April 27, 2026
Managing Partner
AI workforce planning should not convert task automation estimates directly into headcount targets; the real economic unit is the redesigned role, where savings depend on which work disappears, which judgment remains, and which institutional knowledge must be retained to make AI productive.
AI does not make workforce planning simple by automating tasks or reducing jobs. The economic unit that matters is the redesigned role. Boards are asking a reasonable question in the wrong unit of measure. If generative AI can automate a meaningful share of work, executives want to know how much headcount the system will free up. The impulse is understandable. Labor is one of the largest controllable cost lines in most companies, and AI investment cases increasingly need to show productivity, not experimentation. But a task-level automation estimate is not a workforce strategy. It is a diagnostic.
The error is treating automatable tasks as if they could be cleanly converted into removable roles. They rarely do. Work is bundled inside roles, governed by controls, connected to customers, shaped by regulation, and learned through experience. Once AI removes or accelerates pieces of that work, the remaining role often becomes less routine, more exception-heavy, and more dependent on judgment. In that environment, the economically relevant question is not “how many tasks can AI do?” It is “which roles need to be redesigned, which knowledge must be retained, and which work should actually disappear?”
This reframing matters because the transition economics can change when institutional knowledge is treated as an asset rather than a cost. That does not mean every worker should be retained or that AI-driven reductions are misplaced. Some work will disappear. Some functions will need fewer people. The point is narrower and more important: companies that set headcount targets directly from task exposure risk cutting the very knowledge they need to capture the AI savings they promised the board.
The evidence base does not support a straight line from exposure to elimination. The U.S. Bureau of Labor Statistics has made this distinction explicitly in its approach to AI-related employment projections. New technologies can change the composition and weighting of tasks, sometimes dramatically, without directly affecting employment. In the BLS view, the employment effect depends on adoption speed, demand response, regulation, customer preferences, and whether productivity gains increase output rather than simply reduce labor.
The International Labor Organization and NASK reach a similar conclusion from the exposure side. Their 2025 global index finds that 1 in 4 workers is in an occupation with some exposure to generative AI, but only 3.3 percent of global employment falls into the highest exposure category. More importantly, the report states that job transformation, not full replacement, is the most likely impact for many occupations because most jobs still require human input.
SHRM’s 2025 analysis of HR employment shows why this distinction matters in a white-collar function. SHRM estimates that 19.1 percent of U.S. HR employment is in jobs where at least half of the tasks are already automated. On a task-exposure reading, that number can look like the beginning of a reduction target. But SHRM also estimates that only 9.3 percent of HR employment is both at least 50 percent automated and lacks definitive nontechnical barriers to displacement. The gap is not trivial. It is the difference between a task screen and a workforce decision.
The HR example should not be overextended. It does not prove that other functions have the same exposure-to-displacement gap, and SHRM’s own conclusion is that some HR workers are likely to be displaced. But it proves the analytical point executives need: automation potential is insufficient. Legal requirements, client preferences, compliance judgment, process exceptions, employee trust, and workflow accountability can all stop an automated task from becoming an eliminated role.
That distinction is especially relevant for CEOs and CHROs because workforce transition decisions are being made before the operating model is stable. A company may know that AI can draft job descriptions, screen resumes, summarize performance data, or answer employee-policy questions. It still does not know, from those facts alone, whether the HR business partner role shrinks, whether a shared-services role is redesigned, whether a compliance review becomes more important, or whether managers require more support because AI has increased the speed of talent decisions.
BLS occupational case studies make the same point at a broader level. Selected occupations that are susceptible to AI impacts show sharply different projected employment paths. Software developers, database architects, personal financial advisors, and financial analysts are projected to grow, even though AI can improve productivity in parts of their work. Other roles, such as insurance appraisers, customer service representatives, and credit analysts, face projected declines or weaker demand because AI can more directly substitute for core tasks.
The lesson is not that AI protects high-skill roles while harming low-skill ones. That is too simple. The lesson is that the same technology can affect roles differently depending on the task's context. If AI helps a software developer write code faster, demand for software and AI implementation capacity can absorb or exceed the productivity effect. If AI helps a financial advisor produce faster analysis, human counsel may still matter for complex households and trust-based decisions. If AI can estimate property damage from images, a larger share of an insurance appraisal workflow may become substitutable.
For a management team, this means role redesign must precede reduction math. A task inventory can identify where AI creates capacity. It cannot determine whether that capacity becomes lower cost, higher throughput, better service, faster controls, or redeployed talent. Those are operating-model choices. They require a view of demand, risk tolerance, quality requirements, customer behavior, and the institutional knowledge embedded in the current workforce.
The board-level danger is false precision. A claim that “30 percent of tasks are automatable” may be analytically useful. A claim that “therefore 30 percent of roles can be removed” is usually not. The first statement describes technical exposure. The second assumes that tasks are separable, that remaining work can be reassigned without loss, that knowledge transfer is free, and that adoption will not slow when experienced workers leave. Those assumptions deserve to be tested, not buried inside a savings target.
Institutional knowledge is often misclassified in AI workforce cases. In a cost-takeout model, experienced employees appear as expensive capacity attached to automatable work. In a transition model, they may also be carriers of process judgment, exception logic, customer context, risk memory, and informal workarounds that explain how the process actually functions. AI can automate explicit tasks more easily than it can reconstruct tacit operating knowledge once it has left the organization.
The research on knowledge loss is strong enough to make this a real economic variable, even if it does not produce a universal dollar value. A systematic review of 91 empirical studies on turnover-induced knowledge loss found that employee departures can damage organizational knowledge bases and performance, particularly when the departing knowledge is tacit rather than explicit. A separate review of turnover-cost studies by the Center for American Progress found a typical replacement cost of about one-fifth of annual salary, driven by hiring, training, productivity loss, and ramp-up. That figure is older and not AI-specific, so it should not be treated as a universal benchmark. It does, however, reinforce a practical point: workforce exits create costs that do not always appear in the first version of an AI savings model.
Generative AI also changes the value of knowledge in a more positive way. In a field study of 5,172 customer-support agents, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond found that AI assistance increased productivity by 15 percent on average, with larger gains for less experienced and lower-skilled workers. The authors found evidence that AI helped disseminate behaviors associated with more productive agents and helped newer workers move faster down the experience curve.
This study does not prove that experienced workers always become more valuable after AI deployment. In that setting, the largest measured productivity gains accrued to less experienced workers, and the authors caution that the study does not estimate aggregate employment effects. The more useful implication is different: AI can turn embedded expertise into a scalable operating asset when that expertise is captured, codified, and built into workflows. That asset is harder to create if the organization removes the people who know where the process breaks.
This is where many workforce transition models underprice risk. They model wage savings from role elimination with spreadsheet clarity, then treat knowledge loss as a qualitative change-management concern. That reverses the economics. In many functions, the workers most familiar with process exceptions are also the workers best positioned to train the model, validate outputs, redesign controls, identify failure modes, and teach less experienced colleagues how to use AI responsibly. Retaining them is not a sentimental choice. It can be a productivity choice.
The current executive debate is not theoretical. The World Economic Forum’s 2025 Future of Jobs survey found that 85 percent of employers plan to prioritize upskilling, 50 percent plan to transition staff from declining to growing roles, 40 percent plan to reduce staff as skills become less relevant, and 40 percent anticipate reducing the workforce where AI can automate tasks. Those numbers describe a portfolio of actions, not a single directional bet.
The portfolio view also handles the strongest counterargument: some displacement is already visible. A Stanford Digital Economy Lab working paper using high-frequency payroll data found that early-career workers in highly AI-exposed occupations experienced a 16 percent relative employment decline, while more experienced workers in the same occupations were more stable or continued to grow. The authors also found that declines were concentrated where AI is more likely to automate rather than augment human labor.
That evidence should make executives more precise, not more cautious in the abstract. AI will reduce demand in some roles, especially those that involve digitized, standardized, and automatable work with minimal relationship, compliance, or judgment requirements. But the Stanford finding also reinforces the need to distinguish automation from augmentation and early-career work from experienced work. A reduction strategy that treats all exposed roles as equivalent will miss those differences.
For CEOs, the practical issue is sequencing. A layoff-led approach can deliver immediate cost reduction, but it can also slow the work required to generate the productivity benefit. The organization still needs redesigned workflows, governance, training, process controls, model evaluation, customer communication, and new performance measures. If the people who understand today’s process leave before tomorrow’s process is built, the company may save on payroll while increasing transformation risk.
For CHROs, the issue is role architecture. Traditional workforce planning starts from jobs, spans, layers, and skill gaps. AI workforce planning must start one layer deeper, with work elements, then move back up to redesigned roles. Which tasks disappear? Which tasks remain but become faster? Which tasks become more important because AI increases volume or risk? Which judgment points require human sign-off? Which employees hold the experience needed to redesign the system? A headcount target cannot answer those questions.
Microsoft Research’s 2025 New Future of Work report reinforces why this is not only an HR concern. The report argues that organizational AI adoption depends on employees as much as leaders, that some of the best uses of AI can come from the edge rather than the center, and that worker involvement in technology design can improve productivity, satisfaction, skill growth, workflow fit, and adoption. That evidence does not prove a universal ROI for participation, but it supports a practical rule: the people closest to the work should be involved before the company finalizes the automation case.
The right executive model starts with task exposure but does not stop there. Santiago & Company’s Workforce Transition Equation reframes the AI business case as a role-redesign case. The value of an AI workforce move equals automation savings minus transition cost, knowledge-loss risk, and adoption drag, plus the value of redeployed capacity. The components are not meant to imply a universal formula with precise coefficients. They are meant to force the right economic conversation before a reduction target becomes policy.
This model changes how the CEO and CHRO should govern AI workforce planning. A business unit should not receive approval for a reduction case simply because a task assessment shows high automation potential. It should show which work will truly disappear, which roles will be redesigned, which controls will change, which knowledge must be retained through the transition, and how redeployed capacity will create value. If the case cannot answer those questions, the savings may be real on paper but fragile in execution.
The model also clarifies where reductions are justified. If work is standardized, demand is stable or declining, quality risk is low, and knowledge can be documented or transferred, elimination may be the right move. If work is automatable but surrounded by exceptions, customer trust, regulatory requirements, or complex handoffs, role redesign may create better economics than immediate reduction. If experienced workers hold knowledge that can train systems, validate outputs, or accelerate adoption, redeployment may outperform both retention-as-usual and layoff-led savings.
This is not an argument for slower transformation. It is an argument for fewer false starts. Companies that rush from task exposure to headcount reduction may capture visible savings while creating invisible liabilities. Companies that redesign roles first can still reduce headcount where the economics support it, but they are less likely to cut the capacity required to make AI productive.
The immediate board question should change. Instead of asking management how many roles AI can eliminate, directors should ask which roles AI will reconstruct and how management has priced the transition. What institutional knowledge is required to redesign the work? Which experienced employees are needed temporarily, which are needed permanently, and which roles should genuinely go away? How will the company measure adoption drag, quality risk, and time-to-productivity? What value will redeployed capacity create, and who owns that value in the P&L?
The companies that get this right will not be the ones that avoid hard workforce decisions. They will be the ones who make those decisions in the right order. Task automation identifies where change is possible. Role redesign determines where value is real. Institutional knowledge decides whether the transition compounds that value or destroy it before the savings arrive.
Aerospace and defense companies will convert record demand into revenue only by managing qualified throughput. The scarce asset is qualified throughput, and it has to be managed node by node. The approved combination of site, process, workforce, and part at the individual production-node level.
Santiago & Company announced the public release of the AI Workforce Stability Architecture (AWSA) at the United Nations WSIS Forum 2026, introducing an open framework that helps governments monitor, prepare for, and respond to the economic and fiscal impacts of AI-driven labor displacement in service-export economies.
AI is moving on a 15-month clock; governments are moving on a five-year one. For service-export economies, the winners will not be those that forecast labor disruption most precisely, but those that pre-wire fiscal and workforce responses before the evidence becomes undeniable
SpaceX does not need to disrupt the wireless industry to reshape it. By turning satellite coverage into negotiating leverage, it is forcing carriers to rethink who owns the customer relationship when coverage comes from space.
