The 45% Hidden ROI of Automation No One Puts in the Business Case: The Vertical Spillover Effect

Research shows that nearly half the gain from capability-building flows upward to managers. The same “vertical spillover” applies even more strongly to process automation, if you know how to calculate it.

A common blind spot in digital transformation is measuring the impact of new tools or training in isolation. We calculate the productivity gain for an individual worker using a new AI copilot or automation platform, but we often overlook the cascading, second-order effects on the rest of the organization.

A rigorous study by Miguel Espinosa and Christopher T. Stanton provides a framework and a stark finding: Nearly half of the total value of an upskilling program came not from the trained workers themselves, but from the time it freed up for their managers.

The Core Experiment and Critical Findings

The researchers studied a training program for frontline workers at a Colombian government agency. They tracked productivity metrics and internal email communications before and after the program.

1. Trained workers’ individual output increased by ~10%. This is the standard first-order effect KPI and a direct gain. In our world, this is the efficiency gain from an engineer using an AI code assistant or an analyst using a generative AI tool.

2. As workers became more capable, their requests for help to managers plummeted by over 60%. This didn’t leave managers idle; it allowed them to refocus on their own “strategic” tasks. Managers with the strongest connections to trained staff saw the largest productivity gains. This second-order Vertical Spillover effect is rarely evaluated in transformation projects.

3. Peer-to-Peer Help Has a Collateral Benefit. Trained workers did start helping their untrained peers more. However, the direct productivity of those untrained peers didn’t significantly increase. The real value was that this informal help acted as a new, unofficial layer of management, further insulating managers from interruptions.

The Stunning Bottom Line

Using a structural economic model, the researchers calculated that vertical spillovers to managers accounted for approximately 45% of the program’s total benefits. Because managers’ strategic tasks had a higher weight in the organization’s production function, freeing their time was disproportionately valuable. The organization would have needed to train nearly twice as many frontline workers to achieve the same output level without these managerial spillovers.

Upskilling increased the range of problems workers could solve alone. Automation and process optimization achieve the same reduction through additional channels: automation can eliminate problems as workers or managers don’t touching processes; automation can reduce exceptions (deviations or errors) through workflow standardization and input error-proofing, reducing the need for manager escalation; automation enables self-service with direct access to information to workers, reducing the need for a manager to pull, interpret, or approve.

Quantifying the Vertical Spillover. After training, aggregate frontline output rose modestly (from ~380 to ~385 tasks). Without parallel improvement in manager productivity (the vertical spillover) the organization would have needed to boost frontline output all the way to 395 to achieve the same total output.

Strategic Imperatives for Your Organization

• Expand the ROI Model for Process Optimization or Automation. Stop evaluating the success of an AI copilot or automation tool purely on the time saved by the direct user. The primary gain may be the reduction of “organizational drag” on senior talent. What strategic work are your managers not doing because they are answering routine questions from junior staff?

• Design for the “Managerial Time Dividend.” When you roll out a new knowledge base or an AI assistant, explicitly track the change in inbound request volume to expert staff. The goal isn’t just to make the junior person faster; it’s to liberate the senior person’s cognitive load for high-leverage work.

• Beware of Mismeasuring Middle Layers. If your new top performers start spending significant time helping others, their individual output metrics may flatline or even decline. Don’t be alarmed. As the study’s model shows, this is often a rational, value-creating reorganization of the “hierarchy of help,” not a failure of the tool.

How to Estimate the Vertical Spillover in the Cost-Benefit Analysis

To predict training or tool-adoption spillovers in a cost-benefit analysis, you can build a practical, data-light projection by following the paper’s core insight: spillover = (manager time saved) × (value of manager time for strategic work).

As a Business Analyst, you can fill the gaps in a few steps.

Step 1: Baseline the “Help Tax” on Managers Measure how much time managers currently spend assisting frontline employees.

• Data source: A one-week time diary, a pulse survey, or analysis of help-desk tickets and email metadata (the scientists did in the paper).

• Action: Ask managers, “What percentage of your week is spent answering questions, reviewing work reactively, or fixing escalated issues from your direct reports?”

• Example result: Suppose a manager spends 10 hours/week (25% of their time) on such reactive help.

Step 2: Project the Reduction in Help Requests Estimate how much the intervention (training, AI copilot, knowledge base) will cut incoming requests. The paper found a ~60% drop in emails from trained workers to managers.

• Conservative approach: Assume requests from the targeted group will fall by a similar percentage as their individual productivity gain. If training boosts worker performance by 10%, assume a 10% reduction in help requests — or more if the training specifically addresses root causes of those requests.

• Data-driven approach: Run a pilot and measure the change in help-desk tickets, Slack/Teams messages, or email volume between the pilot group and a control group.

• Example calculation: 10 hours/week of help time × 40% projected reduction = 4 hours/week saved per manager.

Step 3: Value the Freed Manager Time The paper’s structural model shows managers’ strategic tasks have a higher output elasticity than frontline tasks (roughly 1.75x greater weight). In practice, value this time at a premium over the manager’s hourly cost.

• Cost-avoidance method: Use the fully-loaded hourly cost of the manager. If the manager’s annual cost is $150,000, the hourly rate is ~$75. Saved time: 4 hrs/week × 48 weeks = 192 hrs/year. Value = 192 × $75 = $14,400 per manager.

• Opportunity-value method: Multiply the cost-avoidance figure by an impact multiplier (1.5–2.0) to reflect that strategic work drives disproportionately more business value. The paper’s estimate of a ~75% higher output weight for managers supports this. This yields a spillover value of $21,600–$28,800 per manager.

Step 4: Aggregate Across the Organization Multiply the per-manager spillover by the number of managers whose teams are affected. This is the vertical spillover benefit. Combine it with the direct productivity gains of the trained/equipped frontline workers to get the full program benefit.

Sanity check: Compare the spillover total to the direct benefits. In the paper, vertical spillovers accounted for ~45% of the total program benefit — a useful benchmark for organizations where managers are a bottleneck and assistance requests are high.

Implications for the Cost-Benefit Analysis

• For projections: Use the 4-step method. If you lack data, start with a conservative assumption: direct worker gains + 25–45% additional spillover value from freed managerial time, depending on how time-constrained and strategic your managers are.

• For pilot measurement: Track “requests for help” (emails, tickets, meetings) as a leading indicator. The paper’s exposure-design approach (measuring connections to treated workers via email volume) can be replicated with modern communication logs to attribute changes.

• Beware of double counting: If you already measure manager output and expect it to rise, the spillover is already captured. But if managerial KPIs are fixed and their freed time is not repurposed, the spillover is unrealized. The intervention must be paired with a deliberate reallocation of manager time toward high-impact strategic work, otherwise the “time dividend” evaporates.

By grounding your analysis in observable time-use shifts rather than vague synergy claims, you turn the paper’s academic insight into a defensible financial case.

Personally, having read the paper just this month, I rushed to apply the concept to my project automating controls on regulatory financial reporting at hand. Its Benefits have already included saving 10 FTE of work efforts. When we asked the managers and reviewed past data, our conservative estimate showed an additional 10 hours per week freed up across the five managers who had previously spent that time onboarding employees, approving steps via email, and sorting out manual issues and escalations. The time they are now redirecting to improving the reporting with long-term initiatives.

In short, nearly half the value of upskilling projects accrues to managers, not the trainees. The same can be expected with Process Transformation, Optimization, or Automation initiatives. As a transformation leader, you can make better decisions on investments and efforts to strengthen both operational efficiency and strategic capacity of the organization.

Reference: “Training, Communications Patterns, and Spillovers Inside Organizations,” by Miguel Espinosa and Christoper Stanton (Journal of Political Economy, 2025)