1951. Source

Pasmore, W., Francis, C., Haldeman, J., and Shani, A. Sociotechnical Systems: A North American Reflection on Empirical Studies of the Seventies. Human Relations, 1982. Source

Galbraith Management Consultants. The Star Model and Organization Design. Source

Ang, K. C. S., et al. Advancing Sociotechnical Systems Theory: New Principles for Human-Robot Team Design and Development. Applied Ergonomics, 2025. Source

McKinsey and Company. The Operating Model Advantage: Why AI Winners Are Rewiring Their Organizations. July 7, 2026. Source

Deloitte Insights. Rewiring the Operating Model for AI. July 2026. Source

Editorial standard: established frameworks, executive application, practical tradeoffs, selective sourcing.