Optimizing Meeting Time: A Mathematical Model of Organizational Productivity — Reggie Chan, CFA, FRICS (March 2026)

paper.md

title	Optimizing Meeting Time: A Mathematical Model of Organizational Productivity
date	2026-03-20
author	Reggie Chan, CFA, FRICS
version	0.1
template	optimization
outline	1. Introduction 2. Model Development 2.1 Time Allocation and Effective Work 2.2 Cognitive Fatigue and Task Complexity 2.3 Employee Engagement 2.4 Communication Quality and Learning 2.5 Stochastic Variability 3. Base Model Optimization 3.1 Log-Derivative Solution 3.2 The Balance Equation 3.3 Closed-Form Solution 4. Model Extensions 4.1 Meeting Fatigue Decay 4.2 Stochastic Term Development 4.3 Numerical Solution 5. Scenario Analysis 5.1 Meeting Overload (Moderate Complexity) 5.2 Documentation-Quality Intervention with Meeting Reduction 5.3 Optimal M* at Moderate and High Complexity 6. Discussion 7. Conclusion References Appendix: Practitioner Guide to Estimating Q_total

Optimizing Meeting Time: A Mathematical Model of Organizational Productivity

Author: Reggie Chan, CFA, FRICS Date: March 20, 2026

A note on methodology: The conceptual framework and domain intuitions in this paper are the author's. The mathematical formalization, validation, and scenario computations were developed collaboratively with AI assistants (Claude, Gemini) serving as pair-mathematician and pair-programmer.

Abstract

Organizations face a fundamental tension: meetings are essential for coordination, yet excessive meeting time cannibalizes the deep work required to execute tasks. This paper presents a mathematical optimization model that identifies the point where the marginal benefit of meetings (employee engagement, knowledge transfer) exactly equals the marginal cost (lost productive time, cognitive fatigue, context switching).

The model has a two-tier architecture. The base model treats communication quality as exogenous, yielding a closed-form quadratic solution for optimal daily meeting hours $M^*$ via log-differentiation of a multiplicative production function. The extension tier relaxes this assumption by introducing meeting fatigue decay on verbal quality and stochastic daily variability, solved numerically.

Under the model's assumptions and representative parameter values, scenario analyses illustrate three implications: (1) increasing meetings from 4 to 6 hours per day reduces modeled output to roughly 25% of baseline; (2) AI-assisted documentation acts as a force multiplier, with the model predicting 82% higher output with fewer synchronous hours; and (3) optimal meeting time for complex tasks falls well below employees' social preference point, revealing a structural tension between individual satisfaction and organizational productivity. These results are analytical consequences of the model — their empirical validity depends on calibration against observed data.

1. Introduction

Organizations today face a "Meeting Paradox": meetings are essential for coordination, alignment, and social cohesion, yet excessive meeting time cannibalizes the deep work required to execute complex tasks. Research in meeting science suggests a Goldilocks zone — too few meetings lead to misalignment and isolation, while too many lead to fatigue and productivity loss (Rogelberg et al., 2007; Lencioni, 2004).

The relationship between meeting hours and output is not linear. It is shaped by three interacting forces: context switching costs (the time lost refocusing after each interruption), cognitive fatigue (diminishing returns on extended work hours, especially for complex tasks), and communication quality (how effectively meeting content translates into retained, actionable knowledge).

Previous approaches to this problem have relied on simple linear trade-offs or arbitrary utility functions. This paper takes a different approach: we define a multiplicative production function for individual worker output and derive the optimal meeting allocation using calculus — specifically, log-differentiation of the production function to obtain a first-order condition that balances marginal engagement gains against marginal time losses.

The model is developed in two tiers:

• Base model (Sections 2–3): Communication quality parameters are treated as exogenous constants. Under this assumption, the log-derivative of the production function yields a clean quadratic equation whose smaller root is the optimal meeting time $M^*$. This tier provides closed-form analytical results that can be computed by hand.

• Extension tier (Section 4): Two simplifying assumptions are relaxed. First, verbal communication quality is allowed to decay with meeting duration (meeting fatigue). Second, a stochastic term captures day-to-day variability. The combined extension requires numerical optimization but produces more realistic predictions for organizations considering policy changes.

Section 5 applies both tiers to three scenarios that illustrate the model's practical implications: meeting overload, AI-assisted documentation as a force multiplier, and the sensitivity of optimal meeting time to task complexity.

Scope and epistemic status: This paper is a structural model and decision framework, not an empirical study. It formalizes a widespread managerial intuition — that excessive meeting load can become net destructive even when individual meetings provide coordination value — and tests whether that intuition survives contact with mathematics. Parameters are illustrative, not estimated from observational data. Scenario outputs are model-implied examples, not observed outcomes. The contribution is theory clarification: converting common-sense beliefs about meeting overload into a formal optimization problem, showing that an interior optimum exists under reasonable assumptions, and demonstrating how interruption costs and documentation quality shift that optimum.

2. Model Development

The total productivity of an individual worker on day $t$ is modeled as a multiplicative function of four components:

$$P_{worker}(M) = \Lambda \cdot f(t_{eff}(M)) \cdot \mu_E(M) \cdot (1 + \epsilon_t)$$

where $M$ is total daily meeting hours (the single decision variable), $\Lambda$ is a learning multiplier driven by communication quality, $f(t_{eff})$ captures productive output as a function of effective work time with cognitive fatigue, $\mu_E(M)$ represents employee engagement as a function of social interaction, and $\epsilon_t$ is a mean-zero stochastic term for daily variability.

The multiplicative structure reflects the assumption that these factors are complementary: high engagement with zero effective time produces nothing, and ample time with zero engagement produces nothing. Each component is developed below.

Plain English: A worker's daily output depends on four things multiplied together: how much they learn from meetings, how much focused time they have left, how engaged they feel, and random daily variation. If any one factor goes to zero, output goes to zero — you can't substitute more meetings for more focus time.

2.1 Time Allocation and Effective Work

The workday is a fixed resource of $H$ hours. Meeting time $M$ is the manager's decision variable, bounded by $0 \leq M \leq H$. The number of discrete meetings is determined by average meeting duration $\bar{d}$:

$$N_{meet} = \frac{M}{\bar{d}}$$

Each meeting incurs a context switching penalty $c_{switch}$ (the time required to refocus on deep work after an interruption). Effective productive time is:

$$t_{eff}(M) = H - M - N_{meet} \cdot c_{switch} = H - M\left(1 + \frac{c_{switch}}{\bar{d}}\right) = H - M\beta$$

where we define the switching cost factor:

$$\beta = 1 + \frac{c_{switch}}{\bar{d}}$$

This collapses the model to a single decision variable $M$. The factor $\beta > 1$ captures the amplifying effect of meeting fragmentation: shorter average meetings (smaller $\bar{d}$) mean more switching events per hour of meeting time, increasing the effective cost of each meeting hour.

The feasibility constraint is $t_{eff} > 0$, which requires $M < H/\beta$.

Plain English: Your actual work time isn't just "8 hours minus meetings." Every time you switch back to work after a meeting, you lose refocus time. If your meetings are short and frequent, the switching tax is brutal — an hour of meetings might cost you 1.25 hours of your day.

2.2 Cognitive Fatigue and Task Complexity

Productivity does not scale linearly with available time. The "eighth hour" of work is less productive than the first, especially for cognitively demanding tasks. We model this using a Cobb-Douglas power law:

$$f(t_{eff}) = t_{eff}^{\alpha}$$

where the fatigue exponent $\alpha$ is a decreasing function of task complexity $TC \in [0, 1]$:

$$\alpha(TC) = 1 - 0.3 \times TC$$

At the boundaries:

• $TC = 0$ (repetitive administration): $\alpha = 1.0$, output scales linearly with time.
• $TC = 1$ (complex engineering or strategy): $\alpha = 0.7$, severe diminishing returns.

The Cobb-Douglas form is standard in labor economics for modeling diminishing marginal productivity of a single input factor. The linear mapping from $TC$ to $\alpha$ is a simplifying assumption; empirical calibration could replace it with a nonlinear relationship.

Plain English: For simple tasks like data entry, two hours of work gives you twice the output of one hour. For complex tasks like system design, the second hour gives you less than the first — your brain fatigues. The $\alpha$ parameter controls how steep this drop-off is.

2.3 Employee Engagement

Engagement is modeled as a function of social interaction time. Both isolation (too few meetings) and burnout (too many meetings) reduce morale, suggesting a unimodal response. We use a Gaussian kernel:

$$\mu_E(M) = \exp\left(-\frac{(M - M_{opt})^2}{2\sigma^2}\right)$$

where $M_{opt}$ is the meeting duration at which social satisfaction peaks (engagement = 1.0), and $\sigma$ controls the width of the bell curve — how tolerant employees are of non-optimal schedules.

Calibration of $M_{opt}$: This parameter is organization-specific and should be estimated from employee surveys or observed team behavior. We adopt $M_{opt} = 2$ hours as a representative value, informed by the Slack Workforce Index (2023, n=10,333), which found that 2 hours per day is the threshold at which a majority of desk workers report feeling overburdened by meetings. We note that this survey establishes a burden threshold, not a measured satisfaction peak — the true engagement optimum may differ. The model's qualitative results are robust to this choice: $M^* < M_{opt}$ holds for any $M_{opt} \geq 1.5$, and the sensitivity table below shows that optimal meeting time scales roughly linearly with $M_{opt}$:

$M_{opt}$	$M^*$	$M^*/M_{opt}$	$t_{eff}$
1.5	0.03 h	0.02	8.0 h
2.0	0.44 h	0.22	7.5 h
2.5	0.84 h	0.34	7.1 h
3.0	1.23 h	0.41	6.6 h
3.5	1.61 h	0.46	6.2 h
4.0	1.97 h	0.49	5.8 h

Regardless of the $M_{opt}$ chosen, the model consistently recommends meeting time at roughly 20–50% of the social satisfaction peak for moderately complex work.

To illustrate the impact of this calibration choice, the following table compares the model's key outputs at $M_{opt} = 2$ (our adopted value) versus $M_{opt} = 3$ (a common informal assumption):

Metric	$M_{opt} = 2$	$M_{opt} = 3$	Difference
$M^*$ (moderate, $\alpha = 0.85$)	0.44 h (26 min)	1.23 h (74 min)	-0.79 h
$M^*$ (high, $\alpha = 0.70$)	0.67 h (40 min)	1.48 h (89 min)	-0.81 h
$t_{eff}$ at $M^*$ (moderate)	7.51 h	6.62 h	+0.89 h
Scenario 5.1: output at $M=6$ vs $M=4$	25% of baseline	30% of baseline	-5 pp
Scenario 5.2: AI output gain	+82%	+55%	+27 pp

The 1-hour difference in $M_{opt}$ shifts optimal meeting time by approximately 48 minutes and changes the AI scenario gain by 27 percentage points. The larger AI gain under $M_{opt} = 2$ occurs because reducing meetings to 2 hours places the team exactly at the engagement peak ($\mu_E = 1.0$), adding an 18% engagement bonus on top of the reclaimed time. At $M_{opt} = 3$, both $M = 2$ and $M = 4$ are equidistant from the peak, and the engagement ratio is neutral.

Key properties:

• At $M = M_{opt}$: $\mu_E = 1.0$ (peak engagement).
• As $M \to 0$ or $M \to H$: $\mu_E \to 0$ (isolation or exhaustion).
• The Gaussian is symmetric around $M_{opt}$, which is a simplification — in practice, the burnout side (too many meetings) may drop faster than the isolation side.

The derivative, which drives the optimization in Section 3, is:

$$\frac{d\mu_E}{dM} = -\frac{M - M_{opt}}{\sigma^2} \cdot \mu_E(M)$$

This is positive when $M < M_{opt}$ (more meetings increase engagement) and negative when $M > M_{opt}$ (more meetings decrease engagement).

Plain English: This draws a bell-shaped curve. At the peak ($M_{opt}$), you have just enough meetings to feel connected and informed — engagement is 100%. Too few meetings and you feel isolated; too many and you're exhausted. The $\sigma$ parameter controls how forgiving the curve is: a wider bell means employees are more resilient to bad scheduling.

2.4 Communication Quality and Learning

Meetings produce learning, but only in proportion to their quality. We define a learning multiplier $\Lambda$ that scales output based on how effectively meeting content is retained and applied:

$$\Lambda(Q_{total}) = 1 + \kappa_L \cdot \frac{Q_{total}}{1 + Q_{total}}$$

where $Q_{total} \in [0, 2.4]$ is aggregate communication quality (a single exogenous parameter combining verbal, written, and asynchronous channels), and $\kappa_L$ is a learning sensitivity factor.

The saturation form $Q_{total}/(1 + Q_{total})$ ensures diminishing returns: doubling communication quality does not double learning. As $Q_{total} \to \infty$, the learning bonus approaches $\kappa_L$.

With $\kappa_L = 0.2$ and $Q_{total} \in [0, 2.4]$:

• Poor communication ($Q_{total} = 0.5$): $\Lambda = 1 + 0.2 \times 0.33 = 1.07$
• Good communication ($Q_{total} = 1.2$): $\Lambda = 1 + 0.2 \times 0.55 = 1.11$
• Excellent communication ($Q_{total} = 2.4$): $\Lambda = 1 + 0.2 \times 0.71 = 1.14$

The range $\Lambda \in [1.0, 1.14]$ reflects a modest but meaningful learning bonus: a well-run meeting organization gets up to 14% more output from the same time allocation.

Critical design choice: $\Lambda$ does not depend on $M$. In the base model, $Q_{total}$ is exogenous — it reflects organizational practices (agenda discipline, documentation quality, tool adoption) rather than meeting duration. This makes $\Lambda$ constant with respect to the decision variable, which is necessary for the log-derivative optimization in Section 3 to be mathematically valid. Section 4.1 relaxes this assumption by allowing verbal quality to decay with $M$.

Plain English: $Q_{total}$ is a single number that captures "how good are your meetings and documentation?" It's not something the model optimizes — it's a given. Better meeting practices (agendas, AI notes, clear follow-ups) raise $Q_{total}$, which raises $\Lambda$, which raises output. The Appendix provides guidance on estimating $Q_{total}$ for your organization.

2.5 Stochastic Variability

Daily productivity varies due to factors outside the model: health, mood, external interruptions, commute quality, and other random shocks. We capture this with a multiplicative stochastic term $(1 + \epsilon_t)$ where $\epsilon_t$ is a mean-zero random variable.

In the base model, we optimize over expected productivity $E[P_{worker}]$. Since $E[\epsilon_t] = 0$, the stochastic term drops out of the first-order condition, and the optimal $M^*$ is determined entirely by the deterministic components. The base model solution is therefore the optimal meeting allocation on average.

Section 4.2 develops $\epsilon_t$ formally: specifying its distribution, autocorrelation structure (bad days tend to follow bad days), and the implications for variance in realized output around the optimal policy.

Plain English: Some days you're sharp, some days you're not — that's life. The base model finds the best meeting schedule on average. The extension in Section 4 asks: "How much does actual output bounce around that average, and does it matter for planning?"

3. Base Model Optimization

This section solves for $M^$, the total daily meeting hours that maximize worker output. It is critical to distinguish $M^$ from $M_{opt}$:

• $M_{opt}$ (Section 2.3) is the meeting time at which employees feel most socially satisfied — peak engagement, morale, and sense of connection. It reflects only the psychological dimension.
• $M^*$ is the meeting time that maximizes the full production function — balancing engagement gains against lost deep work time and cognitive fatigue. It reflects the organizational productivity optimum.

The model's central finding is that $M^* < M_{opt}$: the schedule that maximizes output requires fewer meetings than the schedule employees would prefer. The gap between the two quantifies the tension between individual social satisfaction and organizational productivity.

In the base model, $\Lambda$ is constant and $E[\epsilon_t] = 0$, so we maximize the deterministic core:

$$E[P_{worker}] \propto f(t_{eff}(M)) \cdot \mu_E(M) = (H - M\beta)^{\alpha} \cdot \exp\left(-\frac{(M - M_{opt})^2}{2\sigma^2}\right)$$

The decision variable is $M \in (0, H/\beta)$.

3.1 Log-Derivative Solution

Instead of maximizing $P$ directly, we maximize $\ln(P)$, which converts the product into a sum:

$$\ln(P) = \overbrace{\ln(\Lambda)}^{\text{constant}} + \overbrace{\alpha \ln(H - M\beta)}^{\text{fatigue-adjusted time}} + \overbrace{\left(-\frac{(M - M_{opt})^2}{2\sigma^2}\right)}^{\text{engagement}}$$

Differentiating with respect to $M$ and setting to zero:

$$\frac{d}{dM}[\ln(P)] = \frac{-\alpha\beta}{H - M\beta} + \frac{M_{opt} - M}{\sigma^2} = 0$$

The first term is the marginal cost of meeting time: each additional hour of meetings costs $\alpha\beta/(H - M\beta)$ in log-productivity, amplified by both cognitive fatigue ($\alpha$) and switching costs ($\beta$). The second term is the marginal benefit of engagement: each additional hour of meetings improves log-engagement by $(M_{opt} - M)/\sigma^2$, which is positive when $M < M_{opt}$.

Plain English: This equation finds the exact tipping point. The left side asks "how much productive time am I losing?" The right side asks "how much engagement am I gaining?" The optimum is where these two forces balance.

3.2 The Balance Equation

Rearranging the first-order condition:

$$\frac{M_{opt} - M}{\sigma^2} = \frac{\alpha\beta}{H - M\beta}$$

This is the balance equation. The left side is the marginal engagement gain (decreasing in $M$). The right side is the marginal time cost (increasing in $M$, since the denominator shrinks as $M$ grows). Their intersection determines $M^*$.

Note the role of $\beta = 1 + c_{switch}/\bar{d}$: it appears on both sides. Higher switching costs (larger $\beta$) increase the marginal cost of meeting time and reduce the feasible range, both pushing $M^*$ downward. Organizations with fragmented meeting schedules (many short meetings) face a steeper penalty than those with fewer, longer meetings.

3.3 Closed-Form Solution

Cross-multiplying the balance equation:

$$(M_{opt} - M)(H - M\beta) = \alpha\beta\sigma^2$$

Expanding:

$$M_{opt}H - M_{opt}M\beta - MH + M^2\beta = \alpha\beta\sigma^2$$

Collecting terms in standard quadratic form $aM^2 + bM + c = 0$:

$$\beta M^2 - (M_{opt}\beta + H)M + (M_{opt}H - \alpha\beta\sigma^2) = 0$$

Applying the quadratic formula:

$$M^* = \frac{(M_{opt}\beta + H) - \sqrt{(M_{opt}\beta + H)^2 - 4\beta(M_{opt}H - \alpha\beta\sigma^2)}}{2\beta}$$

We take the smaller root ($-$ branch), which lies in the feasible region $(0, H/\beta)$. The larger root exceeds $H/\beta$ and violates the positive-time constraint.

The discriminant simplifies to:

$$\Delta = (M_{opt}\beta - H)^2 + 4\alpha\beta^2\sigma^2$$

Since $\alpha, \beta, \sigma > 0$, we have $\Delta > 0$ always — a real solution exists for any valid parameter combination.

Worked example: Using representative values $H = 8$, $M_{opt} = 2$, $\sigma = 3.5$, $\alpha = 0.85$ (moderate complexity), $\beta = 1.125$:

$$1.125 M^2 - 10.25 M + 4.286 = 0$$

$$M^* = \frac{10.25 - \sqrt{85.78}}{2.25} = \frac{10.25 - 9.262}{2.25} \approx 0.44 \text{ hours}$$

The model predicts an optimal meeting allocation of approximately 26 minutes per day — roughly 6% of the workday. This is well below the engagement peak ($M_{opt} = 2$), confirming that the time cost of meetings dominates the engagement benefit for moderately complex work. Section 5.3 explores how this result changes with task complexity.

A critical caveat: $M^$ is the optimal quantity of meeting time, but the model's output also depends on meeting quality ($Q_{total}$) through the learning multiplier $\Lambda$. A 26-minute daily meeting budget only maximizes output if those meetings are high quality — clear agendas, effective facilitation, and strong documentation. Reducing meeting hours while allowing quality to deteriorate would lower $\Lambda$ and offset the time gains. The practical implication is that $M^$ and $Q_{total}$ must be optimized together: fewer meetings, but better ones. The Appendix provides guidance on estimating and improving $Q_{total}$.

Plain English: The optimal meeting time is the solution to a quadratic equation. It always has a real answer. The formula balances four forces: how much employees want meetings ($M_{opt}$), how tolerant they are of non-ideal schedules ($\sigma$), how cognitively demanding the work is ($\alpha$), and how fragmented the meeting schedule is ($\beta$).

4. Model Extensions

The base model treats communication quality as exogenous and ignores daily variability. This section relaxes both assumptions, producing an extended production function that requires numerical optimization.

4.1 Meeting Fatigue Decay

In practice, verbal communication quality degrades as meetings lengthen. Attention wanes, discussion quality drops, and later meetings in a long day are less productive than early ones. We model this with an exponential decay:

$$Q_{verbal}(M) = Q_{verbal}^{base} \cdot e^{-\gamma M}$$

where $Q_{verbal}^{base}$ is verbal quality at $M = 0$ (the quality of the first few minutes of a fresh meeting) and $\gamma > 0$ is the fatigue decay rate.

This modifies the total quality function. Decomposing $Q_{total}$ into its components:

$$Q_{total}(M) = Q_{verbal}(M) + \delta \cdot Q_{written} + \theta \cdot Q_{async}$$

where $\delta$ and $\theta$ weight written documentation and asynchronous tools respectively. Only the verbal component depends on $M$; written and async quality remain exogenous.

The learning multiplier becomes:

$$\Lambda(M) = 1 + \kappa_L \cdot \frac{Q_{total}(M)}{1 + Q_{total}(M)}$$

Since $\Lambda$ is now a function of $M$, the log-derivative picks up an additional term $d\ln(\Lambda)/dM \neq 0$, and the base model's closed-form quadratic no longer applies.

Plain English: Meeting quality fades as the day drags on. Your 7th hour of meetings is not as sharp as your 1st. This extension captures that decay — but the price is that we can no longer solve for $M^*$ with a formula. We need a computer.

4.2 Stochastic Term Development

Daily productivity fluctuates around the deterministic prediction. We model this with an AR(1) process:

$$\epsilon_t = \rho \cdot \epsilon_{t-1} + \nu_t, \quad \nu_t \sim \mathcal{N}(0, \sigma_\nu^2)$$

where $\rho \in [0, 1)$ is the autocorrelation coefficient (persistence of good/bad days) and $\sigma_\nu$ is the innovation standard deviation.

The stationary variance of $\epsilon_t$ is:

$$\text{Var}(\epsilon_t) = \frac{\sigma_\nu^2}{1 - \rho^2}$$

This does not affect the optimal $M^*$ (which maximizes expected output), but it characterizes the variance of realized output around the optimal policy. High $\rho$ means productivity shocks persist across days — a bad Monday predicts a bad Tuesday. High $\sigma_\nu$ means larger day-to-day swings regardless of persistence.

For planning purposes, a manager using $M^*$ can expect realized daily output to fall within $\pm 2\sigma_\epsilon$ of the deterministic prediction roughly 95% of the time.

Limitation: In the current model, the stochastic term does not affect the optimal policy $M^$ because we optimize over expected output $E[P_{worker}]$. A risk-averse decision maker — one who penalizes output variance — would shift $M^$ upward (more meetings), since meetings reduce variance by improving coordination. Incorporating risk aversion via a mean-variance objective $E[P] - \lambda \cdot \text{Var}(P)$ is a natural extension that would make the stochastic term policy-relevant. This development is left for future work.

Plain English: Some days are good, some are bad, and bad days tend to cluster. This doesn't change the best meeting schedule, but it tells you how much actual output will bounce around. If your team has high day-to-day variability ($\sigma_\nu$) and bad days are sticky ($\rho$ near 1), plan for wider swings.

4.3 Numerical Solution

The extended production function combines all components:

$$P_{worker}^{ext}(M) = \Lambda(M) \cdot (H - M\beta)^{\alpha} \cdot \mu_E(M) \cdot (1 + \epsilon_t)$$

where $\Lambda(M)$ now depends on $M$ through meeting fatigue decay. Maximizing $E[P_{worker}^{ext}]$ requires numerical optimization since $d\ln(\Lambda)/dM \neq 0$ introduces a term that does not reduce to a polynomial.

The numerical first-order condition is:

$$\frac{d\ln(\Lambda(M))}{dM} + \frac{M_{opt} - M}{\sigma^2} - \frac{\alpha\beta}{H - M\beta} = 0$$

where the first term (absent in the base model) captures the marginal cost of meeting fatigue on learning quality. This is solved via standard root-finding (e.g., Brent's method) on $M \in (0, H/\beta)$.

The extended model nests the base model: setting $\gamma = 0$ eliminates meeting fatigue decay, $\Lambda(M)$ becomes constant, and the numerical solution converges to the closed-form $M^*$.

Plain English: The extended model is the base model plus two realistic complications. A computer solves it in milliseconds. If you set the fatigue decay to zero, you get exactly the same answer as the formula in Section 3 — the extension generalizes rather than replaces the base model.

5. Scenario Analysis

We apply the model to three scenarios using representative parameter values: $H = 8$, $\sigma = 3.5$, $M_{opt} = 2$, $\bar{d} = 2$, $c_{switch} = 0.25$ ($\beta = 1.125$).

5.1 Meeting Overload (Moderate Complexity)

Situation: A team increases meetings from 4 hours to 6 hours per day. Task complexity is moderate ($TC = 0.5$, $\alpha = 0.85$). Communication quality is unchanged.

Engagement change ($\mu_E$):

• At $M = 4$: $\mu_E = \exp(-(4-2)^2 / 2(3.5)^2) = \exp(-4/24.5) \approx 0.849$
• At $M = 6$: $\mu_E = \exp(-(6-2)^2 / 2(3.5)^2) = \exp(-16/24.5) \approx 0.520$
• Engagement ratio: $0.520 / 0.849 \approx 0.613$ (39% drop)

Effective time change ($t_{eff}$):

• At $M = 4$: $t_{eff} = 8 - 4 \times 1.125 = 3.5$ hours
• At $M = 6$: $t_{eff} = 8 - 6 \times 1.125 = 1.25$ hours
• Output ratio: $(1.25/3.5)^{0.85} = 0.357^{0.85} \approx 0.417$ (58% drop)

Total impact:

Holding communication quality constant ($\Lambda$ cancels in the ratio):

$$\frac{P(M=6)}{P(M=4)} = \frac{\mu_E(6)}{\mu_E(4)} \times \frac{f(t_{eff}(6))}{f(t_{eff}(4))} = 0.613 \times 0.417 \approx 0.256$$

Under these parameters, modeled output falls to roughly 25% of baseline output. The $1/0.25 \approx 4\times$ workforce increase needed to compensate for a 2-hour schedule change illustrates the nonlinear cost of meeting overload.

Plain English: Going from 4 to 6 hours of meetings doesn't cut output by 25% — it cuts it by 75%. The engagement drop and the time loss compound multiplicatively. This is the "double whammy."

5.2 Documentation-Quality Intervention with Meeting Reduction

Situation: AI-generated meeting summaries are introduced, raising $Q_{written}$ from 0.3 to 1.0. This scenario bundles three simultaneous changes — improved documentation quality, reduced meeting hours, and altered engagement — and the gains should be understood as the combined effect of this intervention, not AI uplift alone. This allows meetings to be cut from 4 to 2 hours ($N_{meet}$ drops from 2 to 1). Moderate complexity ($\alpha = 0.85$).

This is a comparative statics scenario: the exogenous parameter $Q_{total}$ changes between the baseline and intervention cases.

Baseline ($M = 4$, pre-AI):

• $Q_{verbal} = 0.7$, $Q_{written} = 0.3$ (pre-AI), $Q_{async} = 0.5$
• $Q_{total} = 0.7 + 0.8 \times 0.3 + 0.6 \times 0.5 = 0.7 + 0.24 + 0.30 = 1.24$
• $\Lambda_{base} = 1 + 0.2 \times 1.24/2.24 = 1 + 0.111 = 1.111$
• $\mu_E(4) = 0.849$
• $t_{eff}(4) = 3.5$ hours

With AI ($M = 2$, $Q_{written} = 1.0$):

• $Q_{total} = 0.7 + 0.8 \times 1.0 + 0.6 \times 0.5 = 0.7 + 0.80 + 0.30 = 1.80$
• $\Lambda_{AI} = 1 + 0.2 \times 1.80/2.80 = 1 + 0.129 = 1.129$
• $\mu_E(2) = \exp(-(2-2)^2/24.5) = \exp(0) = 1.000$
• $t_{eff}(2) = 8 - 2 \times 1.125 = 5.75$ hours

Ratio analysis:

$$\frac{P_{AI}}{P_{base}} = \frac{\Lambda_{AI}}{\Lambda_{base}} \times \frac{\mu_E(2)}{\mu_E(4)} \times \frac{t_{eff}(2)^{\alpha}}{t_{eff}(4)^{\alpha}}$$

$$= \frac{1.129}{1.111} \times \frac{1.000}{0.849} \times \frac{5.75^{0.85}}{3.5^{0.85}}$$

$$= 1.016 \times 1.177 \times \frac{4.65}{3.02} = 1.016 \times 1.177 \times 1.525 \approx 1.824$$

Result: The modeled output gain decomposes into three factors:

• Time recovery ($t_{eff}$ ratio): $1.525\times$ — the dominant effect, contributing 52 percentage points
• Engagement improvement ($\mu_E$ ratio): $1.177\times$ — landing at $M_{opt}$ adds 18 percentage points
• Documentation quality ($\Lambda$ ratio): $1.016\times$ — a modest 1.6 percentage point contribution

Combined: $1.016 \times 1.177 \times 1.525 \approx 1.82$, an 82% modeled output gain. The gain is driven overwhelmingly by reclaimed deep work time, with a secondary boost from improved engagement. The learning multiplier contribution is minor. This underscores that AI documentation tools should be evaluated primarily as time-recovery mechanisms that enable meeting reduction, rather than as standalone knowledge-management improvements.

Plain English: AI meeting notes don't just save time — they let you cut meetings in half while improving information quality. With $M_{opt} = 2$, cutting to 2 hours lands you right at peak engagement. The gains compound: more deep work time, higher morale, and better documentation.

5.3 Optimal $M^*$ at Moderate and High Complexity

We solve the closed-form quadratic from Section 3.3 at two complexity levels.

Moderate complexity ($TC = 0.5$, $\alpha = 0.85$):

$$\beta M^2 - (M_{opt}\beta + H)M + (M_{opt}H - \alpha\beta\sigma^2) = 0$$

$$1.125 M^2 - (2 \times 1.125 + 8)M + (2 \times 8 - 0.85 \times 1.125 \times 12.25) = 0$$

$$1.125 M^2 - 10.25 M + (16 - 11.716) = 0$$

$$1.125 M^2 - 10.25 M + 4.284 = 0$$

$$M^* = \frac{10.25 - \sqrt{10.25^2 - 4(1.125)(4.284)}}{2(1.125)} = \frac{10.25 - \sqrt{105.06 - 19.28}}{2.25}$$

$$= \frac{10.25 - \sqrt{85.78}}{2.25} = \frac{10.25 - 9.262}{2.25} = \frac{0.988}{2.25} \approx 0.44 \text{ hours}$$

At moderate complexity, $M^ \approx 0.44$ hours* (26 minutes, 6% of the workday).

High complexity ($TC = 1.0$, $\alpha = 0.7$):

$$1.125 M^2 - 10.25 M + (16 - 0.7 \times 1.125 \times 12.25) = 0$$

$$1.125 M^2 - 10.25 M + (16 - 9.647) = 0$$

$$1.125 M^2 - 10.25 M + 6.353 = 0$$

$$M^* = \frac{10.25 - \sqrt{105.06 - 28.59}}{2.25} = \frac{10.25 - \sqrt{76.47}}{2.25} = \frac{10.25 - 8.745}{2.25} = \frac{1.505}{2.25} \approx 0.67 \text{ hours}$$

At high complexity, $M^ \approx 0.67$ hours* (40 minutes, 8% of the workday).

Interpretation: Counter-intuitively, $M^*$ is higher for more complex tasks. This is because higher complexity ($\alpha$ lower) reduces the marginal cost of time loss — when each hour of work already suffers diminishing returns, losing an hour to meetings is relatively less costly. The engagement benefit of that extra meeting time then tips the balance. However, the difference is small (0.23 hours), and both values are well below the satisfaction peak $M_{opt} = 2$.

Plain English: For complex work, the total daily meeting budget that maximizes output ($M^*$) is only 25–40 minutes. That's far less than the 2 hours where employees feel most connected ($M_{opt}$). Under these assumptions, for complex work, protect deep work time aggressively, even if it means scheduling fewer meetings than people would socially prefer.

6. Discussion

The model reveals three structural insights about meeting allocation in knowledge work:

1. The engagement-productivity tension is real and quantifiable. Employees may prefer approximately $M_{opt} = 2$ hours of daily social interaction, but the productivity-maximizing allocation is well below that ($M^* \approx 0.4$–$0.7$ hours). This gap is not a management failure — it is a mathematical consequence of the multiplicative production function. The cost of lost deep work time compounds with cognitive fatigue, creating a steeper penalty curve than the engagement benefit can offset.

2. Meeting fragmentation matters as much as meeting volume. The switching cost factor $\beta = 1 + c_{switch}/\bar{d}$ amplifies the effective cost of each meeting hour. An organization with 30-minute meetings ($\bar{d} = 0.5$, $\beta = 1.5$) faces a 33% higher effective meeting cost than one with 2-hour meetings ($\bar{d} = 2$, $\beta = 1.125$). The model suggests that consolidating meetings into fewer, longer blocks is independently valuable — separate from reducing total meeting hours.

3. AI documentation shifts the frontier, not just the optimum. In Scenario 5.2, AI-generated meeting summaries produce an 82% output gain. The gain comes from three reinforcing effects: reclaimed deep work time, an engagement boost from landing at the $M_{opt} = 2$ peak, and a modest learning improvement ($\Lambda$ ratio contributes only 1.6%). The strategic implication is that AI documentation tools should be evaluated primarily as time-recovery and engagement-optimization mechanisms.

Limitations:

• The Gaussian engagement function is symmetric, but real-world engagement may decline faster with excess meetings (burnout) than with insufficient meetings (isolation). An asymmetric kernel (e.g., skew-normal) could capture this.
• The model treats individual workers in isolation. Team-level effects — coordination, information sharing, collective decision-making — are outside scope.
• All parameter values used in scenarios are illustrative. Empirical calibration against observed meeting-output data would strengthen the practical claims.
• The base model assumes a fixed workday $H$. In practice, workers may extend hours to compensate for meeting-heavy schedules, introducing a second decision variable.
• The multiplicative production function implies low substitutability across factors — a collapse in any one component (engagement, time, or quality) cannot be offset by increases in another. This structural choice contributes to the severe scenario magnitudes; an additive or CES specification would produce less extreme results. The multiplicative form is chosen for analytical tractability and because it captures the intuition that meetings with zero engagement or zero available work time produce zero output.

7. Conclusion

This paper presents a mathematical framework for optimizing daily meeting time allocation. The core contribution is a closed-form quadratic solution for optimal meeting hours $M^*$, derived from log-differentiation of a multiplicative production function that combines cognitive fatigue (Cobb-Douglas), employee engagement (Gaussian), and a learning multiplier driven by communication quality.

Under illustrative calibration with representative parameter values, the model yields three findings: (1) increasing total daily meetings from 4 to 6 hours reduces modeled output to approximately 25% of baseline, illustrating the nonlinear cost of meeting overload; (2) AI-assisted documentation enables roughly 82% higher output by reclaiming deep work time and optimizing engagement; and (3) the productivity-maximizing total daily meeting budget ($M^*$) falls in the range of 0.4–0.7 hours (25–40 minutes), well below the 2-hour social satisfaction peak ($M_{opt}$). This gap — between the meeting schedule employees would prefer and the one that maximizes output — is the model's central finding.

The two-tier architecture separates what can be solved analytically (base model with exogenous quality) from what requires numerical methods (extensions with meeting fatigue decay and stochastic variability). This provides both a tractable framework for reasoning about meeting policy and a more realistic model for simulation and calibration.

The model's practical value lies not in its specific numerical outputs — which depend on parameter calibration — but in subjecting a common managerial intuition to formal scrutiny. That intuition — that excessive meeting load can become net destructive even when individual meetings provide coordination value — survives four tests:

1. Non-contradiction: the variables interact coherently within the multiplicative structure without producing logical inconsistencies.
2. Mechanistic plausibility: the model has a causal structure (time loss, switching costs, cognitive fatigue, engagement trade-off) rather than a curve-fit.
3. Comparative statics: increasing interruption costs or reducing documentation quality shifts the optimum in the expected direction.
4. Organizational interpretability: a manager can inspect the balance equation and understand why the result occurs.

The core earned claim is this: excess meeting load can be modeled as a genuine optimization failure rather than a mere cultural complaint, and asynchronous documentation can be modeled as a partial substitute for synchronous coordination. The model does not identify the true empirical optimum for any given firm. It formalizes a widespread belief and shows that, under reasonable structural assumptions, the belief has mathematical legitimacy.

References

1. Rogelberg, S. G., Scott, C., & Kello, J. (2007). The Science and Fiction of Meetings. MIT Sloan Management Review, 48(2), 18–21. https://sloanreview.mit.edu/article/the-science-and-fiction-of-meetings/
2. Lencioni, P. (2004). Death by Meeting. Jossey-Bass. https://www.tablegroup.com/product/dbm/
3. Newport, C. (2016). Deep Work: Rules for Focused Success in a Distracted World. Grand Central Publishing. https://www.hachettebookgroup.com/titles/cal-newport/deep-work/9781455586691/
4. Mark, G., Gudith, D., & Klocke, U. (2008). The Cost of Interrupted Work: More Speed and Stress. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 107–110. https://dl.acm.org/doi/10.1145/1357054.1357072
5. Cobb, C. W., & Douglas, P. H. (1928). A Theory of Production. American Economic Review, 18(1), 139–165. https://www.jstor.org/stable/1811556
6. Allen, J. A., & Rogelberg, S. G. (2013). Manager-Led Group Meetings: A Context for Promoting Employee Engagement. Group & Organization Management, 38(5), 543–569. https://journals.sagepub.com/doi/abs/10.1177/1059601113503040

Appendix: Practitioner Guide to Estimating $Q_{total}$

The formal model treats $Q_{total} \in [0, 2.4]$ as a single exogenous parameter. For practitioners who want to estimate $Q_{total}$ for their organization, the following decomposition provides a structured approach.

Step 1: Rate verbal meeting quality ($Q_{verbal} \in [0, 1]$)

Average four sub-scores:

• Agenda clarity (0–1): Are meetings structured with clear objectives?
• Facilitation quality (0–1): Does someone manage time, participation, and decisions?
• Participation breadth (0–1): Do relevant people contribute, or do 1–2 voices dominate?
• Decision quality (0–1): Do meetings produce clear decisions and action items?

$Q_{verbal} = 0.25 \times (\text{Agenda} + \text{Facilitation} + \text{Participation} + \text{Decisions})$

Step 2: Rate documentation quality ($Q_{written} \in [0, 1]$)

Multiply three factors:

• Retention ($K_R$, 0–1): What fraction of meeting content is captured? (Human notes: ~0.3; AI transcription: ~0.9)
• Accuracy ($I_A$, 0–1): How faithfully do notes reflect what was said?
• Completeness ($I_C$, 0–1): Are action items, decisions, and context all captured?

$Q_{written} = K_R \times I_A \times I_C$

Step 3: Rate async communication quality ($Q_{async} \in [0, 1]$)

Single score reflecting Slack/email/project-management tool effectiveness. High score = information is findable, timely, and reduces the need for synchronous meetings.

Step 4: Compute $Q_{total}$

$$Q_{total} = Q_{verbal} + 0.8 \times Q_{written} + 0.6 \times Q_{async}$$

The weights ($\delta = 0.8$, $\theta = 0.6$) reflect the relative importance of written and async channels. These can be adjusted based on organizational context — a remote-first company might weight $Q_{async}$ higher.

Typical ranges:

Organization type	$Q_{verbal}$	$Q_{written}$	$Q_{async}$	$Q_{total}$
No structure, no notes	0.3	0.1	0.3	0.56
Good facilitation, manual notes	0.7	0.3	0.5	1.24
Good facilitation, AI notes, strong async	0.7	0.9	0.8	1.90